Development of deep learning segmentation models for coronary X-ray angiography: Quality assessment by a new global segmentation score and comparison with human performance

array:24 ["pii" => "S0870255122001354" "issn" => "08702551" "doi" => "10.1016/j.repc.2022.04.001" "estado" => "S300" "fechaPublicacion" => "2022-12-01" "aid" => "1927" "copyright" => "Sociedade Portuguesa de Cardiologia" "copyrightAnyo" => "2022" "documento" => "article" "crossmark" => 1 "licencia" => "http://creativecommons.org/licenses/by-nc-nd/4.0/" "subdocumento" => "fla" "cita" => "Rev Port Cardiol. 2022;41:1011-21" "abierto" => array:3 [ "ES" => true "ES2" => true "LATM" => true ] "gratuito" => true "lecturas" => array:1 [ "total" => 0 ] "itemSiguiente" => array:19 [ "pii" => "S0870255122002864" "issn" => "08702551" "doi" => "10.1016/j.repc.2022.07.007" "estado" => "S300" "fechaPublicacion" => "2022-12-01" "aid" => "2006" "copyright" => "Sociedade Portuguesa de Cardiologia" "documento" => "simple-article" "crossmark" => 1 "licencia" => "http://creativecommons.org/licenses/by-nc-nd/4.0/" "subdocumento" => "dis" "cita" => "Rev Port Cardiol. 2022;41:1023-4" "abierto" => array:3 [ "ES" => true "ES2" => true "LATM" => true ] "gratuito" => true "lecturas" => array:1 [ "total" => 0 ] "en" => array:10 [ "idiomaDefecto" => true "cabecera" => "Editorial comment" "titulo" => "Machine learning – coming soon to a cath lab near you? Or ‘don’t hold your breath’?" "tienePdf" => "en" "tieneTextoCompleto" => "en" "paginas" => array:1 [ 0 => array:2 [ "paginaInicial" => "1023" "paginaFinal" => "1024" ] ] "titulosAlternativos" => array:1 [ "pt" => array:1 [ "titulo" => "Machine learning – brevemente num laboratório de hemodinâmica perto de si? Ou é melhor esperar sentado?" ] ] "contieneTextoCompleto" => array:1 [ "en" => true ] "contienePdf" => array:1 [ "en" => true ] "autores" => array:1 [ 0 => array:2 [ "autoresLista" => "António Miguel Ferreira" "autores" => array:1 [ 0 => array:2 [ "nombre" => "António Miguel" "apellidos" => "Ferreira" ] ] ] ] ] "idiomaDefecto" => "en" "EPUB" => "https://multimedia.elsevier.es/PublicationsMultimediaV1/item/epub/S0870255122002864?idApp=UINPBA00004E" "url" => "/08702551/0000004100000012/v2_202212230532/S0870255122002864/v2_202212230532/en/main.assets" ] "itemAnterior" => array:19 [ "pii" => "S0870255122002876" "issn" => "08702551" "doi" => "10.1016/j.repc.2022.07.008" "estado" => "S300" "fechaPublicacion" => "2022-12-01" "aid" => "2007" "copyright" => "Sociedade Portuguesa de Cardiologia" "documento" => "simple-article" "crossmark" => 1 "licencia" => "http://creativecommons.org/licenses/by-nc-nd/4.0/" "subdocumento" => "dis" "cita" => "Rev Port Cardiol. 2022;41:1009-10" "abierto" => array:3 [ "ES" => true "ES2" => true "LATM" => true ] "gratuito" => true "lecturas" => array:1 [ "total" => 0 ] "en" => array:10 [ "idiomaDefecto" => true "cabecera" => "Editorial comment" "titulo" => "Thromboembolic risk in mitral stenosis: Are we underestimating the potential role of inflammation?" "tienePdf" => "en" "tieneTextoCompleto" => "en" "paginas" => array:1 [ 0 => array:2 [ "paginaInicial" => "1009" "paginaFinal" => "1010" ] ] "titulosAlternativos" => array:1 [ "pt" => array:1 [ "titulo" => "Risco trombo-embólico na estenose mitral: estaremos a subestimar o papel da inflamação?" ] ] "contieneTextoCompleto" => array:1 [ "en" => true ] "contienePdf" => array:1 [ "en" => true ] "autores" => array:1 [ 0 => array:2 [ "autoresLista" => "Marisa Trabulo" "autores" => array:1 [ 0 => array:2 [ "nombre" => "Marisa" "apellidos" => "Trabulo" ] ] ] ] ] "idiomaDefecto" => "en" "EPUB" => "https://multimedia.elsevier.es/PublicationsMultimediaV1/item/epub/S0870255122002876?idApp=UINPBA00004E" "url" => "/08702551/0000004100000012/v2_202212230532/S0870255122002876/v2_202212230532/en/main.assets" ] "en" => array:21 [ "idiomaDefecto" => true "cabecera" => "Original Article" "titulo" => "Development of deep learning segmentation models for coronary X-ray angiography: Quality assessment by a new global segmentation score and comparison with human performance" "tieneTextoCompleto" => true "paginas" => array:1 [ 0 => array:2 [ "paginaInicial" => "1011" "paginaFinal" => "1021" ] ] "autores" => array:1 [ 0 => array:4 [ "autoresLista" => "Miguel Nobre Menezes, João Lourenço-Silva, Beatriz Silva, Tiago Rodrigues, Ana Rita G. Francisco, Pedro Carrilho Ferreira, Arlindo L. Oliveira, Fausto J. Pinto" "autores" => array:8 [ 0 => array:4 [ "nombre" => "Miguel" "apellidos" => "Nobre Menezes" "email" => array:1 [ 0 => "mnmenezes.gm@gmail.com" ] "referencia" => array:3 [ 0 => array:2 [ "etiqueta" => "a" "identificador" => "aff0005" ] 1 => array:2 [ "etiqueta" => "b" "identificador" => "aff0010" ] 2 => array:2 [ "etiqueta" => "*" "identificador" => "cor0005" ] ] ] 1 => array:3 [ "nombre" => "João" "apellidos" => "Lourenço-Silva" "referencia" => array:1 [ 0 => array:2 [ "etiqueta" => "c" "identificador" => "aff0015" ] ] ] 2 => array:3 [ "nombre" => "Beatriz" "apellidos" => "Silva" "referencia" => array:2 [ 0 => array:2 [ "etiqueta" => "a" "identificador" => "aff0005" ] 1 => array:2 [ "etiqueta" => "b" "identificador" => "aff0010" ] ] ] 3 => array:3 [ "nombre" => "Tiago" "apellidos" => "Rodrigues" "referencia" => array:2 [ 0 => array:2 [ "etiqueta" => "a" "identificador" => "aff0005" ] 1 => array:2 [ "etiqueta" => "b" "identificador" => "aff0010" ] ] ] 4 => array:3 [ "nombre" => "Ana Rita G." "apellidos" => "Francisco" "referencia" => array:2 [ 0 => array:2 [ "etiqueta" => "a" "identificador" => "aff0005" ] 1 => array:2 [ "etiqueta" => "b" "identificador" => "aff0010" ] ] ] 5 => array:3 [ "nombre" => "Pedro" "apellidos" => "Carrilho Ferreira" "referencia" => array:2 [ 0 => array:2 [ "etiqueta" => "a" "identificador" => "aff0005" ] 1 => array:2 [ "etiqueta" => "b" "identificador" => "aff0010" ] ] ] 6 => array:3 [ "nombre" => "Arlindo L." "apellidos" => "Oliveira" "referencia" => array:1 [ 0 => array:2 [ "etiqueta" => "c" "identificador" => "aff0015" ] ] ] 7 => array:3 [ "nombre" => "Fausto J." "apellidos" => "Pinto" "referencia" => array:2 [ 0 => array:2 [ "etiqueta" => "a" "identificador" => "aff0005" ] 1 => array:2 [ "etiqueta" => "b" "identificador" => "aff0010" ] ] ] ] "afiliaciones" => array:3 [ 0 => array:3 [ "entidad" => "Structural and Coronary Heart Disease Unit, Cardiovascular Center of the University of Lisbon, Faculdade de Medicina, Universidade de Lisboa, Lisboa, Portugal" "etiqueta" => "a" "identificador" => "aff0005" ] 1 => array:3 [ "entidad" => "Serviço de Cardiologia, Departamento de Coração e Vasos, CHULN Hospital de Santa Maria, Lisboa, Portugal" "etiqueta" => "b" "identificador" => "aff0010" ] 2 => array:3 [ "entidad" => "INESC-ID/Instituto Superior Técnico, University of Lisbon, Portugal" "etiqueta" => "c" "identificador" => "aff0015" ] ] "correspondencia" => array:1 [ 0 => array:3 [ "identificador" => "cor0005" "etiqueta" => "⁎" "correspondencia" => "Corresponding author." ] ] ] ] "titulosAlternativos" => array:1 [ "pt" => array:1 [ "titulo" => "Desenvolvimento de modelos de deep learning para segmentação de coronariografias: aferição de qualidade por um novo modelo de segmentação global e comparação com desempenho humano" ] ] "resumenGrafico" => array:2 [ "original" => 0 "multimedia" => array:7 [ "identificador" => "fig0090" "etiqueta" => "Figure 5" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => true "mostrarDisplay" => false "figura" => array:1 [ 0 => array:4 [ "imagen" => "gr5.jpeg" "Alto" => 1305 "Ancho" => 1305 "Tamanyo" => 103564 ] ] "descripcion" => array:1 [ "en" => "Crossovers in spider (above) and extreme RAO cranial (below) views generating artifacts." ] ] ] "textoCompleto" => "IntroductionArtificial intelligence (AI) has shown great potential in medicine, in applications such as predictive data analysis,<a class="elsevierStyleCrossRef" href="#bib0125">1</a> decision making support<a class="elsevierStyleCrossRef" href="#bib0130">2</a> or even medical education/awareness improvement,<a class="elsevierStyleCrossRef" href="#bib0135">3</a> and especially in image analysis. Several publications have demonstrated impressive results with regards to electrocardiogram,<a class="elsevierStyleCrossRef" href="#bib0140">4</a> echocardiography,<a class="elsevierStyleCrossRefs" href="#bib0145">5,6</a> or magnetic resonance imaging.<a class="elsevierStyleCrossRefs" href="#bib0155">7,8</a>The use of AI in Interventional Cardiology (IC) is, however, still a vastly underexplored field. Its application to coronary angiography (CAG) has been explored in very few medical or biology publication.<a class="elsevierStyleCrossRefs" href="#bib0165">9–12</a> There are, nonetheless, many possibilities,<a class="elsevierStyleCrossRef" href="#bib0185">13</a> ranging from automatic anatomical identification, stenosis analysis, lesion subset characterization and perhaps even physiological index derivation. Regardless of the task, arguably the first step in applying AI to CAG is separating and identifying relevant information – the coronary tree – from non-relevant information (bones, other structures). This task is called segmentation.<a class="elsevierStyleCrossRef" href="#bib0190">14</a>In this paper, we explore the development of AI models for automatic coronary artery segmentation from CAG, and assess the results from a clinical perspective, using a new set of criteria and score clinically defined by a panel of Interventional Cardiologists.MethodsDataset selectionInclusion criteriaWe retrospectively and randomly included patients who had undergone CAG and invasive physiology assessment (fractional flow reserve and/or other indexes) during the procedure at a single center (tertiary university hospital).These patients have at least intermediate lesions in one or more vessels. Around one third usually undergo revascularization due to the severity of their disease.<a class="elsevierStyleCrossRefs" href="#bib0195">15,16</a> Therefore, a dataset focusing on these patients comprises a wide spectrum of obstructive coronary artery disease in a relatively balanced way.Exclusion criteriaWe excluded cases where any of the following applied:<ul class="elsevierStyleList" id="lis0005"><li class="elsevierStyleListItem" id="lsti0005">1)Major occluded vessels (acute or chronic)</li><li class="elsevierStyleListItem" id="lsti0010">2)Poor image quality</li><li class="elsevierStyleListItem" id="lsti0015">3)Less than two orthogonal views in the left coronary artery (LCA) - one caudal and one cranial - or absence of at least one left oblique (LAO) view - either cranial or simple - in the right coronary artery (RCA)</li><li class="elsevierStyleListItem" id="lsti0020">4)Patients with previous cardiac surgery, cardiac devices or other sources of potential artifact.</li></ul>Image selectionA single best frame was selected for each diagnostic angulation incidence in each patient.Dataset sizeThe dataset size was the result of a trade-off between two opposing criteria: dimension large enough for successful training of a deep convolutional neural network, estimated from published data<a class="elsevierStyleCrossRefs" href="#bib0165">9,12,17,18</a> vs. expected time required to complete the annotation. We estimated the latter based on a short period of annotation testing prior to formal dataset creation. The trade-off pointed to a training set size of roughly 400.We then randomly and consecutively selected patients until a total of at least 400 annotated images were obtained.Baseline annotation processBaseline human dataset images were annotated by two senior Cardiology Fellows (TR/BS) previously trained in CAG interpretation, under the supervision of an Interventional Cardiologist (MNM), who also annotated. Images were periodically reviewed and perfected by all three. This meant that any initial heterogeneity between annotators was corrected by consensus. The small size of the team was aimed at reducing heterogeneity, as we noticed during the preparatory phase that some operators tended to annotate too much (Supplementary figure 1, Appendix A), while others did the opposite (Supplementary figure 2, Appendix A).Both the catheter (labeled red) and the coronaries (labeled white) were to be segmented.The coronary tree was to be fully segmented up to branches of approximately 2 mm in caliper at their origin (as the vessel became smaller, it was to be segmented until discernible), using the catheter as reference (without formal measurements – eyeball appreciation was used). There were several reasons for this: (1) when performing percutaneous coronary intervention, vessels <2 mm are usually approached conservatively, as the risk of target lesion failure increases significantly<a class="elsevierStyleCrossRefs" href="#bib0215">19,20</a>; (2) human annotation is cumbersome – segmenting every single vessel would increase the risk of errors significantly; (3) including very small vessels might increase the chances of artifacts from bone or other structures when training and applying AI models.Baseline artificial intelligence model trainingWe performed segmentation using an encoder-decoder fully convolutional neural network based on the U-Net,<a class="elsevierStyleCrossRef" href="#bib0225">21</a> commonly used in medical image segmentation. As the name suggests, these neural networks are composed of an encoder, responsible for extracting image features, and a decoder, which processes those features to produce segmentation masks. To derive the best approach for this task, we conducted a comparative study of encoder and decoder architectures, which resulted in the proposal of the EfficientUNet++, a computationally efficient and high-performing decoder architecture<a class="elsevierStyleCrossRef" href="#bib0230">22</a> that, in this work, we combine with an EfficientNet-B5 encoder<a class="elsevierStyleCrossRef" href="#bib0235">23</a> (<a class="elsevierStyleCrossRef" href="#fig0070">Figure 1</a>).<elsevierMultimedia ident="fig0070"></elsevierMultimedia>To ensure fair evaluation, it was necessary to guarantee that each model was tested on data that it had not seen during training. Therefore, the dataset was split, at the patient level, into 13 subsets of approximately 32 angiograms each. Each subset segmentation was performed using a neural network trained exclusively on the remaining data. This enabled the assessment of the segmentation results for the entire cohort, as the usual splitting into a training and testing dataset would have yielded a much smaller group of images for result assessment.The training hyperparameters, including the number of training epochs and the learning rate decay schedule, were set on the first train-test split, using one of the 12 training data subsets for validation. The selected values were then used on every other train-test split, and to train the model on the whole training set of the first split.Enhanced human modelThe results of the baseline AI training were reviewed by the annotating team, without any formal grading, which would be performed subsequently (see below). For each image, both human and AI segmentation were compared with the original. Each annotation was then perfected using a mixture of the best of baseline human segmentation and baseline AI, with additional de novo manual segmentation as needed.Enhanced artificial intelligence modelThe neural network architecture and training procedure were identical for both the baseline and enhanced AI model (<a class="elsevierStyleCrossRef" href="#fig0070">Figure 1</a>). The sole difference was the dataset. The baseline AI model was trained using the baseline human annotations, whereas the enhanced AI model was trained using enhanced human annotations.<a class="elsevierStyleCrossRef" href="#fig0075">Figure 2</a> outlines the development stages.<elsevierMultimedia ident="fig0075"></elsevierMultimedia>Performance assessmentNon-medical metricsAI models were assessed using the Dice Similarity Coefficient (DSC) and Generalized Dice Score (GDS), measures of the overlap between segmentations. Given two segmentations, the DSC has a value between 0: no overlap and 1: total overlap, corresponding to the ratio between the area of their intersection and the sum of their areas. GDS<a class="elsevierStyleCrossRef" href="#bib0240">24</a> is a weighted sum of each class's DSC that attributes the same importance to all classes, regardless of their frequency. While DSC and GDS alone do not reflect clinical usefulness, they are helpful and entirely objective metrics that enable a simple comparison between models.Clinical performance criteriaThe DSC objectively assesses model performance. However, it does not provide a medically meaningful impression of whether segmentation is appropriate. Also, because the DCS can only be calculated based on previously annotated images, it cannot be applied to new, unannotated datasets in the future. To overcome these limitations, we created a set of criteria to assess performance as interpreted by expert physicians.The following 11 criteria are as objectively defined as possible and were analyzed for each image. Each was independently met or not. A “perfect” example is shown in <a class="elsevierStyleCrossRef" href="#fig0080">Figure 3</a>. Supplementary Figures 3 to 13 (Appendix A) show error examples for each.<ul class="elsevierStyleList" id="lis0010"><li class="elsevierStyleListItem" id="lsti0025">1)Catheter segmentation:</li><li class="elsevierStyleListItem" id="lsti0030">a.Main segmentation: The distal part of the catheter (i.e. the closest discernible portion to the coronary artery in the ascending aorta) is correctly segmented and labeled (supplementary figure 3, Appendix A). If minor gaps are present, this criterion should be scored as met.</li><li class="elsevierStyleListItem" id="lsti0035">b.Gaps (minor) are absent (supplementary figure 4, Appendix A).</li><li class="elsevierStyleListItem" id="lsti0040">c.Catheter thickness is accurate, by visual appreciation (supplementary figure 5, Appendix A).</li><li class="elsevierStyleListItem" id="lsti0045">d.Location: if parts of the catheter far from the coronary ostia (ascending and/or descending aorta) are segmented, there are no major gaps or artifacts (supplementary 6, Appendix A).</li><li class="elsevierStyleListItem" id="lsti0050">2)Vessel segmentation:</li><li class="elsevierStyleListItem" id="lsti0055">a.Main vessels are correctly segmented and labeled. For the RCA, this includes the segments from the ostium to the crux (supplementary figure 7, Appendix A). For the LCA, this includes the segments from the left main ostium to the visually discernible distal segments of the left anterior descending or the circumflex (or most important obtuse marginal branch), depending on incidence. Branches are excluded from this criterion. If minor gaps are present, this criterion should be scored as met.</li><li class="elsevierStyleListItem" id="lsti0060">b.Branch segmentation: branches with a luminal diameter of at least approximately 2 mm (using the catheter size as reference) are correctly segmented and labeled (supplementary figure 8, Appendix A). Size is estimated by visual appreciation. If minor gaps are present, this criterion should be scored as met.</li><li class="elsevierStyleListItem" id="lsti0065">c.Main vessel gaps (minor) are absent (supplementary figure 9, Appendix A).</li><li class="elsevierStyleListItem" id="lsti0070">d.Branch gaps (minor) are absent (supplementary figure 10, Appendix A).</li><li class="elsevierStyleListItem" id="lsti0075">e.Catheter to artery transition: correct labeling of the catheter tip vs. coronary artery origin (supplementary figure 11, Appendix A).</li><li class="elsevierStyleListItem" id="lsti0080">3)Artifacts</li><li class="elsevierStyleListItem" id="lsti0085">a.Coronary: No non-coronary structures are labeled as part of the coronary tree (supplementary figure 12, Appendix A).</li><li class="elsevierStyleListItem" id="lsti0090">b.Catheter: No non-catheter structures are incorrectly labeled as part of the catheter (supplementary figure 13, Appendix A).</li></ul><elsevierMultimedia ident="fig0080"></elsevierMultimedia>The criteria for these two artifacts are not applicable to the small catheter-artery transition area.To provide an objective assessment, these criteria were scored by a panel of three Interventional Cardiologists (MNM, ARF, PCF), of whom two (ARF, PCF) took no part in any stage of the annotation/training process. Discrepancies were solved by agreement. All images were graded across all groups: baseline human segmentation, enhanced human segmentation, baseline AI and enhanced AI. During the grading process, the image group was blinded.Lastly, because the abovementioned criteria are not equally important, a Global Segmentation Score (GSS, 1.5 to 100 points) was devised, taking into account the relevance of each criterion as defined by the three experts (<a class="elsevierStyleCrossRef" href="#tbl0005">Table 1</a>). The panel was also asked to select which of the two AI models was preferred for each image, regardless of the final score.<elsevierMultimedia ident="tbl0005"></elsevierMultimedia>Statistical analysisDescriptive variables are shown in absolute and relative (percentage) numbers. To assess the association between qualitative (categorical) variables the Chi-Square test was used. To assess differences in quantitative variables we used the Mann-Whitney test (two independent groups) or the Kruskal-Wallis test (multiple independent groups). A p<0.05 was used for statistical significance, except for multiple group comparisons, where we used a p<0.01. IBM SPSS Statistics 27 was used for statistical analysis.Ethical issuesThis study complies with the Declaration of Helsinki and was approved by the local ethics committee.ResultsBaseline datasetWe included 416 images from 69 patients (<a class="elsevierStyleCrossRef" href="#tbl0010">Table 2</a>). With two human and two AI datasets, 1664 processed images were generated.<elsevierMultimedia ident="tbl0010"></elsevierMultimedia>Performance assessmentNon-medical metricsResults are outlined in <a class="elsevierStyleCrossRef" href="#tbl0015">Table 3</a>. These scores indicate that enhanced AI was generally superior to baseline AI. Segmentation performance was good and consistent across arteries, as indicated by the high mean and low standard deviation of the DSC. For the catheter, performance was lower and much less consistent.<elsevierMultimedia ident="tbl0015"></elsevierMultimedia>Clinical performanceOverall performance – individual criteria assessment (Supplementary Table 1, Appendix A).Coronary segmentationThe main vessels were correctly segmented in almost all cases across groups. Minor gaps occurred rarely in the baseline human segmentation and both AI models, although there was a small but non-significant improvement with the enhanced AI vs. baseline AI.Branch segmentation was also correct almost always in all groups, albeit less so than main vessel segmentation. There was a small, yet significant, improvement with the enhanced AI vs. baseline AI.Minor branch gaps were quite common, revealing very significant differences between AI and human models. While enhanced AI performed numerically better than baseline AI, it still produced small gaps in nearly two thirds of cases.Coronary artifacts were very uncommon in human annotations and were usually minor imperfections in catheter/coronary crossovers. They were common and usually minor in both AI models, although there was a very significant improvement with the enhanced AI vs. baseline AI (14.4% vs. 25.7%).Catheter/artery transitionBaseline human segmentation failed in 12% of cases and enhanced human segmentation missed 3.8%. Baseline AI produced a higher error rate (19.7%), but enhanced AI was numerically more often correct than baseline human segmentation, sometimes correctly identifying the transition where humans failed (<a class="elsevierStyleCrossRef" href="#fig0085">Figure 4</a>).<elsevierMultimedia ident="fig0085"></elsevierMultimedia>Catheter segmentationBaseline human segmentation produced thickness imperfections (usually mildly engorged catheter) in 13.9% of cases, but otherwise, segmentation was almost always correct regarding other criteria. Baseline AI produced low error rates in main body segmentation. However, artifacts, usually quite minor and in the vicinity of coronary segments, occurred very frequently (41.1%). Another common error was catheter thickness (36.3%), often resulting in an overestimation of catheter size.Enhanced human segmentation significantly improved on thickness issues, although imperfections persisted in 6.2% of cases.Enhanced AI produced better results than the baseline AI model for catheter thickness (correct in 96.4%), also surpassing both human models (although the difference was not statistically significant when compared to the enhanced human segmentation). However, the performance of the enhanced AI otherwise decreased in all other criteria, especially regarding minor gaps, which became much more common (3.1% in the baseline AI model to 23.3%). Even main body segmentation was significantly affected, although successful in the vast majority of cases (86.5%). Despite this, in most failures catheter identification was still possible, as major gaps often occurred distally in areas of contrast backflow. There was a slight numerical worsening in artifact and location issues in enhanced AI vs. baseline AI.Overall performance – Global Segmentation Score assessment and expert preference (<a class="elsevierStyleCrossRef" href="#tbl0020">Table 4</a>)Human models outperformed AI models. Enhanced models surpassed baseline models. The difference was statistically significant for all comparisons. GSS was very high for both AI models; the enhanced AI reached an average of 90 points.<elsevierMultimedia ident="tbl0020"></elsevierMultimedia>With regards to expert preference, the enhanced AI model was preferred in 300 (72%) cases, the baseline AI model in 100 (24%) and in 16 (4%) cases no AI model was preferred.Performance according to coronary artery – individual criteria assessment (Supplementary Table 2, Appendix A)There was a trend toward better performance in the RCA, both regarding human and AI groups. The most notable and statistically significant differences occurred in catheter transition (regarding both AI models and the baseline human segmentation) and catheter segmentation (both AI models performed better in the RCA). Branch gaps were quite less frequent in the RCA with the enhanced AI model. Other differences, even if statistically significant, were very small.Performance by coronary artery – Global Segmentation Score assessment (Supplementary Table 3, Appendix A)All models scored very high for both arteries. There were very minor statistically significant differences for the baseline AI model only, favoring RCA segmentation.Considering expert preference:<ul class="elsevierStyleList" id="lis0015"><li class="elsevierStyleListItem" id="lsti0095">-RCA: Enhanced AI was preferred in 109 (68.6%) cases, the baseline AI was preferred in 43 (27%) and in 7 (4.4%) cases no AI model was preferred.</li><li class="elsevierStyleListItem" id="lsti0100">-LCA: Enhanced AI was preferred in 191 (74.3%) cases, the baseline AI was preferred in 57 (22.2%) and in 9 (3.5%) cases no AI was preferred.</li></ul>Performance according to angulation incidence – individual criteria assessment (Supplementary Tables 4 and 5, Appendix A)Given the large amount of data, there being no significant differences in the vast majority of cases and for the sake of readability, only statistically significant differences are shown in the tables. Overall, the impact of incidences on model performance was limited, and affected almost exclusively the AI models.Performance according to angulation incidence – Global Segmentation Score assessment (Supplementary Tables 6 and 7, Appendix A)Differences were minor and only statistically significant for human performance in less common incidences (PA views for the LCA and PA cranial for the RCA).DiscussionOverall considerationsBaseline human segmentation was generally correct. Catheter/coronary transition and catheter thickness errors were the most common. Poor individualization due to contrast backflow, catheter curves and human fatigue all likely contributed.Enhanced human segmentation was nearly perfect. Mild transition issues remained, highlighting the difficulty of the task. As this model was actually a combination of the best of baseline human segmentation and baseline AI, it also demonstrates how AI can help improve human performance. Even these slight human imperfections highlight the need for rigorous quality control during and after the final results, rather than assuming human annotation is a “perfect” ground truth. This an inherent limitation to the annotation of medical images, as the sheer amount of cumbersome work is error prone.Baseline AI performed CAG segmentation successfully yet was affected by the same two issues of the baseline human segmentation – transition and catheter thickness. The effort to correct these when developing the enhanced AI was fruitful in the case of transition but produced mixed results for catheter thickness. Impact on transition performance was impressive, as, at times, the enhanced AI even achieved correct assessments where humans failed (<a class="elsevierStyleCrossRef" href="#fig0085">Figure 4</a>). However, it seems the gain in catheter thickness accuracy was offset by losses in other catheter segmentation tasks. Lastly, every aspect of coronary segmentation improved in the enhanced AI, which performed better than baseline AI. The differences between the two AI models also highlight how relatively small differences in the ground truth can impact relevantly on AI training.It may seem surprising that catheter segmentation was less successful than coronary segmentation. However, while intuitively one may think that catheter segmentation is an easier task and therefore the results would have been better for this task, from a machine learning perspective that is not the case. In particular, segmentation performance is highly dependent on the frequency of each class. Rarer classes, or ones that occupy smaller areas, are interpreted by the model as being less likely to appear. Furthermore, during training, the lower the number of pixels belonging to a particular class, the lower the penalty for segmenting that class incorrectly. Even though we used a loss function designed to mitigate this phenomenon, the poorer segmentation of less common classes (the catheter, in this case) is still evident in the results.Right coronary artery segmentation was easier than LCA, however the differences were quite small and there were fewer than expected, considering its greater anatomical simplicity. Angulations also had a relatively small impact both on human and AI performance and small observed differences may be attributed to specific issues that are more common in certain incidences: contrast backflow (less problematic in PA or RAO caudal); coronary/catheter crossovers (such as spider or extreme RAO cranial – <a class="elsevierStyleCrossRef" href="#fig0090">Figure 5</a>); proximity of bone (such as RCA LAO views); smaller samples of some incidences, such as PA cranial; uncommon catheter pathways, such as the femoral approach, which sometimes produces a central vertical outline.<elsevierMultimedia ident="fig0090"></elsevierMultimedia>Globally, both AI models achieved a very high DSC, with higher performance in artery segmentation than in catheter segmentation, supporting the results of qualitative clinical assessment. When factors are weighed up based on their perceived relevance – as assessed by GSS – both performed very well. The enhanced AI scored an average of 90 points, meaning it provided 90% of what experts deemed most relevant when viewing a CAG. By all measures, the enhanced AI was the better model. However, the fact that differences between the two AI models were not large and that the enhanced AI was preferred in most, but not all cases, highlights the difficulty in improving an already good performance.Other studies with artificial intelligence applied to coronary angiography segmentation/interpretationFew studies regarding coronary artery segmentation based on AI technologies have been published in medical/biology journals to date. Yang et al.<a class="elsevierStyleCrossRef" href="#bib0180">12</a> successfully developed AI models capable of segmenting CAG. Their dataset was larger (3302 images/2042 patients) and was also annotated by two expert physicians. Different incidences were also used. They also focused exclusively on segmenting specific segments of major vessels with at least mild (>30%) stenotic lesions. Neither the branches nor the catheter were segmented, leading to a much simpler problem than the one addressed in this article.Two other works<a class="elsevierStyleCrossRefs" href="#bib0165">9,10</a> from the same baseline dataset, also developed AI-based CAG segmentation. Their dataset was also larger (4904 images from 170 videos). However, the annotations were performed by medical students and no details are provided regarding patient subset, target vessel or incidence.Very recently, Du et al.<a class="elsevierStyleCrossRef" href="#bib0175">11</a> published the results of a broad study. They focused on two tasks: CAG segmentation and special lesion morphology identification (calcium, thrombus, among others). For the former task, which overlaps with ours, they used a very large dataset of 13 373 images distributed across ten incidences (six LCA and four RCA), annotated by ten qualified analysts. This was an all-comers study, rather than focusing on patient subsets. They too annotated catheter/arteries and additionally marked different coronary segments. Their model is impressive as judged by the presented images, as they even distinguished between contrast backflow, catheter and coronary. However, they did not specify the exact criteria for segmenting the coronary tree and their exact metrics make it difficult to assess exactly how their models performed in detail regarding segmentation.While all the abovementioned groups have worked with datasets larger than ours, our study has several unique features: (1) there was medical rationale for vessel size segmentation; (2) results were assessed from a set of criteria defined by experts, capturing the quality of the segmentation from an Interventional Cardiologist's eyes; (3) human annotations were also graded, rather than assuming a perfect human ground truth; (4) specific segmentation tasks were appraised individually, enabling insights into strengths and weaknesses of AI and human models alike; (5) results were also considered globally with the GSS, by factoring the relevance of each criterion, enabling a broad, simple appreciation of the results. Furthermore, the ability to perform high-quality segmentation in a system trained using less data provides relevant evidence that more advanced AI systems can be effectively applied even in situations where the available data are limited.LimitationsThis is a single center retrospective dataset, involving a single image per projection and a smaller sample size than some previously published manuscripts. The images come from the same angiography devices (Siemens Artis) and thus we have not yet tested our models on images obtained from other equipment or image settings.We have not yet conducted formal assessment on how well the models perform in segmenting specific degrees of stenosis severity. Our models are also yet to be tested for specific vessel disease types (calcium, thrombus), clinical settings (chronic total occlusion, ST-elevation myocardial infarction).We have not yet assessed the performance of AI models on an external validation cohort. There are several reasons for this. We aimed to compare AI and human results in detail first and assess the exact performance of AI models for each segmentation task. A validation dataset would comprise a new set of images, which would not undergo human segmentation, thus impeding comparison with human performance. Also, validation implies that a metric be available for comparing results. Because the Dice methods require a ground truth human annotation for comparison, and the GSS was developed and applied for the first time for this paper, we felt a suitable metric was not yet available for performing validation prior to the current analysis. In addition, AI models are continuously and dynamically improving. As we are currently working on further testing and enhancing current AI models (view Future direction and implications section below), we felt performing external validation at this stage was premature.The exclusion of cardiac devices/cardiac surgery and other foreign objects renders our models not yet applicable to such cases. We did not, however, exclude cases with previously implanted stents.Lastly, focusing specifically on patients undergoing invasive physiology assessment may have created bias, limiting a broader application of the models to other patient subsets.We are currently working to address all these issues in future research.Future direction and implicationsCoronary angiography segmentation in itself is not end objective but rather an essential milestone for developing AI systems capable of CAG analysis and interpretation. These results should, therefore, be regarded as a first step, rather than a final deployment tool. While not yet ready for immediate clinical application, the results of both AI models are already relevant, providing a framework that can be built upon in the future.Further steps include testing the models for stenosed segments, which will be critical for clinical application. In the future, we aim to test our models with a validation cohort using new angiograms. Sub-segmentation, automatic anatomical identification and physiology are also areas for future research.We will also strengthen the capabilities of our models further by broadening our training base to other patient and lesion subsets, focusing on particular issues where there is still room for improvement, as identified by our uniquely detailed analysis.Our results also provide insight into which human tasks are most challenging, which may be of use to other researchers.Global Segmentation Score is the first of its kind for assessing the quality of segmentations in CAG. By providing a reasonably objective and quantitative clinical measurement, it can be used as a benchmark for comparing and validating results across research groups.Lastly, while conventional segmentation software does exist, it is not without limitations, and only by developing AI systems can we compare and improve both in the future. The potential implications of AI for Interventional Cardiology are immense, and we envisage a catherization lab of the future where all of these insights render the human eye more objective, thus improving patient care.ConclusionsWe successfully developed two AI models capable of good quality automatic CAG segmentation, as assessed by GDS, DSC and the GSS. From an expert's perspective, the latter and its individual criteria provided a feasible, reasonably objective and quantifiable way of assessing the results.The enhanced AI model outperformed the baseline AI model in coronary segmentation tasks as well as globally. With regards to catheter segmentation tasks, the enhanced AI model improved on the task of catheter thickness, but performed less well in other catheter segmentation tasks. Both human segmentations were superior to both AI models, but only the enhanced human segmentation, built by combining the best of baseline human segmentation and baseline AI, achieved a near perfect GSS.These results provide a relevant framework for building upon, potentially leading to future clinical application.Conflicts of interestThe authors have no conflicts of interest to declare." "textoCompletoSecciones" => array:1 [ "secciones" => array:12 [ 0 => array:3 [ "identificador" => "xres1823844" "titulo" => "Abstract" "secciones" => array:4 [ 0 => array:2 [ "identificador" => "abst0005" "titulo" => "Introduction and objectives" ] 1 => array:2 [ "identificador" => "abst0010" "titulo" => "Methods" ] 2 => array:2 [ "identificador" => "abst0015" "titulo" => "Results" ] 3 => array:2 [ "identificador" => "abst0020" "titulo" => "Conclusions" ] ] ] 1 => array:2 [ "identificador" => "xpalclavsec1590897" "titulo" => "Keywords" ] 2 => array:3 [ "identificador" => "xres1823843" "titulo" => "Resumo" "secciones" => array:4 [ 0 => array:2 [ "identificador" => "abst0025" "titulo" => "Introdução e objetivos" ] 1 => array:2 [ "identificador" => "abst0030" "titulo" => "Métodos" ] 2 => array:2 [ "identificador" => "abst0035" "titulo" => "Resultados" ] 3 => array:2 [ "identificador" => "abst0040" "titulo" => "Conclusões" ] ] ] 3 => array:2 [ "identificador" => "xpalclavsec1590898" "titulo" => "Palavras-chave" ] 4 => array:2 [ "identificador" => "sec0005" "titulo" => "Introduction" ] 5 => array:3 [ "identificador" => "sec0010" "titulo" => "Methods" "secciones" => array:8 [ 0 => array:3 [ "identificador" => "sec0015" "titulo" => "Dataset selection" "secciones" => array:4 [ 0 => array:2 [ "identificador" => "sec0020" "titulo" => "Inclusion criteria" ] 1 => array:2 [ "identificador" => "sec0025" "titulo" => "Exclusion criteria" ] 2 => array:2 [ "identificador" => "sec0030" "titulo" => "Image selection" ] 3 => array:2 [ "identificador" => "sec0035" "titulo" => "Dataset size" ] ] ] 1 => array:2 [ "identificador" => "sec0040" "titulo" => "Baseline annotation process" ] 2 => array:2 [ "identificador" => "sec0045" "titulo" => "Baseline artificial intelligence model training" ] 3 => array:2 [ "identificador" => "sec0050" "titulo" => "Enhanced human model" ] 4 => array:2 [ "identificador" => "sec0055" "titulo" => "Enhanced artificial intelligence model" ] 5 => array:3 [ "identificador" => "sec0060" "titulo" => "Performance assessment" "secciones" => array:2 [ 0 => array:2 [ "identificador" => "sec0065" "titulo" => "Non-medical metrics" ] 1 => array:2 [ "identificador" => "sec0070" "titulo" => "Clinical performance criteria" ] ] ] 6 => array:2 [ "identificador" => "sec0075" "titulo" => "Statistical analysis" ] 7 => array:2 [ "identificador" => "sec0080" "titulo" => "Ethical issues" ] ] ] 6 => array:3 [ "identificador" => "sec0085" "titulo" => "Results" "secciones" => array:11 [ 0 => array:2 [ "identificador" => "sec0090" "titulo" => "Baseline dataset" ] 1 => array:2 [ "identificador" => "sec0095" "titulo" => "Performance assessment" ] 2 => array:2 [ "identificador" => "sec0100" "titulo" => "Clinical performance" ] 3 => array:2 [ "identificador" => "sec0105" "titulo" => "Coronary segmentation" ] 4 => array:2 [ "identificador" => "sec0110" "titulo" => "Catheter/artery transition" ] 5 => array:2 [ "identificador" => "sec0115" "titulo" => "Catheter segmentation" ] 6 => array:2 [ "identificador" => "sec0120" "titulo" => "Overall performance – Global Segmentation Score assessment and expert preference (Table 4)" ] 7 => array:2 [ "identificador" => "sec0125" "titulo" => "Performance according to coronary artery – individual criteria assessment (Supplementary Table 2, Appendix A)" ] 8 => array:2 [ "identificador" => "sec0130" "titulo" => "Performance by coronary artery – Global Segmentation Score assessment (Supplementary Table 3, Appendix A)" ] 9 => array:2 [ "identificador" => "sec0135" "titulo" => "Performance according to angulation incidence – individual criteria assessment (Supplementary Tables 4 and 5, Appendix A)" ] 10 => array:2 [ "identificador" => "sec0140" "titulo" => "Performance according to angulation incidence – Global Segmentation Score assessment (Supplementary Tables 6 and 7, Appendix A)" ] ] ] 7 => array:3 [ "identificador" => "sec0145" "titulo" => "Discussion" "secciones" => array:4 [ 0 => array:2 [ "identificador" => "sec0150" "titulo" => "Overall considerations" ] 1 => array:2 [ "identificador" => "sec0155" "titulo" => "Other studies with artificial intelligence applied to coronary angiography segmentation/interpretation" ] 2 => array:2 [ "identificador" => "sec0160" "titulo" => "Limitations" ] 3 => array:2 [ "identificador" => "sec0165" "titulo" => "Future direction and implications" ] ] ] 8 => array:2 [ "identificador" => "sec0170" "titulo" => "Conclusions" ] 9 => array:2 [ "identificador" => "sec0175" "titulo" => "Conflicts of interest" ] 10 => array:2 [ "identificador" => "xack643450" "titulo" => "Acknowledgments" ] 11 => array:1 [ "titulo" => "References" ] ] ] "pdfFichero" => "main.pdf" "tienePdf" => true "fechaRecibido" => "2022-01-27" "fechaAceptado" => "2022-04-03" "PalabrasClave" => array:2 [ "en" => array:1 [ 0 => array:4 [ "clase" => "keyword" "titulo" => "Keywords" "identificador" => "xpalclavsec1590897" "palabras" => array:6 [ 0 => "Deep learning" 1 => "Artificial intelligence" 2 => "Machine learning" 3 => "Coronary angiography" 4 => "Coronary artery disease" 5 => "Percutaneous coronary intervention" ] ] ] "pt" => array:1 [ 0 => array:4 [ "clase" => "keyword" "titulo" => "Palavras-chave" "identificador" => "xpalclavsec1590898" "palabras" => array:6 [ 0 => "Aprendizagem profunda" 1 => "Inteligência artifical" 2 => "Aprendizagem de máquinas" 3 => "Coronariografia" 4 => "Doença coronária" 5 => "Intervenção coronária percutânea" ] ] ] ] "tieneResumen" => true "resumen" => array:2 [ "en" => array:3 [ "titulo" => "Abstract" "resumen" => "Introduction and objectivesAlthough automatic artificial intelligence (AI) coronary angiography (CAG) segmentation is arguably the first step toward future clinical application, it is underexplored. We aimed to (1) develop AI models for CAG segmentation and (2) assess the results using similarity scores and a set of criteria defined by expert physicians. MethodsPatients undergoing CAG were randomly selected in a retrospective study at a single center. Per incidence, an ideal frame was segmented, forming a baseline human dataset (BH), used for training a baseline AI model (BAI). Enhanced human segmentation (EH) was created by combining the best of both. An enhanced AI model (EAI) was trained using the EH. Results were assessed by experts using 11 weighted criteria, combined into a Global Segmentation Score (GSS: 0–100 points). Generalized Dice Score (GDS) and Dice Similarity Coefficient (DSC) were also used for AI models assessment. Results1664 processed images were generated. GSS for BH, EH, BAI and EAI were 96.9+/-5.7; 98.9+/-3.1; 86.1+/-10.1 and 90+/-7.6, respectively (95% confidence interval, p<0.001 for both paired and global differences). The GDS for the BAI and EAI was 0.9234±0.0361 and 0.9348±0.0284, respectively. The DSC for the coronary tree was 0.8904±0.0464 and 0.9134±0.0410 for the BAI and EAI, respectively. The EAI outperformed the BAI in all coronary segmentation tasks, but performed less well in some catheter segmentation tasks. ConclusionsWe successfully developed AI models capable of CAG segmentation, with good performance as assessed by all scores." "secciones" => array:4 [ 0 => array:2 [ "identificador" => "abst0005" "titulo" => "Introduction and objectives" ] 1 => array:2 [ "identificador" => "abst0010" "titulo" => "Methods" ] 2 => array:2 [ "identificador" => "abst0015" "titulo" => "Results" ] 3 => array:2 [ "identificador" => "abst0020" "titulo" => "Conclusions" ] ] ] "pt" => array:3 [ "titulo" => "Resumo" "resumen" => "Introdução e objetivosA segmentação automática de coronariografia (CRG) por inteligência artificial (IA) encontra-se pouco explorada na literatura médica. Os objetivos do presente estudo são (1) desenvolver modelos de IA para segmentação de CRG e (2) aferir os resultados por scores de similaridade e critérios definidos por peritos. MétodosDoentes submetidos a CRG foram retrospetivamente selecionados aleatoriamente num centro. Por incidência, segmentou-se um frame ideal, formando uma segmentação humana basal (HB), usada para treinar um modelo de IA basal (IAB). Da combinação de ambos acrescentou-se uma segmentação humana aperfeiçoada (HA), utilizada para treinar um modelo de IA aperfeiçoado (IAA). Os resultados foram aferidos com 11 critérios balanceados definidos por peritos, combinados num Scorede Segmentação Global (SSC – 0–100 pontos). O Scorede Dice Generalizado (SDG) e Score de Dice de Similaridade (SDS) aplicaram-se aos modelos de IA. ResultadosGeraram-se 1664 imagens processadas. Os SCC para a HB, HA, IAB e IAA foram 96,9+/-5,7; 98,9+/-3,1; 86,1+/-10,1 e 90+/-7,6, respetivamente (IC 95%, p<0,001 - diferenças globais e emparelhadas). O SDG para o IAB e IAA foi 0,9234±0,0361 e 0,9348±0,0284, respetivamente. O SDS foi 0,8904±0,0464 e 0,9134±0,0410 para o IAB e IAA, respetivamente. O IAA exibiu superior desempenho ao IAB para as todas tarefas de segmentação coronária, mas não para todas as de cateter. ConclusõesDesenvolvemos modelos de IA de segmentação automática de CRG, com bom desempenho de acordo com aferição por todos os scores." "secciones" => array:4 [ 0 => array:2 [ "identificador" => "abst0025" "titulo" => "Introdução e objetivos" ] 1 => array:2 [ "identificador" => "abst0030" "titulo" => "Métodos" ] 2 => array:2 [ "identificador" => "abst0035" "titulo" => "Resultados" ] 3 => array:2 [ "identificador" => "abst0040" "titulo" => "Conclusões" ] ] ] ] "apendice" => array:1 [ 0 => array:1 [ "seccion" => array:1 [ 0 => array:4 [ "apendice" => "The following are the supplementary material to this article: <elsevierMultimedia ident="fig0005"></elsevierMultimedia><elsevierMultimedia ident="fig0010"></elsevierMultimedia><elsevierMultimedia ident="fig0015"></elsevierMultimedia><elsevierMultimedia ident="fig0020"></elsevierMultimedia><elsevierMultimedia ident="fig0025"></elsevierMultimedia><elsevierMultimedia ident="fig0030"></elsevierMultimedia><elsevierMultimedia ident="fig0035"></elsevierMultimedia><elsevierMultimedia ident="fig0040"></elsevierMultimedia><elsevierMultimedia ident="fig0045"></elsevierMultimedia><elsevierMultimedia ident="fig0050"></elsevierMultimedia><elsevierMultimedia ident="fig0055"></elsevierMultimedia><elsevierMultimedia ident="fig0060"></elsevierMultimedia><elsevierMultimedia ident="fig0065"></elsevierMultimedia><elsevierMultimedia ident="upi0005"></elsevierMultimedia>" "etiqueta" => "Appendix A" "titulo" => "Supplementary material" "identificador" => "sec0185" ] ] ] ] "multimedia" => array:23 [ 0 => array:7 [ "identificador" => "fig0070" "etiqueta" => "Figure 1" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => true "mostrarDisplay" => false "figura" => array:1 [ 0 => array:4 [ "imagen" => "gr1.jpeg" "Alto" => 405 "Ancho" => 2508 "Tamanyo" => 63361 ] ] "descripcion" => array:1 [ "en" => "Segmentation model composed of an EfficientNet-B5 encoder and an EfficientUNet++ decoder." ] ] 1 => array:7 [ "identificador" => "fig0075" "etiqueta" => "Figure 2" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => true "mostrarDisplay" => false "figura" => array:1 [ 0 => array:4 [ "imagen" => "gr2.jpeg" "Alto" => 2146 "Ancho" => 2508 "Tamanyo" => 198619 ] ] "descripcion" => array:1 [ "en" => "Annotation and training process." ] ] 2 => array:7 [ "identificador" => "fig0080" "etiqueta" => "Figure 3" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => true "mostrarDisplay" => false "figura" => array:1 [ 0 => array:4 [ "imagen" => "gr3.jpeg" "Alto" => 626 "Ancho" => 1305 "Tamanyo" => 45030 ] ] "descripcion" => array:1 [ "en" => "A segmentation case fulfilling all 11 criteria." ] ] 3 => array:7 [ "identificador" => "fig0085" "etiqueta" => "Figure 4" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => true "mostrarDisplay" => false "figura" => array:1 [ 0 => array:4 [ "imagen" => "gr4.jpeg" "Alto" => 752 "Ancho" => 1505 "Tamanyo" => 66385 ] ] "descripcion" => array:1 [ "en" => "(left to right): The first human segmentation incorrectly labels contrast backflow as coronary. The baseline AI model improves on the human segmentation but is still not perfect. The enhanced human model segments the transition perfectly. The enhanced AI model is hampered in catheter segmentation but identifies the transition correctly." ] ] 4 => array:7 [ "identificador" => "fig0090" "etiqueta" => "Figure 5" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => true "mostrarDisplay" => false "figura" => array:1 [ 0 => array:4 [ "imagen" => "gr5.jpeg" "Alto" => 1305 "Ancho" => 1305 "Tamanyo" => 103564 ] ] "descripcion" => array:1 [ "en" => "Crossovers in spider (above) and extreme RAO cranial (below) views generating artifacts." ] ] 5 => array:8 [ "identificador" => "tbl0005" "etiqueta" => "Table 1" "tipo" => "MULTIMEDIATABLA" "mostrarFloat" => true "mostrarDisplay" => false "detalles" => array:1 [ 0 => array:3 [ "identificador" => "at1" "detalle" => "Table " "rol" => "short" ] ] "tabla" => array:1 [ "tablatextoimagen" => array:1 [ 0 => array:1 [ "tabla" => array:1 [ 0 => """ <table border="0" frame="\n \t\t\t\t\tvoid\n \t\t\t\t" class=""><thead title="thead"><tr title="table-row"><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Criteria \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Catheter vs. CoronaryRelative Weight \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Individual CriteriaRelative Weight \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Points \t\t\t\t\t\t\n \t\t\t\t\t\t</th></tr></thead><tbody title="tbody"><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Main vessel segmentation \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">70% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">40% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">28.0 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Main vessel gaps \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">10% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">7.0 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Catheter to artery transition \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">15% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">10.5 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Branch segmentation \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">20% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">14.0 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">BranchGaps \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">5% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">3.5 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Coronary artifacts \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">10% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">7.0 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Catheter segmentation \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">30% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">40% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">12.0 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Catheter gaps \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">10% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">3.0 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Catheter artifacts \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">15% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">4.5 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Catheter location \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">5% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">1.5 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Catheter thickness \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">30% \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">9.0 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Total \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">100 \t\t\t\t\t\t\n \t\t\t\t</td></tr></tbody></table> """ ] ] ] ] "descripcion" => array:1 [ "en" => "scoring metrics for application of the Global Segmentation Score." ] ] 6 => array:8 [ "identificador" => "tbl0010" "etiqueta" => "Table 2" "tipo" => "MULTIMEDIATABLA" "mostrarFloat" => true "mostrarDisplay" => false "detalles" => array:1 [ 0 => array:3 [ "identificador" => "at2" "detalle" => "Table " "rol" => "short" ] ] "tabla" => array:2 [ "leyenda" => "CAG: coronary angiography." "tablatextoimagen" => array:1 [ 0 => array:1 [ "tabla" => array:1 [ 0 => """ <table border="0" frame="\n \t\t\t\t\tvoid\n \t\t\t\t" class=""><thead title="thead"><tr title="table-row"><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Factor \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">N+/-SD or N(%) \t\t\t\t\t\t\n \t\t\t\t\t\t</th></tr></thead><tbody title="tbody"><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Age \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">67+/-11 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Sex (male) \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">54 (78%) \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Hypertension \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">56 (81.2%) \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Diabetes mellitus \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">27 (39.1%) \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Dyslipidemia \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">39 (56.5%) \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Smoker (past or present) \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">26 (37.7%) \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Chronic coronary syndromes \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">50 (72.5%) \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Acute coronary syndrome \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">19 (27.5%) \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Revascularization during/after CAG \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">21 (30.4%) \t\t\t\t\t\t\n \t\t\t\t</td></tr></tbody></table> """ ] ] ] ] "descripcion" => array:1 [ "en" => "Baseline clinical characteristics of patients from whom images were analyzed." ] ] 7 => array:8 [ "identificador" => "tbl0015" "etiqueta" => "Table 3" "tipo" => "MULTIMEDIATABLA" "mostrarFloat" => true "mostrarDisplay" => false "detalles" => array:1 [ 0 => array:3 [ "identificador" => "at3" "detalle" => "Table " "rol" => "short" ] ] "tabla" => array:2 [ "leyenda" => "BAI: baseline AI model; GDS: Generalized Dice Score; EAI; enhanced AI model." "tablatextoimagen" => array:1 [ 0 => array:1 [ "tabla" => array:1 [ 0 => """ <table border="0" frame="\n \t\t\t\t\tvoid\n \t\t\t\t" class=""><thead title="thead"><tr title="table-row"><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black"> \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BAI \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EAI \t\t\t\t\t\t\n \t\t\t\t\t\t</th></tr></thead><tbody title="tbody"><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">GDS \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">0.9234±0.0361 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">0.9348±0.0284 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Artery DSC \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">0.8904±0.0464 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">0.9134±0.0410 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Catheter DSC \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">0.7526±0.1998 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">0.7975±0.1836 \t\t\t\t\t\t\n \t\t\t\t</td></tr></tbody></table> """ ] ] ] ] "descripcion" => array:1 [ "en" => "Generalized Dice Score and class-wise Dice Similarity Coefficient obtained by the baseline and enhanced AI models. Results presented as mean ± standard deviation." ] ] 8 => array:8 [ "identificador" => "tbl0020" "etiqueta" => "Table 4" "tipo" => "MULTIMEDIATABLA" "mostrarFloat" => true "mostrarDisplay" => false "detalles" => array:1 [ 0 => array:3 [ "identificador" => "at4" "detalle" => "Table " "rol" => "short" ] ] "tabla" => array:3 [ "leyenda" => "BAI: baseline AI model; BH: baseline human model; EAI: enhanced AI model; EH: enhanced human model; GSS: Global Segmentation Score; IQR: interquartile range; SD: standard deviation." "tablatextoimagen" => array:1 [ 0 => array:1 [ "tabla" => array:1 [ 0 => """ <table border="0" frame="\n \t\t\t\t\tvoid\n \t\t\t\t" class=""><thead title="thead"><tr title="table-row"><th class="td-with-role" title="\n \t\t\t\t\ttable-head\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col">GSS \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " colspan="4" align="center" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Group</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " colspan="7" align="center" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">p-value</th></tr><tr title="table-row"><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black"> \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BH \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EH \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BAI \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EAI \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Between all<a class="elsevierStyleCrossRef" href="#tblfn0005">*</a> \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BH vs EH<a class="elsevierStyleCrossRef" href="#tblfn0010">**</a> \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BAI vs EAI<a class="elsevierStyleCrossRef" href="#tblfn0010">**</a> \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BH vs BAI<a class="elsevierStyleCrossRef" href="#tblfn0010">**</a> \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EH vs EAI<a class="elsevierStyleCrossRef" href="#tblfn0010">**</a> \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BH vs EAI<a class="elsevierStyleCrossRef" href="#tblfn0010">**</a> \t\t\t\t\t\t\n \t\t\t\t\t\t</th><th class="td" title="\n \t\t\t\t\ttable-head\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EH vs BAI<a class="elsevierStyleCrossRef" href="#tblfn0010">**</a> \t\t\t\t\t\t\n \t\t\t\t\t\t</th></tr></thead><tbody title="tbody"><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Mean+/-SD \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">96.9+/-5.7 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">98.9+/-3.1 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">86.1+/-10.1 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">90+/-7.6 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t"><0.001 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t"><0.001 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t"><0.001 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t"><0.001 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t"><0.001 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t"><0.001 \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t"><0.001 \t\t\t\t\t\t\n \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">Median (IQR) \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">100 (9) \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">100(0) \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">87.5 (9) \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="left" valign="\n \t\t\t\t\ttop\n \t\t\t\t">92 (9.5) \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td><td class="td" title="\n \t\t\t\t\ttable-entry\n \t\t\t\t " align="" valign="\n \t\t\t\t\ttop\n \t\t\t\t"> \t\t\t\t\t\t\n \t\t\t\t</td></tr></tbody></table> """ ] ] ] "notaPie" => array:2 [ 0 => array:3 [ "identificador" => "tblfn0005" "etiqueta" => "*" "nota" => "Kruskal-Wallis Test." ] 1 => array:3 [ "identificador" => "tblfn0010" "etiqueta" => "**" "nota" => "Mann-Whitney Test." ] ] ] "descripcion" => array:1 [ "en" => "Performance by group according to Global Segmentation Score (significance at p<0.05 for paired differences and p<0.01 for multiple comparisons)." ] ] 9 => array:5 [ "identificador" => "upi0005" "tipo" => "MULTIMEDIAECOMPONENTE" "mostrarFloat" => false "mostrarDisplay" => true "Ecomponente" => array:2 [ "fichero" => "mmc14.doc" "ficheroTamanyo" => 43765 ] ] 10 => array:7 [ "identificador" => "fig0005" "etiqueta" => "Supplementary Figure 1" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc1.jpeg" "Alto" => 648 "Ancho" => 1300 "Tamanyo" => 51205 ] ] "descripcion" => array:1 [ "en" => "A test case with too many annotated vessels (not used for training)." ] ] 11 => array:7 [ "identificador" => "fig0010" "etiqueta" => "Supplementary Figure 2" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc2.jpeg" "Alto" => 650 "Ancho" => 1300 "Tamanyo" => 39593 ] ] "descripcion" => array:1 [ "en" => "A test case with too few annotations (not used for training)." ] ] 12 => array:7 [ "identificador" => "fig0015" "etiqueta" => "Supplementary Figure 3" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc3.jpeg" "Alto" => 651 "Ancho" => 1300 "Tamanyo" => 51620 ] ] "descripcion" => array:1 [ "en" => "Contrast backflow leads the artificial intelligence model to disregard most of the catheter." ] ] 13 => array:7 [ "identificador" => "fig0020" "etiqueta" => "Supplementary Figure 4" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc4.jpeg" "Alto" => 653 "Ancho" => 1300 "Tamanyo" => 46187 ] ] "descripcion" => array:1 [ "en" => "Minor catheter gaps, possibly facilitated by contrast backflow." ] ] 14 => array:7 [ "identificador" => "fig0025" "etiqueta" => "Supplementary Figure 5" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc5.jpeg" "Alto" => 649 "Ancho" => 1300 "Tamanyo" => 46546 ] ] "descripcion" => array:1 [ "en" => "Catheter thickness is overestimated." ] ] 15 => array:7 [ "identificador" => "fig0030" "etiqueta" => "Supplementary Figure 6" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc6.jpeg" "Alto" => 650 "Ancho" => 1300 "Tamanyo" => 53256 ] ] "descripcion" => array:1 [ "en" => "A small part of the catheter in the descending aorta was segmented in a femoral access case." ] ] 16 => array:7 [ "identificador" => "fig0035" "etiqueta" => "Supplementary Figure 7" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc7.jpeg" "Alto" => 650 "Ancho" => 1300 "Tamanyo" => 47821 ] ] "descripcion" => array:1 [ "en" => "A major part of the left main was not segmented." ] ] 17 => array:7 [ "identificador" => "fig0040" "etiqueta" => "Supplementary Figure 8" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc8.jpeg" "Alto" => 650 "Ancho" => 1300 "Tamanyo" => 43312 ] ] "descripcion" => array:1 [ "en" => "The posterolateral branch and much of the posterior descending artery were missed. The enhanced artificial intelligence model did not miss these vessels." ] ] 18 => array:7 [ "identificador" => "fig0045" "etiqueta" => "Supplementary Figure 9" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc9.jpeg" "Alto" => 650 "Ancho" => 1300 "Tamanyo" => 48885 ] ] "descripcion" => array:1 [ "en" => "A small gap is visible in the left anterior descending artery." ] ] 19 => array:7 [ "identificador" => "fig0050" "etiqueta" => "Supplementary Figure 10" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc10.jpeg" "Alto" => 650 "Ancho" => 1300 "Tamanyo" => 51326 ] ] "descripcion" => array:1 [ "en" => "Small gaps are visible in branches." ] ] 20 => array:7 [ "identificador" => "fig0055" "etiqueta" => "Supplementary Figure 11" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc11.jpeg" "Alto" => 651 "Ancho" => 1300 "Tamanyo" => 46390 ] ] "descripcion" => array:1 [ "en" => "Contrast backflow renders the transition less discernible, leading the model to miss the transition zone." ] ] 21 => array:7 [ "identificador" => "fig0060" "etiqueta" => "Supplementary Figure 12" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc12.jpeg" "Alto" => 650 "Ancho" => 1300 "Tamanyo" => 44232 ] ] "descripcion" => array:1 [ "en" => "A part of the intervertebral disk and vertebra are mislabeled as coronary." ] ] 22 => array:7 [ "identificador" => "fig0065" "etiqueta" => "Supplementary Figure 13" "tipo" => "MULTIMEDIAFIGURA" "mostrarFloat" => false "mostrarDisplay" => true "figura" => array:1 [ 0 => array:4 [ "imagen" => "mmc13.jpeg" "Alto" => 650 "Ancho" => 1300 "Tamanyo" => 47287 ] ] "descripcion" => array:1 [ "en" => "Contrast backflow is mislabeled as catheter." ] ] ] "bibliografia" => array:2 [ "titulo" => "References" "seccion" => array:1 [ 0 => array:2 [ "identificador" => "bibs0015" "bibliografiaReferencia" => array:24 [ 0 => array:3 [ "identificador" => "bib0125" "etiqueta" => "1" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Phenomapping for novel classification of heart failure with preserved ejection fraction" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "S.J. Shah" 1 => "D.H. Katz" 2 => "S. Selvaraj" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1161/CIRCULATIONAHA.114.010637" "Revista" => array:6 [ "tituloSerie" => "Circulation" "fecha" => "2015" "volumen" => "131" "paginaInicial" => "269" "paginaFinal" => "279" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/25398313" "web" => "Medline" ] ] ] ] ] ] ] ] 1 => array:3 [ "identificador" => "bib0130" "etiqueta" => "2" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Early experience with IBM Watson for Oncology (WFO) cognitive computing system for lung and colorectal cancer treatment" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "S.P. Somashekhar" 1 => "M.-J. Sepúlveda" 2 => "A.D. Norden" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "Revista" => array:5 [ "tituloSerie" => "J Clin Oncol" "fecha" => "2017" "volumen" => "35" "numero" => "suppl." "paginaInicial" => "8527" ] ] ] ] ] ] 2 => array:3 [ "identificador" => "bib0135" "etiqueta" => "3" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Acute myocardial infarction on YouTube – is it all fake news?" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "I. Fialho" 1 => "M. Beringuilho" 2 => "D. Madeira" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "Revista" => array:5 [ "tituloSerie" => "Rev Port Cardiol (English Ed)" "fecha" => "2021" "volumen" => "40" "paginaInicial" => "815" "paginaFinal" => "825" ] ] ] ] ] ] 3 => array:3 [ "identificador" => "bib0140" "etiqueta" => "4" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "A.Y. Hannun" 1 => "P. Rajpurkar" 2 => "M. Haghpanahi" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "Revista" => array:4 [ "tituloSerie" => "Nat Med" "fecha" => "2019" "paginaInicial" => "65" "paginaFinal" => "69" ] ] ] ] ] ] 4 => array:3 [ "identificador" => "bib0145" "etiqueta" => "5" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Machine-learning algorithms to automate morphological and functional assessments in 2D echocardiography" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "S. Narula" 1 => "K. Shameer" 2 => "A.M. Salem Omar" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1016/j.jacc.2016.08.062" "Revista" => array:6 [ "tituloSerie" => "J Am Coll Cardiol" "fecha" => "2016" "volumen" => "68" "paginaInicial" => "2287" "paginaFinal" => "2295" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/27884247" "web" => "Medline" ] ] ] ] ] ] ] ] 5 => array:3 [ "identificador" => "bib0150" "etiqueta" => "6" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Automated echocardiographic quantification of left ventricular ejection fraction without volume measurements using a machine learning algorithm mimicking a human expert" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "F.M. Asch" 1 => "N. Poilvert" 2 => "T. Abraham" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1161/CIRCIMAGING.119.009303" "Revista" => array:5 [ "tituloSerie" => "Circ Cardiovasc Imaging" "fecha" => "2019" "volumen" => "12" "paginaInicial" => "e009303" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/31522550" "web" => "Medline" ] ] ] ] ] ] ] ] 6 => array:3 [ "identificador" => "bib0155" "etiqueta" => "7" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Combining deep learning and level set for the automated segmentation of the left ventricle of the heart from cardiac cine magnetic resonance" "autores" => array:1 [ 0 => array:2 [ "etal" => false "autores" => array:3 [ 0 => "T.A. Ngo" 1 => "Z. Lu" 2 => "G. Carneiro" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1016/j.media.2016.05.009" "Revista" => array:6 [ "tituloSerie" => "Med Image Anal" "fecha" => "2017" "volumen" => "35" "paginaInicial" => "159" "paginaFinal" => "171" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/27423113" "web" => "Medline" ] ] ] ] ] ] ] ] 7 => array:3 [ "identificador" => "bib0160" "etiqueta" => "8" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Automated cardiovascular magnetic resonance image analysis with fully convolutional networks 08 Information and Computing Sciences 0801 Artificial Intelligence and Image Processing" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "W. Bai" 1 => "M. Sinclair" 2 => "G. Tarroni" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "Revista" => array:4 [ "tituloSerie" => "J Cardiovasc Magn Reson" "fecha" => "2018" "volumen" => "20" "paginaInicial" => "65" ] ] ] ] ] ] 8 => array:3 [ "identificador" => "bib0165" "etiqueta" => "9" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Coronary artery segmentation in angiographic videos utilizing spatial-temporal information" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "L. Wang" 1 => "D. Liang" 2 => "X. Yin" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1186/s12880-019-0399-0" "Revista" => array:6 [ "tituloSerie" => "BMC Med Imaging" "fecha" => "2020" "volumen" => "20" "paginaInicial" => "1" "paginaFinal" => "10" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/31892314" "web" => "Medline" ] ] ] ] ] ] ] ] 9 => array:3 [ "identificador" => "bib0170" "etiqueta" => "10" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Coronary angiography video segmentation method for assisting cardiovascular disease interventional treatment" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "D. Liang" 1 => "J. Qiu" 2 => "L. Wang" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1186/s12880-019-0399-0" "Revista" => array:6 [ "tituloSerie" => "BMC Med Imaging" "fecha" => "2020" "volumen" => "20" "paginaInicial" => "1" "paginaFinal" => "8" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/31892314" "web" => "Medline" ] ] ] ] ] ] ] ] 10 => array:3 [ "identificador" => "bib0175" "etiqueta" => "11" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Training and validation of a deep learning architecture for the automatic analysis of coronary angiography" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "T. Du" 1 => "L. Xie" 2 => "H. Zhang" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.4244/EIJ-D-20-00570" "Revista" => array:6 [ "tituloSerie" => "EuroIntervention" "fecha" => "2021" "volumen" => "17" "paginaInicial" => "32" "paginaFinal" => "40" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/32830647" "web" => "Medline" ] ] ] ] ] ] ] ] 11 => array:3 [ "identificador" => "bib0180" "etiqueta" => "12" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Deep learning segmentation of major vessels in X-ray coronary angiography" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "S. Yang" 1 => "J. Kweon" 2 => "J.H. Roh" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1038/s41598-018-37186-2" "Revista" => array:6 [ "tituloSerie" => "Sci Rep" "fecha" => "2019" "volumen" => "9" "paginaInicial" => "1" "paginaFinal" => "11" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/30626917" "web" => "Medline" ] ] ] ] ] ] ] ] 12 => array:3 [ "identificador" => "bib0185" "etiqueta" => "13" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Implementing machine learning in interventional cardiology: the benefits are worth the trouble" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "W. Ben Ali" 1 => "A. Pesaranghader" 2 => "R. Avram" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "Revista" => array:3 [ "tituloSerie" => "Front Cardiovasc Med" "fecha" => "2021" "paginaInicial" => "1775" ] ] ] ] ] ] 13 => array:3 [ "identificador" => "bib0190" "etiqueta" => "14" "referencia" => array:1 [ 0 => array:1 [ "referenciaCompleta" => "Gonzalez & Woods, Digital image processing, 4th ed. | Pearson [Internet]. Available from: <a target="_blank" href="https://www.pearson.com/us/higher-education/program/Gonzalez-Digital-Image-Processing-4th-Edition/PGM241219.html">https://www.pearson.com/us/higher-education/program/Gonzalez-Digital-Image-Processing-4th-Edition/PGM241219.html</a> [cited 02.09.21]" ] ] ] 14 => array:3 [ "identificador" => "bib0195" "etiqueta" => "15" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Use of the instantaneous wave-free ratio or fractional flow reserve in PCI" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "J.E. Davies" 1 => "S. Sen" 2 => "H.-M. Dehbi" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1056/NEJMoa1700445" "Revista" => array:6 [ "tituloSerie" => "N Engl J Med" "fecha" => "2017" "volumen" => "376" "paginaInicial" => "1824" "paginaFinal" => "1834" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/28317458" "web" => "Medline" ] ] ] ] ] ] ] ] 15 => array:3 [ "identificador" => "bib0200" "etiqueta" => "16" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Instantaneous wave-free ratio versus fractional flow reserve to guide PCI" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "M. Götberg" 1 => "E.H. Christiansen" 2 => "I.J. Gudmundsdottir" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1056/NEJMoa1616540" "Revista" => array:6 [ "tituloSerie" => "N Engl J Med" "fecha" => "2017" "volumen" => "376" "paginaInicial" => "1813" "paginaFinal" => "1823" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/28317438" "web" => "Medline" ] ] ] ] ] ] ] ] 16 => array:3 [ "identificador" => "bib0205" "etiqueta" => "17" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Coronary angiography image segmentation based on PSPNet" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "X. Zhu" 1 => "Z. Cheng" 2 => "S. Wang" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1016/j.cmpb.2020.105897" "Revista" => array:5 [ "tituloSerie" => "Comput Methods Programs Biomed" "fecha" => "2021" "volumen" => "200" "paginaInicial" => "105897" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/33317873" "web" => "Medline" ] ] ] ] ] ] ] ] 17 => array:3 [ "identificador" => "bib0210" "etiqueta" => "18" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "et: Nested encoder–decoder architecture for the main vessel segmentation in coronary angiography" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "T.J. Jun" 1 => "J. Kweon" 2 => "Y.H. Kim" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1016/j.neunet.2020.05.002" "Revista" => array:6 [ "tituloSerie" => "Neural Networks" "fecha" => "2020" "volumen" => "128" "paginaInicial" => "216" "paginaFinal" => "233" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/32447265" "web" => "Medline" ] ] ] ] ] ] ] ] 18 => array:3 [ "identificador" => "bib0215" "etiqueta" => "19" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Treatment of very small de novo coronary artery disease with 2.0 mm drug-coated balloons showed 1-year clinical outcome comparable with 2.0 mm drug-eluting stents [Internet]" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "H.W. Sim" 1 => "R. Ananthakrishna" 2 => "S.P. Chan" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "Revista" => array:6 [ "tituloSerie" => "J Invasive Cardiol" "fecha" => "2018" "volumen" => "30" "paginaInicial" => "256" "paginaFinal" => "261" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/29656281" "web" => "Medline" ] ] ] ] ] ] ] ] 19 => array:3 [ "identificador" => "bib0220" "etiqueta" => "20" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Small-vessel treatment with contemporary newer-generation drug-eluting coronary stents in all-comers: Insights from 2-year DUTCH PEERS (TWENTE II) randomized trial" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "L.C.H. van der" 1 => "K. MM" 2 => "D. PW" ] ] ] ] ] "host" => array:1 [ 0 => array:2 [ "doi" => "10.1016/j.ahj.2016.02.020" "Revista" => array:6 [ "tituloSerie" => "Am Heart J" "fecha" => "2016" "volumen" => "176" "paginaInicial" => "28" "paginaFinal" => "35" "link" => array:1 [ 0 => array:2 [ "url" => "https://www.ncbi.nlm.nih.gov/pubmed/27264217" "web" => "Medline" ] ] ] ] ] ] ] ] 20 => array:3 [ "identificador" => "bib0225" "etiqueta" => "21" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "U-Net: convolutional networks for biomedical image segmentation" "autores" => array:1 [ 0 => array:2 [ "etal" => false "autores" => array:3 [ 0 => "O. Ronneberger" 1 => "P. Fischer" 2 => "T. Brox" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "Revista" => array:5 [ "tituloSerie" => "Lect Notes Comput Sci (including Subser Lect Notes Artif Intell Lect Notes Bioinformatics)" "fecha" => "2015" "volumen" => "9351" "paginaInicial" => "234" "paginaFinal" => "241" ] ] ] ] ] ] 21 => array:3 [ "identificador" => "bib0230" "etiqueta" => "22" "referencia" => array:1 [ 0 => array:1 [ "referenciaCompleta" => "Silva JL, Nobre Menezes M, Rodrigues T, et al. Encoder-decoder architectures for clinically relevant coronary artery segmentation; 2021 Jun 21, arXiv:210611447 [eessIV]." ] ] ] 22 => array:3 [ "identificador" => "bib0235" "etiqueta" => "23" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "EfficientNet: rethinking model scaling for convolutional neural networks" "autores" => array:1 [ 0 => array:2 [ "etal" => false "autores" => array:2 [ 0 => "M. Tan" 1 => "Q.V. Le" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "LibroEditado" => array:5 [ "titulo" => "36th Int Conf Mach Learn ICML 2019" "paginaInicial" => "10691" "paginaFinal" => "10700" "conferencia" => "International Machine Learning Society (IMLS); 2019 May 28" "serieFecha" => "2019 June" ] ] ] ] ] ] 23 => array:3 [ "identificador" => "bib0240" "etiqueta" => "24" "referencia" => array:1 [ 0 => array:2 [ "contribucion" => array:1 [ 0 => array:2 [ "titulo" => "Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations" "autores" => array:1 [ 0 => array:2 [ "etal" => true "autores" => array:3 [ 0 => "C.H. Sudre" 1 => "W. Li" 2 => "T. Vercauteren" ] ] ] ] ] "host" => array:1 [ 0 => array:1 [ "LibroEditado" => array:3 [ "titulo" => "Deep Learn Med image Anal multimodal Learn Clin Decis Support Third Int Work DLMIA 2017, 7th Int Work ML-CDS 2017, held conjunction with MICCAI 2017 Quebec City, QC, vol. 2017" "paginaInicial" => "240" "serieFecha" => "2017" ] ] ] ] ] ] ] ] ] ] "agradecimientos" => array:1 [ 0 => array:4 [ "identificador" => "xack643450" "titulo" => "Acknowledgments" "texto" => "MNM was responsible for conceptualization, data gathering, processing and analysis, interpretating the results and paper drafting. JLS was responsible for technical and AI tasks, data and image processing, model implementation and training. TR and BS were responsible for data gathering, processing and analysis. ARF and PCF were responsible for data analysis and results interpretation. ALO was responsible for supervising the work of JLS, and participated directly in the same tasks. FJP was responsible for supervising the work of MNM, and participated directly in the same tasks.All authors revised the paper critically for important intellectual content, gave final approval for its publication and agreed to be accountable for all respects of its accuracy and integrity." "vista" => "all" ] ] ] "idiomaDefecto" => "en" "url" => "/08702551/0000004100000012/v2_202212230532/S0870255122001354/v2_202212230532/en/main.assets" "Apartado" => array:4 [ "identificador" => "93359" "tipo" => "SECCION" "pt" => array:2 [ "titulo" => "Original articles" "idiomaDefecto" => true ] "idiomaDefecto" => "pt" ] "PDF" => "https://static.elsevier.es/multimedia/08702551/0000004100000012/v2_202212230532/S0870255122001354/v2_202212230532/en/main.pdf?idApp=UINPBA00004E&text.app=https://revportcardiol.org/" "EPUB" => "https://multimedia.elsevier.es/PublicationsMultimediaV1/item/epub/S0870255122001354?idApp=UINPBA00004E"]

Estatísticas

Original Article

Desenvolvimento de modelos de deep learning para segmentação de coronariografias: aferição de qualidade por um novo modelo de segmentação global e comparação com desempenho humano

Miguel Nobre Menezesa,b,

Autor para correspondência

mnmenezes.gm@gmail.com

Corresponding author.

, João Lourenço-Silvac, Beatriz Silvaa,b, Tiago Rodriguesa,b, Ana Rita G. Franciscoa,b, Pedro Carrilho Ferreiraa,b, Arlindo L. Oliveirac, Fausto J. Pintoa,b

a Structural and Coronary Heart Disease Unit, Cardiovascular Center of the University of Lisbon, Faculdade de Medicina, Universidade de Lisboa, Lisboa, Portugal

b Serviço de Cardiologia, Departamento de Coração e Vasos, CHULN Hospital de Santa Maria, Lisboa, Portugal

c INESC-ID/Instituto Superior Técnico, University of Lisbon, Portugal

Lido

2794

Vezes
que se leu este artigo

1137

Total PDF

1657

Total HTML

Compartilhar estatísticas

 array:24 [
  "pii" => "S0870255122001354"
  "issn" => "08702551"
  "doi" => "10.1016/j.repc.2022.04.001"
  "estado" => "S300"
  "fechaPublicacion" => "2022-12-01"
  "aid" => "1927"
  "copyright" => "Sociedade Portuguesa de Cardiologia"
  "copyrightAnyo" => "2022"
  "documento" => "article"
  "crossmark" => 1
  "licencia" => "http://creativecommons.org/licenses/by-nc-nd/4.0/"
  "subdocumento" => "fla"
  "cita" => "Rev Port Cardiol. 2022;41:1011-21"
  "abierto" => array:3 [
    "ES" => true
    "ES2" => true
    "LATM" => true
  ]
  "gratuito" => true
  "lecturas" => array:1 [
    "total" => 0
  ]
  "itemSiguiente" => array:19 [
    "pii" => "S0870255122002864"
    "issn" => "08702551"
    "doi" => "10.1016/j.repc.2022.07.007"
    "estado" => "S300"
    "fechaPublicacion" => "2022-12-01"
    "aid" => "2006"
    "copyright" => "Sociedade Portuguesa de Cardiologia"
    "documento" => "simple-article"
    "crossmark" => 1
    "licencia" => "http://creativecommons.org/licenses/by-nc-nd/4.0/"
    "subdocumento" => "dis"
    "cita" => "Rev Port Cardiol. 2022;41:1023-4"
    "abierto" => array:3 [
      "ES" => true
      "ES2" => true
      "LATM" => true
    ]
    "gratuito" => true
    "lecturas" => array:1 [
      "total" => 0
    ]
    "en" => array:10 [
      "idiomaDefecto" => true
      "cabecera" => "<span class="elsevierStyleTextfn">Editorial comment</span>"
      "titulo" => "Machine learning &#8211; coming soon to a cath lab near you&#63; Or &#8216;don&#8217;t hold your breath&#8217;&#63;"
      "tienePdf" => "en"
      "tieneTextoCompleto" => "en"
      "paginas" => array:1 [
        0 => array:2 [
          "paginaInicial" => "1023"
          "paginaFinal" => "1024"
        ]
      ]
      "titulosAlternativos" => array:1 [
        "pt" => array:1 [
          "titulo" => "<span class="elsevierStyleItalic">Machine learning</span> &#8211; brevemente num laborat&#243;rio de hemodin&#226;mica perto de si&#63; Ou &#233; melhor esperar sentado&#63;"
        ]
      ]
      "contieneTextoCompleto" => array:1 [
        "en" => true
      ]
      "contienePdf" => array:1 [
        "en" => true
      ]
      "autores" => array:1 [
        0 => array:2 [
          "autoresLista" => "Ant&#243;nio Miguel Ferreira"
          "autores" => array:1 [
            0 => array:2 [
              "nombre" => "Ant&#243;nio Miguel"
              "apellidos" => "Ferreira"
            ]
          ]
        ]
      ]
    ]
    "idiomaDefecto" => "en"
    "EPUB" => "https://multimedia.elsevier.es/PublicationsMultimediaV1/item/epub/S0870255122002864?idApp=UINPBA00004E"
    "url" => "/08702551/0000004100000012/v2_202212230532/S0870255122002864/v2_202212230532/en/main.assets"
  ]
  "itemAnterior" => array:19 [
    "pii" => "S0870255122002876"
    "issn" => "08702551"
    "doi" => "10.1016/j.repc.2022.07.008"
    "estado" => "S300"
    "fechaPublicacion" => "2022-12-01"
    "aid" => "2007"
    "copyright" => "Sociedade Portuguesa de Cardiologia"
    "documento" => "simple-article"
    "crossmark" => 1
    "licencia" => "http://creativecommons.org/licenses/by-nc-nd/4.0/"
    "subdocumento" => "dis"
    "cita" => "Rev Port Cardiol. 2022;41:1009-10"
    "abierto" => array:3 [
      "ES" => true
      "ES2" => true
      "LATM" => true
    ]
    "gratuito" => true
    "lecturas" => array:1 [
      "total" => 0
    ]
    "en" => array:10 [
      "idiomaDefecto" => true
      "cabecera" => "<span class="elsevierStyleTextfn">Editorial comment</span>"
      "titulo" => "Thromboembolic risk in mitral stenosis&#58; Are we underestimating the potential role of inflammation&#63;"
      "tienePdf" => "en"
      "tieneTextoCompleto" => "en"
      "paginas" => array:1 [
        0 => array:2 [
          "paginaInicial" => "1009"
          "paginaFinal" => "1010"
        ]
      ]
      "titulosAlternativos" => array:1 [
        "pt" => array:1 [
          "titulo" => "Risco trombo-emb&#243;lico na estenose mitral&#58; estaremos a subestimar o papel da inflama&#231;&#227;o&#63;"
        ]
      ]
      "contieneTextoCompleto" => array:1 [
        "en" => true
      ]
      "contienePdf" => array:1 [
        "en" => true
      ]
      "autores" => array:1 [
        0 => array:2 [
          "autoresLista" => "Marisa Trabulo"
          "autores" => array:1 [
            0 => array:2 [
              "nombre" => "Marisa"
              "apellidos" => "Trabulo"
            ]
          ]
        ]
      ]
    ]
    "idiomaDefecto" => "en"
    "EPUB" => "https://multimedia.elsevier.es/PublicationsMultimediaV1/item/epub/S0870255122002876?idApp=UINPBA00004E"
    "url" => "/08702551/0000004100000012/v2_202212230532/S0870255122002876/v2_202212230532/en/main.assets"
  ]
  "en" => array:21 [
    "idiomaDefecto" => true
    "cabecera" => "<span class="elsevierStyleTextfn">Original Article</span>"
    "titulo" => "Development of deep learning segmentation models for coronary X-ray angiography&#58; Quality assessment by a new global segmentation score and comparison with human performance"
    "tieneTextoCompleto" => true
    "paginas" => array:1 [
      0 => array:2 [
        "paginaInicial" => "1011"
        "paginaFinal" => "1021"
      ]
    ]
    "autores" => array:1 [
      0 => array:4 [
        "autoresLista" => "Miguel Nobre Menezes, Jo&#227;o Louren&#231;o-Silva, Beatriz Silva, Tiago Rodrigues, Ana Rita G&#46; Francisco, Pedro Carrilho Ferreira, Arlindo L&#46; Oliveira, Fausto J&#46; Pinto"
        "autores" => array:8 [
          0 => array:4 [
            "nombre" => "Miguel"
            "apellidos" => "Nobre Menezes"
            "email" => array:1 [
              0 => "mnmenezes.gm@gmail.com"
            ]
            "referencia" => array:3 [
              0 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">a</span>"
                "identificador" => "aff0005"
              ]
              1 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">b</span>"
                "identificador" => "aff0010"
              ]
              2 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">&#42;</span>"
                "identificador" => "cor0005"
              ]
            ]
          ]
          1 => array:3 [
            "nombre" => "Jo&#227;o"
            "apellidos" => "Louren&#231;o-Silva"
            "referencia" => array:1 [
              0 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">c</span>"
                "identificador" => "aff0015"
              ]
            ]
          ]
          2 => array:3 [
            "nombre" => "Beatriz"
            "apellidos" => "Silva"
            "referencia" => array:2 [
              0 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">a</span>"
                "identificador" => "aff0005"
              ]
              1 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">b</span>"
                "identificador" => "aff0010"
              ]
            ]
          ]
          3 => array:3 [
            "nombre" => "Tiago"
            "apellidos" => "Rodrigues"
            "referencia" => array:2 [
              0 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">a</span>"
                "identificador" => "aff0005"
              ]
              1 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">b</span>"
                "identificador" => "aff0010"
              ]
            ]
          ]
          4 => array:3 [
            "nombre" => "Ana Rita G&#46;"
            "apellidos" => "Francisco"
            "referencia" => array:2 [
              0 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">a</span>"
                "identificador" => "aff0005"
              ]
              1 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">b</span>"
                "identificador" => "aff0010"
              ]
            ]
          ]
          5 => array:3 [
            "nombre" => "Pedro"
            "apellidos" => "Carrilho Ferreira"
            "referencia" => array:2 [
              0 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">a</span>"
                "identificador" => "aff0005"
              ]
              1 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">b</span>"
                "identificador" => "aff0010"
              ]
            ]
          ]
          6 => array:3 [
            "nombre" => "Arlindo L&#46;"
            "apellidos" => "Oliveira"
            "referencia" => array:1 [
              0 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">c</span>"
                "identificador" => "aff0015"
              ]
            ]
          ]
          7 => array:3 [
            "nombre" => "Fausto J&#46;"
            "apellidos" => "Pinto"
            "referencia" => array:2 [
              0 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">a</span>"
                "identificador" => "aff0005"
              ]
              1 => array:2 [
                "etiqueta" => "<span class="elsevierStyleSup">b</span>"
                "identificador" => "aff0010"
              ]
            ]
          ]
        ]
        "afiliaciones" => array:3 [
          0 => array:3 [
            "entidad" => "Structural and Coronary Heart Disease Unit&#44; Cardiovascular Center of the University of Lisbon&#44; Faculdade de Medicina&#44; Universidade de Lisboa&#44; Lisboa&#44; Portugal"
            "etiqueta" => "a"
            "identificador" => "aff0005"
          ]
          1 => array:3 [
            "entidad" => "Servi&#231;o de Cardiologia&#44; Departamento de Cora&#231;&#227;o e Vasos&#44; CHULN Hospital de Santa Maria&#44; Lisboa&#44; Portugal"
            "etiqueta" => "b"
            "identificador" => "aff0010"
          ]
          2 => array:3 [
            "entidad" => "INESC-ID&#47;Instituto Superior T&#233;cnico&#44; University of Lisbon&#44; Portugal"
            "etiqueta" => "c"
            "identificador" => "aff0015"
          ]
        ]
        "correspondencia" => array:1 [
          0 => array:3 [
            "identificador" => "cor0005"
            "etiqueta" => "&#8270;"
            "correspondencia" => "Corresponding author&#46;"
          ]
        ]
      ]
    ]
    "titulosAlternativos" => array:1 [
      "pt" => array:1 [
        "titulo" => "Desenvolvimento de modelos de <span class="elsevierStyleItalic">deep learning</span> para segmenta&#231;&#227;o de coronariografias&#58; aferi&#231;&#227;o de qualidade por um novo modelo de segmenta&#231;&#227;o global e compara&#231;&#227;o com desempenho humano"
      ]
    ]
    "resumenGrafico" => array:2 [
      "original" => 0
      "multimedia" => array:7 [
        "identificador" => "fig0090"
        "etiqueta" => "Figure 5"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "gr5.jpeg"
            "Alto" => 1305
            "Ancho" => 1305
            "Tamanyo" => 103564
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0130" class="elsevierStyleSimplePara elsevierViewall">Crossovers in spider &#40;above&#41; and extreme RAO cranial &#40;below&#41; views generating artifacts&#46;</p>"
        ]
      ]
    ]
    "textoCompleto" => "<span class="elsevierStyleSections"><span id="sec0005" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0065">Introduction</span><p id="par0005" class="elsevierStylePara elsevierViewall">Artificial intelligence &#40;AI&#41; has shown great potential in medicine&#44; in applications such as predictive data analysis&#44;<a class="elsevierStyleCrossRef" href="#bib0125"><span class="elsevierStyleSup">1</span></a> decision making support<a class="elsevierStyleCrossRef" href="#bib0130"><span class="elsevierStyleSup">2</span></a> or even medical education&#47;awareness improvement&#44;<a class="elsevierStyleCrossRef" href="#bib0135"><span class="elsevierStyleSup">3</span></a> and especially in image analysis&#46; Several publications have demonstrated impressive results with regards to electrocardiogram&#44;<a class="elsevierStyleCrossRef" href="#bib0140"><span class="elsevierStyleSup">4</span></a> echocardiography&#44;<a class="elsevierStyleCrossRefs" href="#bib0145"><span class="elsevierStyleSup">5&#44;6</span></a> or magnetic resonance imaging&#46;<a class="elsevierStyleCrossRefs" href="#bib0155"><span class="elsevierStyleSup">7&#44;8</span></a></p><p id="par0010" class="elsevierStylePara elsevierViewall">The use of AI in Interventional Cardiology &#40;IC&#41; is&#44; however&#44; still a vastly underexplored field&#46; Its application to coronary angiography &#40;CAG&#41; has been explored in very few medical or biology publication&#46;<a class="elsevierStyleCrossRefs" href="#bib0165"><span class="elsevierStyleSup">9&#8211;12</span></a> There are&#44; nonetheless&#44; many possibilities&#44;<a class="elsevierStyleCrossRef" href="#bib0185"><span class="elsevierStyleSup">13</span></a> ranging from automatic anatomical identification&#44; stenosis analysis&#44; lesion subset characterization and perhaps even physiological index derivation&#46; Regardless of the task&#44; arguably the first step in applying AI to CAG is separating and identifying relevant information &#8211; the coronary tree &#8211; from non-relevant information &#40;bones&#44; other structures&#41;&#46; This task is called segmentation&#46;<a class="elsevierStyleCrossRef" href="#bib0190"><span class="elsevierStyleSup">14</span></a></p><p id="par0015" class="elsevierStylePara elsevierViewall">In this paper&#44; we explore the development of AI models for automatic coronary artery segmentation from CAG&#44; and assess the results from a clinical perspective&#44; using a new set of criteria and score clinically defined by a panel of Interventional Cardiologists&#46;</p></span><span id="sec0010" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0070">Methods</span><span id="sec0015" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0075">Dataset selection</span><span id="sec0020" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0080">Inclusion criteria</span><p id="par0020" class="elsevierStylePara elsevierViewall">We retrospectively and randomly included patients who had undergone CAG and invasive physiology assessment &#40;fractional flow reserve and&#47;or other indexes&#41; during the procedure at a single center &#40;tertiary university hospital&#41;&#46;</p><p id="par0025" class="elsevierStylePara elsevierViewall">These patients have at least intermediate lesions in one or more vessels&#46; Around one third usually undergo revascularization due to the severity of their disease&#46;<a class="elsevierStyleCrossRefs" href="#bib0195"><span class="elsevierStyleSup">15&#44;16</span></a> Therefore&#44; a dataset focusing on these patients comprises a wide spectrum of obstructive coronary artery disease in a relatively balanced way&#46;</p></span><span id="sec0025" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0085">Exclusion criteria</span><p id="par0030" class="elsevierStylePara elsevierViewall">We excluded cases where any of the following applied&#58;<ul class="elsevierStyleList" id="lis0005"><li class="elsevierStyleListItem" id="lsti0005"><span class="elsevierStyleLabel">1&#41;</span><p id="par0035" class="elsevierStylePara elsevierViewall">Major occluded vessels &#40;acute or chronic&#41;</p></li><li class="elsevierStyleListItem" id="lsti0010"><span class="elsevierStyleLabel">2&#41;</span><p id="par0040" class="elsevierStylePara elsevierViewall">Poor image quality</p></li><li class="elsevierStyleListItem" id="lsti0015"><span class="elsevierStyleLabel">3&#41;</span><p id="par0045" class="elsevierStylePara elsevierViewall">Less than two orthogonal views in the left coronary artery &#40;LCA&#41; - one caudal and one cranial - or absence of at least one left oblique &#40;LAO&#41; view - either cranial or simple - in the right coronary artery &#40;RCA&#41;</p></li><li class="elsevierStyleListItem" id="lsti0020"><span class="elsevierStyleLabel">4&#41;</span><p id="par0050" class="elsevierStylePara elsevierViewall">Patients with previous cardiac surgery&#44; cardiac devices or other sources of potential artifact&#46;</p></li></ul></p></span><span id="sec0030" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0090">Image selection</span><p id="par0055" class="elsevierStylePara elsevierViewall">A single best frame was selected for each diagnostic angulation incidence in each patient&#46;</p></span><span id="sec0035" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0095">Dataset size</span><p id="par0060" class="elsevierStylePara elsevierViewall">The dataset size was the result of a trade-off between two opposing criteria&#58; dimension large enough for successful training of a deep convolutional neural network&#44; estimated from published data<a class="elsevierStyleCrossRefs" href="#bib0165"><span class="elsevierStyleSup">9&#44;12&#44;17&#44;18</span></a> vs&#46; expected time required to complete the annotation&#46; We estimated the latter based on a short period of annotation testing prior to formal dataset creation&#46; The trade-off pointed to a training set size of roughly 400&#46;</p><p id="par0065" class="elsevierStylePara elsevierViewall">We then randomly and consecutively selected patients until a total of at least 400 annotated images were obtained&#46;</p></span></span><span id="sec0040" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0100">Baseline annotation process</span><p id="par0070" class="elsevierStylePara elsevierViewall">Baseline human dataset images were annotated by two senior Cardiology Fellows &#40;TR&#47;BS&#41; previously trained in CAG interpretation&#44; under the supervision of an Interventional Cardiologist &#40;MNM&#41;&#44; who also annotated&#46; Images were periodically reviewed and perfected by all three&#46; This meant that any initial heterogeneity between annotators was corrected by consensus&#46; The small size of the team was aimed at reducing heterogeneity&#44; as we noticed during the preparatory phase that some operators tended to annotate too much &#40;Supplementary figure 1&#44; Appendix A&#41;&#44; while others did the opposite &#40;Supplementary figure 2&#44; Appendix A&#41;&#46;</p><p id="par0075" class="elsevierStylePara elsevierViewall">Both the catheter &#40;labeled red&#41; and the coronaries &#40;labeled white&#41; were to be segmented&#46;</p><p id="par0080" class="elsevierStylePara elsevierViewall">The coronary tree was to be fully segmented up to branches of approximately 2 mm in caliper at their origin &#40;as the vessel became smaller&#44; it was to be segmented until discernible&#41;&#44; using the catheter as reference &#40;without formal measurements &#8211; eyeball appreciation was used&#41;&#46; There were several reasons for this&#58; &#40;1&#41; when performing percutaneous coronary intervention&#44; vessels &#60;2 mm are usually approached conservatively&#44; as the risk of target lesion failure increases significantly<a class="elsevierStyleCrossRefs" href="#bib0215"><span class="elsevierStyleSup">19&#44;20</span></a>&#59; &#40;2&#41; human annotation is cumbersome &#8211; segmenting every single vessel would increase the risk of errors significantly&#59; &#40;3&#41; including very small vessels might increase the chances of artifacts from bone or other structures when training and applying AI models&#46;</p></span><span id="sec0045" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0105">Baseline artificial intelligence model training</span><p id="par0085" class="elsevierStylePara elsevierViewall">We performed segmentation using an encoder-decoder fully convolutional neural network based on the U-Net&#44;<a class="elsevierStyleCrossRef" href="#bib0225"><span class="elsevierStyleSup">21</span></a> commonly used in medical image segmentation&#46; As the name suggests&#44; these neural networks are composed of an encoder&#44; responsible for extracting image features&#44; and a decoder&#44; which processes those features to produce segmentation masks&#46; To derive the best approach for this task&#44; we conducted a comparative study of encoder and decoder architectures&#44; which resulted in the proposal of the EfficientUNet&#43;&#43;&#44; a computationally efficient and high-performing decoder architecture<a class="elsevierStyleCrossRef" href="#bib0230"><span class="elsevierStyleSup">22</span></a> that&#44; in this work&#44; we combine with an EfficientNet-B5 encoder<a class="elsevierStyleCrossRef" href="#bib0235"><span class="elsevierStyleSup">23</span></a> &#40;<a class="elsevierStyleCrossRef" href="#fig0070">Figure 1</a>&#41;&#46;</p><elsevierMultimedia ident="fig0070"></elsevierMultimedia><p id="par0090" class="elsevierStylePara elsevierViewall">To ensure fair evaluation&#44; it was necessary to guarantee that each model was tested on data that it had not seen during training&#46; Therefore&#44; the dataset was split&#44; at the patient level&#44; into 13 subsets of approximately 32 angiograms each&#46; Each subset segmentation was performed using a neural network trained exclusively on the remaining data&#46; This enabled the assessment of the segmentation results for the entire cohort&#44; as the usual splitting into a training and testing dataset would have yielded a much smaller group of images for result assessment&#46;</p><p id="par0095" class="elsevierStylePara elsevierViewall">The training hyperparameters&#44; including the number of training epochs and the learning rate decay schedule&#44; were set on the first train-test split&#44; using one of the 12 training data subsets for validation&#46; The selected values were then used on every other train-test split&#44; and to train the model on the whole training set of the first split&#46;</p></span><span id="sec0050" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0110">Enhanced human model</span><p id="par0100" class="elsevierStylePara elsevierViewall">The results of the baseline AI training were reviewed by the annotating team&#44; without any formal grading&#44; which would be performed subsequently &#40;see below&#41;&#46; For each image&#44; both human and AI segmentation were compared with the original&#46; Each annotation was then perfected using a mixture of the best of baseline human segmentation and baseline AI&#44; with additional de novo manual segmentation as needed&#46;</p></span><span id="sec0055" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0115">Enhanced artificial intelligence model</span><p id="par0105" class="elsevierStylePara elsevierViewall">The neural network architecture and training procedure were identical for both the baseline and enhanced AI model &#40;<a class="elsevierStyleCrossRef" href="#fig0070">Figure 1</a>&#41;&#46; The sole difference was the dataset&#46; The baseline AI model was trained using the baseline human annotations&#44; whereas the enhanced AI model was trained using enhanced human annotations&#46;</p><p id="par0110" class="elsevierStylePara elsevierViewall"><a class="elsevierStyleCrossRef" href="#fig0075">Figure 2</a> outlines the development stages&#46;</p><elsevierMultimedia ident="fig0075"></elsevierMultimedia></span><span id="sec0060" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0120">Performance assessment</span><span id="sec0065" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0125">Non-medical metrics</span><p id="par0115" class="elsevierStylePara elsevierViewall">AI models were assessed using the Dice Similarity Coefficient &#40;DSC&#41; and Generalized Dice Score &#40;GDS&#41;&#44; measures of the overlap between segmentations&#46; Given two segmentations&#44; the DSC has a value between 0&#58; no overlap and 1&#58; total overlap&#44; corresponding to the ratio between the area of their intersection and the sum of their areas&#46; GDS<a class="elsevierStyleCrossRef" href="#bib0240"><span class="elsevierStyleSup">24</span></a> is a weighted sum of each class&#39;s DSC that attributes the same importance to all classes&#44; regardless of their frequency&#46; While DSC and GDS alone do not reflect clinical usefulness&#44; they are helpful and entirely objective metrics that enable a simple comparison between models&#46;</p></span><span id="sec0070" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0130">Clinical performance criteria</span><p id="par0120" class="elsevierStylePara elsevierViewall">The DSC objectively assesses model performance&#46; However&#44; it does not provide a medically meaningful impression of whether segmentation is appropriate&#46; Also&#44; because the DCS can only be calculated based on previously annotated images&#44; it cannot be applied to new&#44; unannotated datasets in the future&#46; To overcome these limitations&#44; we created a set of criteria to assess performance as interpreted by expert physicians&#46;</p><p id="par0125" class="elsevierStylePara elsevierViewall">The following 11 criteria are as objectively defined as possible and were analyzed for each image&#46; Each was independently met or not&#46; A &#8220;perfect&#8221; example is shown in <a class="elsevierStyleCrossRef" href="#fig0080">Figure 3</a>&#46; Supplementary Figures 3 to 13 &#40;Appendix A&#41; show error examples for each&#46;<ul class="elsevierStyleList" id="lis0010"><li class="elsevierStyleListItem" id="lsti0025"><span class="elsevierStyleLabel">1&#41;</span><p id="par0130" class="elsevierStylePara elsevierViewall">Catheter segmentation&#58;</p></li><li class="elsevierStyleListItem" id="lsti0030"><span class="elsevierStyleLabel">a&#46;</span><p id="par0135" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Main segmentation</span>&#58; The distal part of the catheter &#40;i&#46;e&#46; the closest discernible portion to the coronary artery in the ascending aorta&#41; is correctly segmented and labeled &#40;supplementary figure 3&#44; Appendix A&#41;&#46; If minor gaps are present&#44; this criterion should be scored as met&#46;</p></li><li class="elsevierStyleListItem" id="lsti0035"><span class="elsevierStyleLabel">b&#46;</span><p id="par0140" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Gaps</span> &#40;minor&#41; are absent &#40;supplementary figure 4&#44; Appendix A&#41;&#46;</p></li><li class="elsevierStyleListItem" id="lsti0040"><span class="elsevierStyleLabel">c&#46;</span><p id="par0145" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Catheter thickness</span> is accurate&#44; by visual appreciation &#40;supplementary figure 5&#44; Appendix A&#41;&#46;</p></li><li class="elsevierStyleListItem" id="lsti0045"><span class="elsevierStyleLabel">d&#46;</span><p id="par0150" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Location</span>&#58; if parts of the catheter far from the coronary ostia &#40;ascending and&#47;or descending aorta&#41; are segmented&#44; there are no major gaps or artifacts &#40;supplementary 6&#44; Appendix A&#41;&#46;</p></li><li class="elsevierStyleListItem" id="lsti0050"><span class="elsevierStyleLabel">2&#41;</span><p id="par0155" class="elsevierStylePara elsevierViewall">Vessel segmentation&#58;</p></li><li class="elsevierStyleListItem" id="lsti0055"><span class="elsevierStyleLabel">a&#46;</span><p id="par0160" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Main vessels</span> are correctly segmented and labeled&#46; For the RCA&#44; this includes the segments from the ostium to the crux &#40;supplementary figure 7&#44; Appendix A&#41;&#46; For the LCA&#44; this includes the segments from the left main ostium to the visually discernible distal segments of the left anterior descending or the circumflex &#40;or most important obtuse marginal branch&#41;&#44; depending on incidence&#46; Branches are excluded from this criterion&#46; If minor gaps are present&#44; this criterion should be scored as met&#46;</p></li><li class="elsevierStyleListItem" id="lsti0060"><span class="elsevierStyleLabel">b&#46;</span><p id="par0165" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Branch segmentation</span>&#58; branches with a luminal diameter of at least approximately 2 mm &#40;using the catheter size as reference&#41; are correctly segmented and labeled &#40;supplementary figure 8&#44; Appendix A&#41;&#46; Size is estimated by visual appreciation&#46; If minor gaps are present&#44; this criterion should be scored as met&#46;</p></li><li class="elsevierStyleListItem" id="lsti0065"><span class="elsevierStyleLabel">c&#46;</span><p id="par0170" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Main vessel gaps</span> &#40;minor&#41; are absent &#40;supplementary figure 9&#44; Appendix A&#41;&#46;</p></li><li class="elsevierStyleListItem" id="lsti0070"><span class="elsevierStyleLabel">d&#46;</span><p id="par0175" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Branch gaps</span> &#40;minor&#41; are absent &#40;supplementary figure 10&#44; Appendix A&#41;&#46;</p></li><li class="elsevierStyleListItem" id="lsti0075"><span class="elsevierStyleLabel">e&#46;</span><p id="par0180" class="elsevierStylePara elsevierViewall">Catheter to artery <span class="elsevierStyleBold">transition</span>&#58; correct labeling of the catheter tip vs&#46; coronary artery origin &#40;supplementary figure 11&#44; Appendix A&#41;&#46;</p></li><li class="elsevierStyleListItem" id="lsti0080"><span class="elsevierStyleLabel">3&#41;</span><p id="par0185" class="elsevierStylePara elsevierViewall">Artifacts</p></li><li class="elsevierStyleListItem" id="lsti0085"><span class="elsevierStyleLabel">a&#46;</span><p id="par0190" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Coronary</span>&#58; No non-coronary structures are labeled as part of the coronary tree &#40;supplementary figure 12&#44; Appendix A&#41;&#46;</p></li><li class="elsevierStyleListItem" id="lsti0090"><span class="elsevierStyleLabel">b&#46;</span><p id="par0195" class="elsevierStylePara elsevierViewall"><span class="elsevierStyleBold">Catheter</span>&#58; No non-catheter structures are incorrectly labeled as part of the catheter &#40;supplementary figure 13&#44; Appendix A&#41;&#46;</p></li></ul></p><elsevierMultimedia ident="fig0080"></elsevierMultimedia><p id="par0200" class="elsevierStylePara elsevierViewall">The criteria for these two artifacts are not applicable to the small catheter-artery transition area&#46;</p><p id="par0205" class="elsevierStylePara elsevierViewall">To provide an objective assessment&#44; these criteria were scored by a panel of three Interventional Cardiologists &#40;MNM&#44; ARF&#44; PCF&#41;&#44; of whom two &#40;ARF&#44; PCF&#41; took no part in any stage of the annotation&#47;training process&#46; Discrepancies were solved by agreement&#46; All images were graded across all groups&#58; baseline human segmentation&#44; enhanced human segmentation&#44; baseline AI and enhanced AI&#46; During the grading process&#44; the image group was blinded&#46;</p><p id="par0210" class="elsevierStylePara elsevierViewall">Lastly&#44; because the abovementioned criteria are not equally important&#44; a Global Segmentation Score &#40;GSS&#44; 1&#46;5 to 100 points&#41; was devised&#44; taking into account the relevance of each criterion as defined by the three experts &#40;<a class="elsevierStyleCrossRef" href="#tbl0005">Table 1</a>&#41;&#46; The panel was also asked to select which of the two AI models was preferred for each image&#44; regardless of the final score&#46;</p><elsevierMultimedia ident="tbl0005"></elsevierMultimedia></span></span><span id="sec0075" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0135">Statistical analysis</span><p id="par0215" class="elsevierStylePara elsevierViewall">Descriptive variables are shown in absolute and relative &#40;percentage&#41; numbers&#46; To assess the association between qualitative &#40;categorical&#41; variables the Chi-Square test was used&#46; To assess differences in quantitative variables we used the Mann-Whitney test &#40;two independent groups&#41; or the Kruskal-Wallis test &#40;multiple independent groups&#41;&#46; A p&#60;0&#46;05 was used for statistical significance&#44; except for multiple group comparisons&#44; where we used a p&#60;0&#46;01&#46; IBM SPSS Statistics 27 was used for statistical analysis&#46;</p></span><span id="sec0080" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0140">Ethical issues</span><p id="par0220" class="elsevierStylePara elsevierViewall">This study complies with the Declaration of Helsinki and was approved by the local ethics committee&#46;</p></span></span><span id="sec0085" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0145">Results</span><span id="sec0090" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0150">Baseline dataset</span><p id="par0225" class="elsevierStylePara elsevierViewall">We included 416 images from 69 patients &#40;<a class="elsevierStyleCrossRef" href="#tbl0010">Table 2</a>&#41;&#46; With two human and two AI datasets&#44; 1664 processed images were generated&#46;</p><elsevierMultimedia ident="tbl0010"></elsevierMultimedia></span><span id="sec0095" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0155">Performance assessment</span><p id="par0230" class="elsevierStylePara elsevierViewall">Non-medical metrics</p><p id="par0235" class="elsevierStylePara elsevierViewall">Results are outlined in <a class="elsevierStyleCrossRef" href="#tbl0015">Table 3</a>&#46; These scores indicate that enhanced AI was generally superior to baseline AI&#46; Segmentation performance was good and consistent across arteries&#44; as indicated by the high mean and low standard deviation of the DSC&#46; For the catheter&#44; performance was lower and much less consistent&#46;</p><elsevierMultimedia ident="tbl0015"></elsevierMultimedia></span><span id="sec0100" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0160">Clinical performance</span><p id="par0240" class="elsevierStylePara elsevierViewall">Overall performance &#8211; individual criteria assessment &#40;Supplementary Table 1&#44; Appendix A&#41;&#46;</p></span><span id="sec0105" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0165">Coronary segmentation</span><p id="par0245" class="elsevierStylePara elsevierViewall">The main vessels were correctly segmented in almost all cases across groups&#46; Minor gaps occurred rarely in the baseline human segmentation and both AI models&#44; although there was a small but non-significant improvement with the enhanced AI vs&#46; baseline AI&#46;</p><p id="par0250" class="elsevierStylePara elsevierViewall">Branch segmentation was also correct almost always in all groups&#44; albeit less so than main vessel segmentation&#46; There was a small&#44; yet significant&#44; improvement with the enhanced AI vs&#46; baseline AI&#46;</p><p id="par0255" class="elsevierStylePara elsevierViewall">Minor branch gaps were quite common&#44; revealing very significant differences between AI and human models&#46; While enhanced AI performed numerically better than baseline AI&#44; it still produced small gaps in nearly two thirds of cases&#46;</p><p id="par0260" class="elsevierStylePara elsevierViewall">Coronary artifacts were very uncommon in human annotations and were usually minor imperfections in catheter&#47;coronary crossovers&#46; They were common and usually minor in both AI models&#44; although there was a very significant improvement with the enhanced AI vs&#46; baseline AI &#40;14&#46;4&#37; vs&#46; 25&#46;7&#37;&#41;&#46;</p></span><span id="sec0110" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0170">Catheter&#47;artery transition</span><p id="par0265" class="elsevierStylePara elsevierViewall">Baseline human segmentation failed in 12&#37; of cases and enhanced human segmentation missed 3&#46;8&#37;&#46; Baseline AI produced a higher error rate &#40;19&#46;7&#37;&#41;&#44; but enhanced AI was numerically more often correct than baseline human segmentation&#44; sometimes correctly identifying the transition where humans failed &#40;<a class="elsevierStyleCrossRef" href="#fig0085">Figure 4</a>&#41;&#46;</p><elsevierMultimedia ident="fig0085"></elsevierMultimedia></span><span id="sec0115" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0175">Catheter segmentation</span><p id="par0270" class="elsevierStylePara elsevierViewall">Baseline human segmentation produced thickness imperfections &#40;usually mildly engorged catheter&#41; in 13&#46;9&#37; of cases&#44; but otherwise&#44; segmentation was almost always correct regarding other criteria&#46; Baseline AI produced low error rates in main body segmentation&#46; However&#44; artifacts&#44; usually quite minor and in the vicinity of coronary segments&#44; occurred very frequently &#40;41&#46;1&#37;&#41;&#46; Another common error was catheter thickness &#40;36&#46;3&#37;&#41;&#44; often resulting in an overestimation of catheter size&#46;</p><p id="par0275" class="elsevierStylePara elsevierViewall">Enhanced human segmentation significantly improved on thickness issues&#44; although imperfections persisted in 6&#46;2&#37; of cases&#46;</p><p id="par0280" class="elsevierStylePara elsevierViewall">Enhanced AI produced better results than the baseline AI model for catheter thickness &#40;correct in 96&#46;4&#37;&#41;&#44; also surpassing both human models &#40;although the difference was not statistically significant when compared to the enhanced human segmentation&#41;&#46; However&#44; the performance of the enhanced AI otherwise decreased in all other criteria&#44; especially regarding minor gaps&#44; which became much more common &#40;3&#46;1&#37; in the baseline AI model to 23&#46;3&#37;&#41;&#46; Even main body segmentation was significantly affected&#44; although successful in the vast majority of cases &#40;86&#46;5&#37;&#41;&#46; Despite this&#44; in most failures catheter identification was still possible&#44; as major gaps often occurred distally in areas of contrast backflow&#46; There was a slight numerical worsening in artifact and location issues in enhanced AI vs&#46; baseline AI&#46;</p></span><span id="sec0120" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0180">Overall performance &#8211; Global Segmentation Score assessment and expert preference &#40;<a class="elsevierStyleCrossRef" href="#tbl0020">Table 4</a>&#41;</span><p id="par0285" class="elsevierStylePara elsevierViewall">Human models outperformed AI models&#46; Enhanced models surpassed baseline models&#46; The difference was statistically significant for all comparisons&#46; GSS was very high for both AI models&#59; the enhanced AI reached an average of 90 points&#46;</p><elsevierMultimedia ident="tbl0020"></elsevierMultimedia><p id="par0290" class="elsevierStylePara elsevierViewall">With regards to expert preference&#44; the enhanced AI model was preferred in 300 &#40;72&#37;&#41; cases&#44; the baseline AI model in 100 &#40;24&#37;&#41; and in 16 &#40;4&#37;&#41; cases no AI model was preferred&#46;</p></span><span id="sec0125" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0185">Performance according to coronary artery &#8211; individual criteria assessment &#40;Supplementary Table 2&#44; Appendix A&#41;</span><p id="par0295" class="elsevierStylePara elsevierViewall">There was a trend toward better performance in the RCA&#44; both regarding human and AI groups&#46; The most notable and statistically significant differences occurred in catheter transition &#40;regarding both AI models and the baseline human segmentation&#41; and catheter segmentation &#40;both AI models performed better in the RCA&#41;&#46; Branch gaps were quite less frequent in the RCA with the enhanced AI model&#46; Other differences&#44; even if statistically significant&#44; were very small&#46;</p></span><span id="sec0130" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0190">Performance by coronary artery &#8211; Global Segmentation Score assessment &#40;Supplementary Table 3&#44; Appendix A&#41;</span><p id="par0300" class="elsevierStylePara elsevierViewall">All models scored very high for both arteries&#46; There were very minor statistically significant differences for the baseline AI model only&#44; favoring RCA segmentation&#46;</p><p id="par0305" class="elsevierStylePara elsevierViewall">Considering expert preference&#58;<ul class="elsevierStyleList" id="lis0015"><li class="elsevierStyleListItem" id="lsti0095"><span class="elsevierStyleLabel">-</span><p id="par0310" class="elsevierStylePara elsevierViewall">RCA&#58; Enhanced AI was preferred in 109 &#40;68&#46;6&#37;&#41; cases&#44; the baseline AI was preferred in 43 &#40;27&#37;&#41; and in 7 &#40;4&#46;4&#37;&#41; cases no AI model was preferred&#46;</p></li><li class="elsevierStyleListItem" id="lsti0100"><span class="elsevierStyleLabel">-</span><p id="par0315" class="elsevierStylePara elsevierViewall">LCA&#58; Enhanced AI was preferred in 191 &#40;74&#46;3&#37;&#41; cases&#44; the baseline AI was preferred in 57 &#40;22&#46;2&#37;&#41; and in 9 &#40;3&#46;5&#37;&#41; cases no AI was preferred&#46;</p></li></ul></p></span><span id="sec0135" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0195">Performance according to angulation incidence &#8211; individual criteria assessment &#40;Supplementary Tables 4 and 5&#44; Appendix A&#41;</span><p id="par0320" class="elsevierStylePara elsevierViewall">Given the large amount of data&#44; there being no significant differences in the vast majority of cases and for the sake of readability&#44; only statistically significant differences are shown in the tables&#46; Overall&#44; the impact of incidences on model performance was limited&#44; and affected almost exclusively the AI models&#46;</p></span><span id="sec0140" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0200">Performance according to angulation incidence &#8211; Global Segmentation Score assessment &#40;Supplementary Tables 6 and 7&#44; Appendix A&#41;</span><p id="par0325" class="elsevierStylePara elsevierViewall">Differences were minor and only statistically significant for human performance in less common incidences &#40;PA views for the LCA and PA cranial for the RCA&#41;&#46;</p></span></span><span id="sec0145" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0205">Discussion</span><span id="sec0150" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0210">Overall considerations</span><p id="par0330" class="elsevierStylePara elsevierViewall">Baseline human segmentation was generally correct&#46; Catheter&#47;coronary transition and catheter thickness errors were the most common&#46; Poor individualization due to contrast backflow&#44; catheter curves and human fatigue all likely contributed&#46;</p><p id="par0335" class="elsevierStylePara elsevierViewall">Enhanced human segmentation was nearly perfect&#46; Mild transition issues remained&#44; highlighting the difficulty of the task&#46; As this model was actually a combination of the best of baseline human segmentation and baseline AI&#44; it also demonstrates how AI can help improve human performance&#46; Even these slight human imperfections highlight the need for rigorous quality control during and after the final results&#44; rather than assuming human annotation is a &#8220;perfect&#8221; ground truth&#46; This an inherent limitation to the annotation of medical images&#44; as the sheer amount of cumbersome work is error prone&#46;</p><p id="par0340" class="elsevierStylePara elsevierViewall">Baseline AI performed CAG segmentation successfully yet was affected by the same two issues of the baseline human segmentation &#8211; transition and catheter thickness&#46; The effort to correct these when developing the enhanced AI was fruitful in the case of transition but produced mixed results for catheter thickness&#46; Impact on transition performance was impressive&#44; as&#44; at times&#44; the enhanced AI even achieved correct assessments where humans failed &#40;<a class="elsevierStyleCrossRef" href="#fig0085">Figure 4</a>&#41;&#46; However&#44; it seems the gain in catheter thickness accuracy was offset by losses in other catheter segmentation tasks&#46; Lastly&#44; every aspect of coronary segmentation improved in the enhanced AI&#44; which performed better than baseline AI&#46; The differences between the two AI models also highlight how relatively small differences in the ground truth can impact relevantly on AI training&#46;</p><p id="par0345" class="elsevierStylePara elsevierViewall">It may seem surprising that catheter segmentation was less successful than coronary segmentation&#46; However&#44; while intuitively one may think that catheter segmentation is an easier task and therefore the results would have been better for this task&#44; from a machine learning perspective that is not the case&#46; In particular&#44; segmentation performance is highly dependent on the frequency of each class&#46; Rarer classes&#44; or ones that occupy smaller areas&#44; are interpreted by the model as being less likely to appear&#46; Furthermore&#44; during training&#44; the lower the number of pixels belonging to a particular class&#44; the lower the penalty for segmenting that class incorrectly&#46; Even though we used a loss function designed to mitigate this phenomenon&#44; the poorer segmentation of less common classes &#40;the catheter&#44; in this case&#41; is still evident in the results&#46;</p><p id="par0350" class="elsevierStylePara elsevierViewall">Right coronary artery segmentation was easier than LCA&#44; however the differences were quite small and there were fewer than expected&#44; considering its greater anatomical simplicity&#46; Angulations also had a relatively small impact both on human and AI performance and small observed differences may be attributed to specific issues that are more common in certain incidences&#58; contrast backflow &#40;less problematic in PA or RAO caudal&#41;&#59; coronary&#47;catheter crossovers &#40;such as spider or extreme RAO cranial &#8211; <a class="elsevierStyleCrossRef" href="#fig0090">Figure 5</a>&#41;&#59; proximity of bone &#40;such as RCA LAO views&#41;&#59; smaller samples of some incidences&#44; such as PA cranial&#59; uncommon catheter pathways&#44; such as the femoral approach&#44; which sometimes produces a central vertical outline&#46;</p><elsevierMultimedia ident="fig0090"></elsevierMultimedia><p id="par0355" class="elsevierStylePara elsevierViewall">Globally&#44; both AI models achieved a very high DSC&#44; with higher performance in artery segmentation than in catheter segmentation&#44; supporting the results of qualitative clinical assessment&#46; When factors are weighed up based on their perceived relevance &#8211; as assessed by GSS &#8211; both performed very well&#46; The enhanced AI scored an average of 90 points&#44; meaning it provided 90&#37; of what experts deemed most relevant when viewing a CAG&#46; By all measures&#44; the enhanced AI was the better model&#46; However&#44; the fact that differences between the two AI models were not large and that the enhanced AI was preferred in most&#44; but not all cases&#44; highlights the difficulty in improving an already good performance&#46;</p></span><span id="sec0155" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0215">Other studies with artificial intelligence applied to coronary angiography segmentation&#47;interpretation</span><p id="par0360" class="elsevierStylePara elsevierViewall">Few studies regarding coronary artery segmentation based on AI technologies have been published in medical&#47;biology journals to date&#46; Yang et al&#46;<a class="elsevierStyleCrossRef" href="#bib0180"><span class="elsevierStyleSup">12</span></a> successfully developed AI models capable of segmenting CAG&#46; Their dataset was larger &#40;3302 images&#47;2042 patients&#41; and was also annotated by two expert physicians&#46; Different incidences were also used&#46; They also focused exclusively on segmenting specific segments of major vessels with at least mild &#40;&#62;30&#37;&#41; stenotic lesions&#46; Neither the branches nor the catheter were segmented&#44; leading to a much simpler problem than the one addressed in this article&#46;</p><p id="par0365" class="elsevierStylePara elsevierViewall">Two other works<a class="elsevierStyleCrossRefs" href="#bib0165"><span class="elsevierStyleSup">9&#44;10</span></a> from the same baseline dataset&#44; also developed AI-based CAG segmentation&#46; Their dataset was also larger &#40;4904 images from 170 videos&#41;&#46; However&#44; the annotations were performed by medical students and no details are provided regarding patient subset&#44; target vessel or incidence&#46;</p><p id="par0370" class="elsevierStylePara elsevierViewall">Very recently&#44; Du et al&#46;<a class="elsevierStyleCrossRef" href="#bib0175"><span class="elsevierStyleSup">11</span></a> published the results of a broad study&#46; They focused on two tasks&#58; CAG segmentation and special lesion morphology identification &#40;calcium&#44; thrombus&#44; among others&#41;&#46; For the former task&#44; which overlaps with ours&#44; they used a very large dataset of 13&#160;373 images distributed across ten incidences &#40;six LCA and four RCA&#41;&#44; annotated by ten qualified analysts&#46; This was an all-comers study&#44; rather than focusing on patient subsets&#46; They too annotated catheter&#47;arteries and additionally marked different coronary segments&#46; Their model is impressive as judged by the presented images&#44; as they even distinguished between contrast backflow&#44; catheter and coronary&#46; However&#44; they did not specify the exact criteria for segmenting the coronary tree and their exact metrics make it difficult to assess exactly how their models performed in detail regarding segmentation&#46;</p><p id="par0375" class="elsevierStylePara elsevierViewall">While all the abovementioned groups have worked with datasets larger than ours&#44; our study has several unique features&#58; &#40;1&#41; there was medical rationale for vessel size segmentation&#59; &#40;2&#41; results were assessed from a set of criteria defined by experts&#44; capturing the quality of the segmentation from an Interventional Cardiologist&#39;s eyes&#59; &#40;3&#41; human annotations were also graded&#44; rather than assuming a perfect human ground truth&#59; &#40;4&#41; specific segmentation tasks were appraised individually&#44; enabling insights into strengths and weaknesses of AI and human models alike&#59; &#40;5&#41; results were also considered globally with the GSS&#44; by factoring the relevance of each criterion&#44; enabling a broad&#44; simple appreciation of the results&#46; Furthermore&#44; the ability to perform high-quality segmentation in a system trained using less data provides relevant evidence that more advanced AI systems can be effectively applied even in situations where the available data are limited&#46;</p></span><span id="sec0160" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0220">Limitations</span><p id="par0380" class="elsevierStylePara elsevierViewall">This is a single center retrospective dataset&#44; involving a single image per projection and a smaller sample size than some previously published manuscripts&#46; The images come from the same angiography devices &#40;Siemens Artis&#41; and thus we have not yet tested our models on images obtained from other equipment or image settings&#46;</p><p id="par0385" class="elsevierStylePara elsevierViewall">We have not yet conducted formal assessment on how well the models perform in segmenting specific degrees of stenosis severity&#46; Our models are also yet to be tested for specific vessel disease types &#40;calcium&#44; thrombus&#41;&#44; clinical settings &#40;chronic total occlusion&#44; ST-elevation myocardial infarction&#41;&#46;</p><p id="par0390" class="elsevierStylePara elsevierViewall">We have not yet assessed the performance of AI models on an external validation cohort&#46; There are several reasons for this&#46; We aimed to compare AI and human results in detail first and assess the exact performance of AI models for each segmentation task&#46; A validation dataset would comprise a new set of images&#44; which would not undergo human segmentation&#44; thus impeding comparison with human performance&#46; Also&#44; validation implies that a metric be available for comparing results&#46; Because the Dice methods require a ground truth human annotation for comparison&#44; and the GSS was developed and applied for the first time for this paper&#44; we felt a suitable metric was not yet available for performing validation prior to the current analysis&#46; In addition&#44; AI models are continuously and dynamically improving&#46; As we are currently working on further testing and enhancing current AI models &#40;view Future direction and implications section below&#41;&#44; we felt performing external validation at this stage was premature&#46;</p><p id="par0395" class="elsevierStylePara elsevierViewall">The exclusion of cardiac devices&#47;cardiac surgery and other foreign objects renders our models not yet applicable to such cases&#46; We did not&#44; however&#44; exclude cases with previously implanted stents&#46;</p><p id="par0400" class="elsevierStylePara elsevierViewall">Lastly&#44; focusing specifically on patients undergoing invasive physiology assessment may have created bias&#44; limiting a broader application of the models to other patient subsets&#46;</p><p id="par0405" class="elsevierStylePara elsevierViewall">We are currently working to address all these issues in future research&#46;</p></span><span id="sec0165" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0225">Future direction and implications</span><p id="par0410" class="elsevierStylePara elsevierViewall">Coronary angiography segmentation in itself is not end objective but rather an essential milestone for developing AI systems capable of CAG analysis and interpretation&#46; These results should&#44; therefore&#44; be regarded as a first step&#44; rather than a final deployment tool&#46; While not yet ready for immediate clinical application&#44; the results of both AI models are already relevant&#44; providing a framework that can be built upon in the future&#46;</p><p id="par0415" class="elsevierStylePara elsevierViewall">Further steps include testing the models for stenosed segments&#44; which will be critical for clinical application&#46; In the future&#44; we aim to test our models with a validation cohort using new angiograms&#46; Sub-segmentation&#44; automatic anatomical identification and physiology are also areas for future research&#46;</p><p id="par0420" class="elsevierStylePara elsevierViewall">We will also strengthen the capabilities of our models further by broadening our training base to other patient and lesion subsets&#44; focusing on particular issues where there is still room for improvement&#44; as identified by our uniquely detailed analysis&#46;</p><p id="par0425" class="elsevierStylePara elsevierViewall">Our results also provide insight into which human tasks are most challenging&#44; which may be of use to other researchers&#46;</p><p id="par0430" class="elsevierStylePara elsevierViewall">Global Segmentation Score is the first of its kind for assessing the quality of segmentations in CAG&#46; By providing a reasonably objective and quantitative clinical measurement&#44; it can be used as a benchmark for comparing and validating results across research groups&#46;</p><p id="par0435" class="elsevierStylePara elsevierViewall">Lastly&#44; while conventional segmentation software does exist&#44; it is not without limitations&#44; and only by developing AI systems can we compare and improve both in the future&#46; The potential implications of AI for Interventional Cardiology are immense&#44; and we envisage a catherization lab of the future where all of these insights render the human eye more objective&#44; thus improving patient care&#46;</p></span></span><span id="sec0170" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0230">Conclusions</span><p id="par0440" class="elsevierStylePara elsevierViewall">We successfully developed two AI models capable of good quality automatic CAG segmentation&#44; as assessed by GDS&#44; DSC and the GSS&#46; From an expert&#39;s perspective&#44; the latter and its individual criteria provided a feasible&#44; reasonably objective and quantifiable way of assessing the results&#46;</p><p id="par0445" class="elsevierStylePara elsevierViewall">The enhanced AI model outperformed the baseline AI model in coronary segmentation tasks as well as globally&#46; With regards to catheter segmentation tasks&#44; the enhanced AI model improved on the task of catheter thickness&#44; but performed less well in other catheter segmentation tasks&#46; Both human segmentations were superior to both AI models&#44; but only the enhanced human segmentation&#44; built by combining the best of baseline human segmentation and baseline AI&#44; achieved a near perfect GSS&#46;</p><p id="par0450" class="elsevierStylePara elsevierViewall">These results provide a relevant framework for building upon&#44; potentially leading to future clinical application&#46;</p></span><span id="sec0175" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0235">Conflicts of interest</span><p id="par0455" class="elsevierStylePara elsevierViewall">The authors have no conflicts of interest to declare&#46;</p></span></span>"
    "textoCompletoSecciones" => array:1 [
      "secciones" => array:12 [
        0 => array:3 [
          "identificador" => "xres1823844"
          "titulo" => "Abstract"
          "secciones" => array:4 [
            0 => array:2 [
              "identificador" => "abst0005"
              "titulo" => "Introduction and objectives"
            ]
            1 => array:2 [
              "identificador" => "abst0010"
              "titulo" => "Methods"
            ]
            2 => array:2 [
              "identificador" => "abst0015"
              "titulo" => "Results"
            ]
            3 => array:2 [
              "identificador" => "abst0020"
              "titulo" => "Conclusions"
            ]
          ]
        ]
        1 => array:2 [
          "identificador" => "xpalclavsec1590897"
          "titulo" => "Keywords"
        ]
        2 => array:3 [
          "identificador" => "xres1823843"
          "titulo" => "Resumo"
          "secciones" => array:4 [
            0 => array:2 [
              "identificador" => "abst0025"
              "titulo" => "Introdu&#231;&#227;o e objetivos"
            ]
            1 => array:2 [
              "identificador" => "abst0030"
              "titulo" => "M&#233;todos"
            ]
            2 => array:2 [
              "identificador" => "abst0035"
              "titulo" => "Resultados"
            ]
            3 => array:2 [
              "identificador" => "abst0040"
              "titulo" => "Conclus&#245;es"
            ]
          ]
        ]
        3 => array:2 [
          "identificador" => "xpalclavsec1590898"
          "titulo" => "Palavras-chave"
        ]
        4 => array:2 [
          "identificador" => "sec0005"
          "titulo" => "Introduction"
        ]
        5 => array:3 [
          "identificador" => "sec0010"
          "titulo" => "Methods"
          "secciones" => array:8 [
            0 => array:3 [
              "identificador" => "sec0015"
              "titulo" => "Dataset selection"
              "secciones" => array:4 [
                0 => array:2 [
                  "identificador" => "sec0020"
                  "titulo" => "Inclusion criteria"
                ]
                1 => array:2 [
                  "identificador" => "sec0025"
                  "titulo" => "Exclusion criteria"
                ]
                2 => array:2 [
                  "identificador" => "sec0030"
                  "titulo" => "Image selection"
                ]
                3 => array:2 [
                  "identificador" => "sec0035"
                  "titulo" => "Dataset size"
                ]
              ]
            ]
            1 => array:2 [
              "identificador" => "sec0040"
              "titulo" => "Baseline annotation process"
            ]
            2 => array:2 [
              "identificador" => "sec0045"
              "titulo" => "Baseline artificial intelligence model training"
            ]
            3 => array:2 [
              "identificador" => "sec0050"
              "titulo" => "Enhanced human model"
            ]
            4 => array:2 [
              "identificador" => "sec0055"
              "titulo" => "Enhanced artificial intelligence model"
            ]
            5 => array:3 [
              "identificador" => "sec0060"
              "titulo" => "Performance assessment"
              "secciones" => array:2 [
                0 => array:2 [
                  "identificador" => "sec0065"
                  "titulo" => "Non-medical metrics"
                ]
                1 => array:2 [
                  "identificador" => "sec0070"
                  "titulo" => "Clinical performance criteria"
                ]
              ]
            ]
            6 => array:2 [
              "identificador" => "sec0075"
              "titulo" => "Statistical analysis"
            ]
            7 => array:2 [
              "identificador" => "sec0080"
              "titulo" => "Ethical issues"
            ]
          ]
        ]
        6 => array:3 [
          "identificador" => "sec0085"
          "titulo" => "Results"
          "secciones" => array:11 [
            0 => array:2 [
              "identificador" => "sec0090"
              "titulo" => "Baseline dataset"
            ]
            1 => array:2 [
              "identificador" => "sec0095"
              "titulo" => "Performance assessment"
            ]
            2 => array:2 [
              "identificador" => "sec0100"
              "titulo" => "Clinical performance"
            ]
            3 => array:2 [
              "identificador" => "sec0105"
              "titulo" => "Coronary segmentation"
            ]
            4 => array:2 [
              "identificador" => "sec0110"
              "titulo" => "Catheter&#47;artery transition"
            ]
            5 => array:2 [
              "identificador" => "sec0115"
              "titulo" => "Catheter segmentation"
            ]
            6 => array:2 [
              "identificador" => "sec0120"
              "titulo" => "Overall performance &#8211; Global Segmentation Score assessment and expert preference &#40;Table 4&#41;"
            ]
            7 => array:2 [
              "identificador" => "sec0125"
              "titulo" => "Performance according to coronary artery &#8211; individual criteria assessment &#40;Supplementary Table 2&#44; Appendix A&#41;"
            ]
            8 => array:2 [
              "identificador" => "sec0130"
              "titulo" => "Performance by coronary artery &#8211; Global Segmentation Score assessment &#40;Supplementary Table 3&#44; Appendix A&#41;"
            ]
            9 => array:2 [
              "identificador" => "sec0135"
              "titulo" => "Performance according to angulation incidence &#8211; individual criteria assessment &#40;Supplementary Tables 4 and 5&#44; Appendix A&#41;"
            ]
            10 => array:2 [
              "identificador" => "sec0140"
              "titulo" => "Performance according to angulation incidence &#8211; Global Segmentation Score assessment &#40;Supplementary Tables 6 and 7&#44; Appendix A&#41;"
            ]
          ]
        ]
        7 => array:3 [
          "identificador" => "sec0145"
          "titulo" => "Discussion"
          "secciones" => array:4 [
            0 => array:2 [
              "identificador" => "sec0150"
              "titulo" => "Overall considerations"
            ]
            1 => array:2 [
              "identificador" => "sec0155"
              "titulo" => "Other studies with artificial intelligence applied to coronary angiography segmentation&#47;interpretation"
            ]
            2 => array:2 [
              "identificador" => "sec0160"
              "titulo" => "Limitations"
            ]
            3 => array:2 [
              "identificador" => "sec0165"
              "titulo" => "Future direction and implications"
            ]
          ]
        ]
        8 => array:2 [
          "identificador" => "sec0170"
          "titulo" => "Conclusions"
        ]
        9 => array:2 [
          "identificador" => "sec0175"
          "titulo" => "Conflicts of interest"
        ]
        10 => array:2 [
          "identificador" => "xack643450"
          "titulo" => "Acknowledgments"
        ]
        11 => array:1 [
          "titulo" => "References"
        ]
      ]
    ]
    "pdfFichero" => "main.pdf"
    "tienePdf" => true
    "fechaRecibido" => "2022-01-27"
    "fechaAceptado" => "2022-04-03"
    "PalabrasClave" => array:2 [
      "en" => array:1 [
        0 => array:4 [
          "clase" => "keyword"
          "titulo" => "Keywords"
          "identificador" => "xpalclavsec1590897"
          "palabras" => array:6 [
            0 => "Deep learning"
            1 => "Artificial intelligence"
            2 => "Machine learning"
            3 => "Coronary angiography"
            4 => "Coronary artery disease"
            5 => "Percutaneous coronary intervention"
          ]
        ]
      ]
      "pt" => array:1 [
        0 => array:4 [
          "clase" => "keyword"
          "titulo" => "Palavras-chave"
          "identificador" => "xpalclavsec1590898"
          "palabras" => array:6 [
            0 => "Aprendizagem profunda"
            1 => "Intelig&#234;ncia artifical"
            2 => "Aprendizagem de m&#225;quinas"
            3 => "Coronariografia"
            4 => "Doen&#231;a coron&#225;ria"
            5 => "Interven&#231;&#227;o coron&#225;ria percut&#226;nea"
          ]
        ]
      ]
    ]
    "tieneResumen" => true
    "resumen" => array:2 [
      "en" => array:3 [
        "titulo" => "Abstract"
        "resumen" => "<span id="abst0005" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0010">Introduction and objectives</span><p id="spar0005" class="elsevierStyleSimplePara elsevierViewall">Although automatic artificial intelligence &#40;AI&#41; coronary angiography &#40;CAG&#41; segmentation is arguably the first step toward future clinical application&#44; it is underexplored&#46; We aimed to &#40;1&#41; develop AI models for CAG segmentation and &#40;2&#41; assess the results using similarity scores and a set of criteria defined by expert physicians&#46;</p></span> <span id="abst0010" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0015">Methods</span><p id="spar0010" class="elsevierStyleSimplePara elsevierViewall">Patients undergoing CAG were randomly selected in a retrospective study at a single center&#46; Per incidence&#44; an ideal frame was segmented&#44; forming a baseline human dataset &#40;BH&#41;&#44; used for training a baseline AI model &#40;BAI&#41;&#46; Enhanced human segmentation &#40;EH&#41; was created by combining the best of both&#46; An enhanced AI model &#40;EAI&#41; was trained using the EH&#46; Results were assessed by experts using 11 weighted criteria&#44; combined into a Global Segmentation Score &#40;GSS&#58; 0&#8211;100 points&#41;&#46; Generalized Dice Score &#40;GDS&#41; and Dice Similarity Coefficient &#40;DSC&#41; were also used for AI models assessment&#46;</p></span> <span id="abst0015" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0020">Results</span><p id="spar0015" class="elsevierStyleSimplePara elsevierViewall">1664 processed images were generated&#46; GSS for BH&#44; EH&#44; BAI and EAI were 96&#46;9&#43;&#47;-5&#46;7&#59; 98&#46;9&#43;&#47;-3&#46;1&#59; 86&#46;1&#43;&#47;-10&#46;1 and 90&#43;&#47;-7&#46;6&#44; respectively &#40;95&#37; confidence interval&#44; p&#60;0&#46;001 for both paired and global differences&#41;&#46; The GDS for the BAI and EAI was 0&#46;9234&#177;0&#46;0361 and 0&#46;9348&#177;0&#46;0284&#44; respectively&#46; The DSC for the coronary tree was 0&#46;8904&#177;0&#46;0464 and 0&#46;9134&#177;0&#46;0410 for the BAI and EAI&#44; respectively&#46; The EAI outperformed the BAI in all coronary segmentation tasks&#44; but performed less well in some catheter segmentation tasks&#46;</p></span> <span id="abst0020" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0025">Conclusions</span><p id="spar0020" class="elsevierStyleSimplePara elsevierViewall">We successfully developed AI models capable of CAG segmentation&#44; with good performance as assessed by all scores&#46;</p></span>"
        "secciones" => array:4 [
          0 => array:2 [
            "identificador" => "abst0005"
            "titulo" => "Introduction and objectives"
          ]
          1 => array:2 [
            "identificador" => "abst0010"
            "titulo" => "Methods"
          ]
          2 => array:2 [
            "identificador" => "abst0015"
            "titulo" => "Results"
          ]
          3 => array:2 [
            "identificador" => "abst0020"
            "titulo" => "Conclusions"
          ]
        ]
      ]
      "pt" => array:3 [
        "titulo" => "Resumo"
        "resumen" => "<span id="abst0025" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0035">Introdu&#231;&#227;o e objetivos</span><p id="spar0025" class="elsevierStyleSimplePara elsevierViewall">A segmenta&#231;&#227;o autom&#225;tica de coronariografia &#40;CRG&#41; por intelig&#234;ncia artificial &#40;IA&#41; encontra-se pouco explorada na literatura m&#233;dica&#46; Os objetivos do presente estudo s&#227;o &#40;1&#41; desenvolver modelos de IA para segmenta&#231;&#227;o de CRG e &#40;2&#41; aferir os resultados por <span class="elsevierStyleItalic">scores</span> de similaridade e crit&#233;rios definidos por peritos&#46;</p></span> <span id="abst0030" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0040">M&#233;todos</span><p id="spar0030" class="elsevierStyleSimplePara elsevierViewall">Doentes submetidos a CRG foram retrospetivamente selecionados aleatoriamente num centro&#46; Por incid&#234;ncia&#44; segmentou-se um <span class="elsevierStyleItalic">frame</span> ideal&#44; formando uma segmenta&#231;&#227;o humana basal &#40;HB&#41;&#44; usada para treinar um modelo de IA basal &#40;IAB&#41;&#46; Da combina&#231;&#227;o de ambos acrescentou-se uma segmenta&#231;&#227;o humana aperfei&#231;oada &#40;HA&#41;&#44; utilizada para treinar um modelo de IA aperfei&#231;oado &#40;IAA&#41;&#46; Os resultados foram aferidos com 11 crit&#233;rios balanceados definidos por peritos&#44; combinados num <span class="elsevierStyleItalic">Score</span><span class="elsevierStyleItalic">de Segmenta&#231;&#227;o Global</span> &#40;SSC &#8211; 0&#8211;100 pontos&#41;&#46; O <span class="elsevierStyleItalic">Score</span><span class="elsevierStyleItalic">de Dice Generalizado &#40;</span>SDG&#41; e <span class="elsevierStyleItalic">Score de Dice de Similaridade</span> &#40;SDS&#41; aplicaram-se aos modelos de IA&#46;</p></span> <span id="abst0035" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0045">Resultados</span><p id="spar0035" class="elsevierStyleSimplePara elsevierViewall">Geraram-se 1664 imagens processadas&#46; Os SCC para a HB&#44; HA&#44; IAB e IAA foram 96&#44;9&#43;&#47;-5&#44;7&#59; 98&#44;9&#43;&#47;-3&#44;1&#59; 86&#44;1&#43;&#47;-10&#44;1 e 90&#43;&#47;-7&#44;6&#44; respetivamente &#40;IC 95&#37;&#44; p&#60;0&#44;001 - diferen&#231;as globais e emparelhadas&#41;&#46; O SDG para o IAB e IAA foi 0&#44;9234&#177;0&#44;0361 e 0&#44;9348&#177;0&#44;0284&#44; respetivamente&#46; O SDS foi 0&#44;8904&#177;0&#44;0464 e 0&#44;9134&#177;0&#44;0410 para o IAB e IAA&#44; respetivamente&#46; O IAA exibiu superior desempenho ao IAB para as todas tarefas de segmenta&#231;&#227;o coron&#225;ria&#44; mas n&#227;o para todas as de cateter&#46;</p></span> <span id="abst0040" class="elsevierStyleSection elsevierViewall"><span class="elsevierStyleSectionTitle" id="sect0050">Conclus&#245;es</span><p id="spar0040" class="elsevierStyleSimplePara elsevierViewall">Desenvolvemos modelos de IA de segmenta&#231;&#227;o autom&#225;tica de CRG&#44; com bom desempenho de acordo com aferi&#231;&#227;o por todos os <span class="elsevierStyleItalic">scores</span>&#46;</p></span>"
        "secciones" => array:4 [
          0 => array:2 [
            "identificador" => "abst0025"
            "titulo" => "Introdu&#231;&#227;o e objetivos"
          ]
          1 => array:2 [
            "identificador" => "abst0030"
            "titulo" => "M&#233;todos"
          ]
          2 => array:2 [
            "identificador" => "abst0035"
            "titulo" => "Resultados"
          ]
          3 => array:2 [
            "identificador" => "abst0040"
            "titulo" => "Conclus&#245;es"
          ]
        ]
      ]
    ]
    "apendice" => array:1 [
      0 => array:1 [
        "seccion" => array:1 [
          0 => array:4 [
            "apendice" => "<p id="par0475" class="elsevierStylePara elsevierViewall">The following are the supplementary material to this article&#58;</p> <p id="par0480" class="elsevierStylePara elsevierViewall"><elsevierMultimedia ident="fig0005"></elsevierMultimedia><elsevierMultimedia ident="fig0010"></elsevierMultimedia><elsevierMultimedia ident="fig0015"></elsevierMultimedia><elsevierMultimedia ident="fig0020"></elsevierMultimedia><elsevierMultimedia ident="fig0025"></elsevierMultimedia><elsevierMultimedia ident="fig0030"></elsevierMultimedia><elsevierMultimedia ident="fig0035"></elsevierMultimedia><elsevierMultimedia ident="fig0040"></elsevierMultimedia><elsevierMultimedia ident="fig0045"></elsevierMultimedia><elsevierMultimedia ident="fig0050"></elsevierMultimedia><elsevierMultimedia ident="fig0055"></elsevierMultimedia><elsevierMultimedia ident="fig0060"></elsevierMultimedia><elsevierMultimedia ident="fig0065"></elsevierMultimedia><elsevierMultimedia ident="upi0005"></elsevierMultimedia></p>"
            "etiqueta" => "Appendix A"
            "titulo" => "Supplementary material"
            "identificador" => "sec0185"
          ]
        ]
      ]
    ]
    "multimedia" => array:23 [
      0 => array:7 [
        "identificador" => "fig0070"
        "etiqueta" => "Figure 1"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "gr1.jpeg"
            "Alto" => 405
            "Ancho" => 2508
            "Tamanyo" => 63361
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0110" class="elsevierStyleSimplePara elsevierViewall">Segmentation model composed of an EfficientNet-B5 encoder and an EfficientUNet&#43;&#43; decoder&#46;</p>"
        ]
      ]
      1 => array:7 [
        "identificador" => "fig0075"
        "etiqueta" => "Figure 2"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "gr2.jpeg"
            "Alto" => 2146
            "Ancho" => 2508
            "Tamanyo" => 198619
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0115" class="elsevierStyleSimplePara elsevierViewall">Annotation and training process&#46;</p>"
        ]
      ]
      2 => array:7 [
        "identificador" => "fig0080"
        "etiqueta" => "Figure 3"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "gr3.jpeg"
            "Alto" => 626
            "Ancho" => 1305
            "Tamanyo" => 45030
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0120" class="elsevierStyleSimplePara elsevierViewall">A segmentation case fulfilling all 11 criteria&#46;</p>"
        ]
      ]
      3 => array:7 [
        "identificador" => "fig0085"
        "etiqueta" => "Figure 4"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "gr4.jpeg"
            "Alto" => 752
            "Ancho" => 1505
            "Tamanyo" => 66385
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0125" class="elsevierStyleSimplePara elsevierViewall">&#40;left to right&#41;&#58; The first human segmentation incorrectly labels contrast backflow as coronary&#46; The baseline AI model improves on the human segmentation but is still not perfect&#46; The enhanced human model segments the transition perfectly&#46; The enhanced AI model is hampered in catheter segmentation but identifies the transition correctly&#46;</p>"
        ]
      ]
      4 => array:7 [
        "identificador" => "fig0090"
        "etiqueta" => "Figure 5"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "gr5.jpeg"
            "Alto" => 1305
            "Ancho" => 1305
            "Tamanyo" => 103564
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0130" class="elsevierStyleSimplePara elsevierViewall">Crossovers in spider &#40;above&#41; and extreme RAO cranial &#40;below&#41; views generating artifacts&#46;</p>"
        ]
      ]
      5 => array:8 [
        "identificador" => "tbl0005"
        "etiqueta" => "Table 1"
        "tipo" => "MULTIMEDIATABLA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "detalles" => array:1 [
          0 => array:3 [
            "identificador" => "at1"
            "detalle" => "Table "
            "rol" => "short"
          ]
        ]
        "tabla" => array:1 [
          "tablatextoimagen" => array:1 [
            0 => array:1 [
              "tabla" => array:1 [
                0 => """
                  <table border="0" frame="\n
                  \t\t\t\t\tvoid\n
                  \t\t\t\t" class=""><thead title="thead"><tr title="table-row"><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Criteria&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Catheter vs&#46; CoronaryRelative Weight&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Individual CriteriaRelative Weight&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Points&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th></tr></thead><tbody title="tbody"><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Main vessel segmentation&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">70&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">40&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">28&#46;0&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Main vessel gaps&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">10&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">7&#46;0&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Catheter to artery transition&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">15&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">10&#46;5&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Branch segmentation&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">20&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">14&#46;0&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">BranchGaps&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">5&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">3&#46;5&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Coronary artifacts&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">10&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">7&#46;0&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Catheter segmentation&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">30&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">40&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">12&#46;0&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Catheter gaps&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">10&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">3&#46;0&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Catheter artifacts&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">15&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">4&#46;5&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Catheter location&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">5&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">1&#46;5&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Catheter thickness&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">30&#37;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">9&#46;0&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Total&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">100&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr></tbody></table>
                  """
              ]
            ]
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0135" class="elsevierStyleSimplePara elsevierViewall">scoring metrics for application of the Global Segmentation Score&#46;</p>"
        ]
      ]
      6 => array:8 [
        "identificador" => "tbl0010"
        "etiqueta" => "Table 2"
        "tipo" => "MULTIMEDIATABLA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "detalles" => array:1 [
          0 => array:3 [
            "identificador" => "at2"
            "detalle" => "Table "
            "rol" => "short"
          ]
        ]
        "tabla" => array:2 [
          "leyenda" => "<p id="spar0165" class="elsevierStyleSimplePara elsevierViewall">CAG&#58; coronary angiography&#46;</p>"
          "tablatextoimagen" => array:1 [
            0 => array:1 [
              "tabla" => array:1 [
                0 => """
                  <table border="0" frame="\n
                  \t\t\t\t\tvoid\n
                  \t\t\t\t" class=""><thead title="thead"><tr title="table-row"><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Factor&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">N&#43;&#47;-SD or N&#40;&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th></tr></thead><tbody title="tbody"><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Age&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">67&#43;&#47;-11&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Sex &#40;male&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">54 &#40;78&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Hypertension&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">56 &#40;81&#46;2&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Diabetes mellitus&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">27 &#40;39&#46;1&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Dyslipidemia&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">39 &#40;56&#46;5&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Smoker &#40;past or present&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">26 &#40;37&#46;7&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Chronic coronary syndromes&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">50 &#40;72&#46;5&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Acute coronary syndrome&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">19 &#40;27&#46;5&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Revascularization during&#47;after CAG&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">21 &#40;30&#46;4&#37;&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr></tbody></table>
                  """
              ]
            ]
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0140" class="elsevierStyleSimplePara elsevierViewall">Baseline clinical characteristics of patients from whom images were analyzed&#46;</p>"
        ]
      ]
      7 => array:8 [
        "identificador" => "tbl0015"
        "etiqueta" => "Table 3"
        "tipo" => "MULTIMEDIATABLA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "detalles" => array:1 [
          0 => array:3 [
            "identificador" => "at3"
            "detalle" => "Table "
            "rol" => "short"
          ]
        ]
        "tabla" => array:2 [
          "leyenda" => "<p id="spar0150" class="elsevierStyleSimplePara elsevierViewall">BAI&#58; baseline AI model&#59; GDS&#58; Generalized Dice Score&#59; EAI&#59; enhanced AI model&#46;</p>"
          "tablatextoimagen" => array:1 [
            0 => array:1 [
              "tabla" => array:1 [
                0 => """
                  <table border="0" frame="\n
                  \t\t\t\t\tvoid\n
                  \t\t\t\t" class=""><thead title="thead"><tr title="table-row"><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BAI&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EAI&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th></tr></thead><tbody title="tbody"><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">GDS&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">0&#46;9234&#177;0&#46;0361&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">0&#46;9348&#177;0&#46;0284&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Artery DSC&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">0&#46;8904&#177;0&#46;0464&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">0&#46;9134&#177;0&#46;0410&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Catheter DSC&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">0&#46;7526&#177;0&#46;1998&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">0&#46;7975&#177;0&#46;1836&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr></tbody></table>
                  """
              ]
            ]
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0145" class="elsevierStyleSimplePara elsevierViewall">Generalized Dice Score and class-wise Dice Similarity Coefficient obtained by the baseline and enhanced AI models&#46; Results presented as mean &#177; standard deviation&#46;</p>"
        ]
      ]
      8 => array:8 [
        "identificador" => "tbl0020"
        "etiqueta" => "Table 4"
        "tipo" => "MULTIMEDIATABLA"
        "mostrarFloat" => true
        "mostrarDisplay" => false
        "detalles" => array:1 [
          0 => array:3 [
            "identificador" => "at4"
            "detalle" => "Table "
            "rol" => "short"
          ]
        ]
        "tabla" => array:3 [
          "leyenda" => "<p id="spar0160" class="elsevierStyleSimplePara elsevierViewall">BAI&#58; baseline AI model&#59; BH&#58; baseline human model&#59; EAI&#58; enhanced AI model&#59; EH&#58; enhanced human model&#59; GSS&#58; Global Segmentation Score&#59; IQR&#58; interquartile range&#59; SD&#58; standard deviation&#46;</p>"
          "tablatextoimagen" => array:1 [
            0 => array:1 [
              "tabla" => array:1 [
                0 => """
                  <table border="0" frame="\n
                  \t\t\t\t\tvoid\n
                  \t\t\t\t" class=""><thead title="thead"><tr title="table-row"><th class="td-with-role" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col">GSS&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " colspan="4" align="center" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Group</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " colspan="7" align="center" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">p-value</th></tr><tr title="table-row"><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BH&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EH&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BAI&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EAI&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">Between all<a class="elsevierStyleCrossRef" href="#tblfn0005">&#42;</a>&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BH vs EH<a class="elsevierStyleCrossRef" href="#tblfn0010"><span class="elsevierStyleSup">&#42;&#42;</span></a>&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BAI vs EAI<a class="elsevierStyleCrossRef" href="#tblfn0010"><span class="elsevierStyleSup">&#42;&#42;</span></a>&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BH vs BAI<a class="elsevierStyleCrossRef" href="#tblfn0010"><span class="elsevierStyleSup">&#42;&#42;</span></a>&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EH vs EAI<a class="elsevierStyleCrossRef" href="#tblfn0010"><span class="elsevierStyleSup">&#42;&#42;</span></a>&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">BH vs EAI<a class="elsevierStyleCrossRef" href="#tblfn0010"><span class="elsevierStyleSup">&#42;&#42;</span></a>&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th><th class="td" title="\n
                  \t\t\t\t\ttable-head\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t" scope="col" style="border-bottom: 2px solid black">EH vs BAI<a class="elsevierStyleCrossRef" href="#tblfn0010"><span class="elsevierStyleSup">&#42;&#42;</span></a>&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t\t\t</th></tr></thead><tbody title="tbody"><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Mean&#43;&#47;-SD&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">96&#46;9&#43;&#47;-5&#46;7&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">98&#46;9&#43;&#47;-3&#46;1&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">86&#46;1&#43;&#47;-10&#46;1&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">90&#43;&#47;-7&#46;6&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&#60;0&#46;001&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&#60;0&#46;001&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&#60;0&#46;001&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&#60;0&#46;001&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&#60;0&#46;001&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&#60;0&#46;001&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&#60;0&#46;001&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr><tr title="table-row"><td class="td-with-role" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t ; entry_with_role_rowhead " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">Median &#40;IQR&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">100 &#40;9&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">100&#40;0&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">87&#46;5 &#40;9&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="left" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">92 &#40;9&#46;5&#41;&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td><td class="td" title="\n
                  \t\t\t\t\ttable-entry\n
                  \t\t\t\t  " align="" valign="\n
                  \t\t\t\t\ttop\n
                  \t\t\t\t">&nbsp;\t\t\t\t\t\t\n
                  \t\t\t\t</td></tr></tbody></table>
                  """
              ]
            ]
          ]
          "notaPie" => array:2 [
            0 => array:3 [
              "identificador" => "tblfn0005"
              "etiqueta" => "&#42;"
              "nota" => "<p class="elsevierStyleNotepara" id="npar0005">Kruskal-Wallis Test&#46;</p>"
            ]
            1 => array:3 [
              "identificador" => "tblfn0010"
              "etiqueta" => "&#42;&#42;"
              "nota" => "<p class="elsevierStyleNotepara" id="npar0010">Mann-Whitney Test&#46;</p>"
            ]
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0155" class="elsevierStyleSimplePara elsevierViewall">Performance by group according to Global Segmentation Score &#40;significance at p&#60;0&#46;05 for paired differences and p&#60;0&#46;01 for multiple comparisons&#41;&#46;</p>"
        ]
      ]
      9 => array:5 [
        "identificador" => "upi0005"
        "tipo" => "MULTIMEDIAECOMPONENTE"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "Ecomponente" => array:2 [
          "fichero" => "mmc14.doc"
          "ficheroTamanyo" => 43765
        ]
      ]
      10 => array:7 [
        "identificador" => "fig0005"
        "etiqueta" => "Supplementary Figure 1"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc1.jpeg"
            "Alto" => 648
            "Ancho" => 1300
            "Tamanyo" => 51205
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0045" class="elsevierStyleSimplePara elsevierViewall">A test case with too many annotated vessels &#40;not used for training&#41;&#46;</p>"
        ]
      ]
      11 => array:7 [
        "identificador" => "fig0010"
        "etiqueta" => "Supplementary Figure 2"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc2.jpeg"
            "Alto" => 650
            "Ancho" => 1300
            "Tamanyo" => 39593
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0050" class="elsevierStyleSimplePara elsevierViewall">A test case with too few annotations &#40;not used for training&#41;&#46;</p>"
        ]
      ]
      12 => array:7 [
        "identificador" => "fig0015"
        "etiqueta" => "Supplementary Figure 3"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc3.jpeg"
            "Alto" => 651
            "Ancho" => 1300
            "Tamanyo" => 51620
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0055" class="elsevierStyleSimplePara elsevierViewall">Contrast backflow leads the artificial intelligence model to disregard most of the catheter&#46;</p>"
        ]
      ]
      13 => array:7 [
        "identificador" => "fig0020"
        "etiqueta" => "Supplementary Figure 4"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc4.jpeg"
            "Alto" => 653
            "Ancho" => 1300
            "Tamanyo" => 46187
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0060" class="elsevierStyleSimplePara elsevierViewall">Minor catheter gaps&#44; possibly facilitated by contrast backflow&#46;</p>"
        ]
      ]
      14 => array:7 [
        "identificador" => "fig0025"
        "etiqueta" => "Supplementary Figure 5"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc5.jpeg"
            "Alto" => 649
            "Ancho" => 1300
            "Tamanyo" => 46546
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0065" class="elsevierStyleSimplePara elsevierViewall">Catheter thickness is overestimated&#46;</p>"
        ]
      ]
      15 => array:7 [
        "identificador" => "fig0030"
        "etiqueta" => "Supplementary Figure 6"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc6.jpeg"
            "Alto" => 650
            "Ancho" => 1300
            "Tamanyo" => 53256
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0070" class="elsevierStyleSimplePara elsevierViewall">A small part of the catheter in the descending aorta was segmented in a femoral access case&#46;</p>"
        ]
      ]
      16 => array:7 [
        "identificador" => "fig0035"
        "etiqueta" => "Supplementary Figure 7"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc7.jpeg"
            "Alto" => 650
            "Ancho" => 1300
            "Tamanyo" => 47821
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0075" class="elsevierStyleSimplePara elsevierViewall">A major part of the left main was not segmented&#46;</p>"
        ]
      ]
      17 => array:7 [
        "identificador" => "fig0040"
        "etiqueta" => "Supplementary Figure 8"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc8.jpeg"
            "Alto" => 650
            "Ancho" => 1300
            "Tamanyo" => 43312
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0080" class="elsevierStyleSimplePara elsevierViewall">The posterolateral branch and much of the posterior descending artery were missed&#46; The enhanced artificial intelligence model did not miss these vessels&#46;</p>"
        ]
      ]
      18 => array:7 [
        "identificador" => "fig0045"
        "etiqueta" => "Supplementary Figure 9"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc9.jpeg"
            "Alto" => 650
            "Ancho" => 1300
            "Tamanyo" => 48885
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0085" class="elsevierStyleSimplePara elsevierViewall">A small gap is visible in the left anterior descending artery&#46;</p>"
        ]
      ]
      19 => array:7 [
        "identificador" => "fig0050"
        "etiqueta" => "Supplementary Figure 10"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc10.jpeg"
            "Alto" => 650
            "Ancho" => 1300
            "Tamanyo" => 51326
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0090" class="elsevierStyleSimplePara elsevierViewall">Small gaps are visible in branches&#46;</p>"
        ]
      ]
      20 => array:7 [
        "identificador" => "fig0055"
        "etiqueta" => "Supplementary Figure 11"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc11.jpeg"
            "Alto" => 651
            "Ancho" => 1300
            "Tamanyo" => 46390
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0095" class="elsevierStyleSimplePara elsevierViewall">Contrast backflow renders the transition less discernible&#44; leading the model to miss the transition zone&#46;</p>"
        ]
      ]
      21 => array:7 [
        "identificador" => "fig0060"
        "etiqueta" => "Supplementary Figure 12"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc12.jpeg"
            "Alto" => 650
            "Ancho" => 1300
            "Tamanyo" => 44232
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0100" class="elsevierStyleSimplePara elsevierViewall">A part of the intervertebral disk and vertebra are mislabeled as coronary&#46;</p>"
        ]
      ]
      22 => array:7 [
        "identificador" => "fig0065"
        "etiqueta" => "Supplementary Figure 13"
        "tipo" => "MULTIMEDIAFIGURA"
        "mostrarFloat" => false
        "mostrarDisplay" => true
        "figura" => array:1 [
          0 => array:4 [
            "imagen" => "mmc13.jpeg"
            "Alto" => 650
            "Ancho" => 1300
            "Tamanyo" => 47287
          ]
        ]
        "descripcion" => array:1 [
          "en" => "<p id="spar0105" class="elsevierStyleSimplePara elsevierViewall">Contrast backflow is mislabeled as catheter&#46;</p>"
        ]
      ]
    ]
    "bibliografia" => array:2 [
      "titulo" => "References"
      "seccion" => array:1 [
        0 => array:2 [
          "identificador" => "bibs0015"
          "bibliografiaReferencia" => array:24 [
            0 => array:3 [
              "identificador" => "bib0125"
              "etiqueta" => "1"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Phenomapping for novel classification of heart failure with preserved ejection fraction"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "S&#46;J&#46; Shah"
                            1 => "D&#46;H&#46; Katz"
                            2 => "S&#46; Selvaraj"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1161/CIRCULATIONAHA.114.010637"
                      "Revista" => array:6 [
                        "tituloSerie" => "Circulation"
                        "fecha" => "2015"
                        "volumen" => "131"
                        "paginaInicial" => "269"
                        "paginaFinal" => "279"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/25398313"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            1 => array:3 [
              "identificador" => "bib0130"
              "etiqueta" => "2"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Early experience with IBM Watson for Oncology &#40;WFO&#41; cognitive computing system for lung and colorectal cancer treatment"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "S&#46;P&#46; Somashekhar"
                            1 => "M&#46;-J&#46; Sep&#250;lveda"
                            2 => "A&#46;D&#46; Norden"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "Revista" => array:5 [
                        "tituloSerie" => "J Clin Oncol"
                        "fecha" => "2017"
                        "volumen" => "35"
                        "numero" => "suppl&#46;"
                        "paginaInicial" => "8527"
                      ]
                    ]
                  ]
                ]
              ]
            ]
            2 => array:3 [
              "identificador" => "bib0135"
              "etiqueta" => "3"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Acute myocardial infarction on YouTube &#8211; is it all fake news&#63;"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "I&#46; Fialho"
                            1 => "M&#46; Beringuilho"
                            2 => "D&#46; Madeira"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "Revista" => array:5 [
                        "tituloSerie" => "Rev Port Cardiol &#40;English Ed&#41;"
                        "fecha" => "2021"
                        "volumen" => "40"
                        "paginaInicial" => "815"
                        "paginaFinal" => "825"
                      ]
                    ]
                  ]
                ]
              ]
            ]
            3 => array:3 [
              "identificador" => "bib0140"
              "etiqueta" => "4"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "A&#46;Y&#46; Hannun"
                            1 => "P&#46; Rajpurkar"
                            2 => "M&#46; Haghpanahi"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "Revista" => array:4 [
                        "tituloSerie" => "Nat Med"
                        "fecha" => "2019"
                        "paginaInicial" => "65"
                        "paginaFinal" => "69"
                      ]
                    ]
                  ]
                ]
              ]
            ]
            4 => array:3 [
              "identificador" => "bib0145"
              "etiqueta" => "5"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Machine-learning algorithms to automate morphological and functional assessments in 2D echocardiography"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "S&#46; Narula"
                            1 => "K&#46; Shameer"
                            2 => "A&#46;M&#46; Salem Omar"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1016/j.jacc.2016.08.062"
                      "Revista" => array:6 [
                        "tituloSerie" => "J Am Coll Cardiol"
                        "fecha" => "2016"
                        "volumen" => "68"
                        "paginaInicial" => "2287"
                        "paginaFinal" => "2295"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/27884247"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            5 => array:3 [
              "identificador" => "bib0150"
              "etiqueta" => "6"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Automated echocardiographic quantification of left ventricular ejection fraction without volume measurements using a machine learning algorithm mimicking a human expert"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "F&#46;M&#46; Asch"
                            1 => "N&#46; Poilvert"
                            2 => "T&#46; Abraham"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1161/CIRCIMAGING.119.009303"
                      "Revista" => array:5 [
                        "tituloSerie" => "Circ Cardiovasc Imaging"
                        "fecha" => "2019"
                        "volumen" => "12"
                        "paginaInicial" => "e009303"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/31522550"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            6 => array:3 [
              "identificador" => "bib0155"
              "etiqueta" => "7"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Combining deep learning and level set for the automated segmentation of the left ventricle of the heart from cardiac cine magnetic resonance"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => false
                          "autores" => array:3 [
                            0 => "T&#46;A&#46; Ngo"
                            1 => "Z&#46; Lu"
                            2 => "G&#46; Carneiro"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1016/j.media.2016.05.009"
                      "Revista" => array:6 [
                        "tituloSerie" => "Med Image Anal"
                        "fecha" => "2017"
                        "volumen" => "35"
                        "paginaInicial" => "159"
                        "paginaFinal" => "171"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/27423113"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            7 => array:3 [
              "identificador" => "bib0160"
              "etiqueta" => "8"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Automated cardiovascular magnetic resonance image analysis with fully convolutional networks 08 Information and Computing Sciences 0801 Artificial Intelligence and Image Processing"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "W&#46; Bai"
                            1 => "M&#46; Sinclair"
                            2 => "G&#46; Tarroni"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "Revista" => array:4 [
                        "tituloSerie" => "J Cardiovasc Magn Reson"
                        "fecha" => "2018"
                        "volumen" => "20"
                        "paginaInicial" => "65"
                      ]
                    ]
                  ]
                ]
              ]
            ]
            8 => array:3 [
              "identificador" => "bib0165"
              "etiqueta" => "9"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Coronary artery segmentation in angiographic videos utilizing spatial-temporal information"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "L&#46; Wang"
                            1 => "D&#46; Liang"
                            2 => "X&#46; Yin"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1186/s12880-019-0399-0"
                      "Revista" => array:6 [
                        "tituloSerie" => "BMC Med Imaging"
                        "fecha" => "2020"
                        "volumen" => "20"
                        "paginaInicial" => "1"
                        "paginaFinal" => "10"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/31892314"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            9 => array:3 [
              "identificador" => "bib0170"
              "etiqueta" => "10"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Coronary angiography video segmentation method for assisting cardiovascular disease interventional treatment"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "D&#46; Liang"
                            1 => "J&#46; Qiu"
                            2 => "L&#46; Wang"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1186/s12880-019-0399-0"
                      "Revista" => array:6 [
                        "tituloSerie" => "BMC Med Imaging"
                        "fecha" => "2020"
                        "volumen" => "20"
                        "paginaInicial" => "1"
                        "paginaFinal" => "8"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/31892314"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            10 => array:3 [
              "identificador" => "bib0175"
              "etiqueta" => "11"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Training and validation of a deep learning architecture for the automatic analysis of coronary angiography"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "T&#46; Du"
                            1 => "L&#46; Xie"
                            2 => "H&#46; Zhang"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.4244/EIJ-D-20-00570"
                      "Revista" => array:6 [
                        "tituloSerie" => "EuroIntervention"
                        "fecha" => "2021"
                        "volumen" => "17"
                        "paginaInicial" => "32"
                        "paginaFinal" => "40"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/32830647"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            11 => array:3 [
              "identificador" => "bib0180"
              "etiqueta" => "12"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Deep learning segmentation of major vessels in X-ray coronary angiography"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "S&#46; Yang"
                            1 => "J&#46; Kweon"
                            2 => "J&#46;H&#46; Roh"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1038/s41598-018-37186-2"
                      "Revista" => array:6 [
                        "tituloSerie" => "Sci Rep"
                        "fecha" => "2019"
                        "volumen" => "9"
                        "paginaInicial" => "1"
                        "paginaFinal" => "11"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/30626917"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            12 => array:3 [
              "identificador" => "bib0185"
              "etiqueta" => "13"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Implementing machine learning in interventional cardiology&#58; the benefits are worth the trouble"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "W&#46; Ben Ali"
                            1 => "A&#46; Pesaranghader"
                            2 => "R&#46; Avram"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "Revista" => array:3 [
                        "tituloSerie" => "Front Cardiovasc Med"
                        "fecha" => "2021"
                        "paginaInicial" => "1775"
                      ]
                    ]
                  ]
                ]
              ]
            ]
            13 => array:3 [
              "identificador" => "bib0190"
              "etiqueta" => "14"
              "referencia" => array:1 [
                0 => array:1 [
                  "referenciaCompleta" => "Gonzalez &#38; Woods&#44; Digital image processing&#44; 4th ed&#46; &#124; Pearson &#91;Internet&#93;&#46; Available from&#58; <a target="_blank" href="https://www.pearson.com/us/higher-education/program/Gonzalez-Digital-Image-Processing-4th-Edition/PGM241219.html">https&#58;&#47;&#47;www&#46;pearson&#46;com&#47;us&#47;higher-education&#47;program&#47;Gonzalez-Digital-Image-Processing-4th-Edition&#47;PGM241219&#46;html</a> &#91;cited 02&#46;09&#46;21&#93;"
                ]
              ]
            ]
            14 => array:3 [
              "identificador" => "bib0195"
              "etiqueta" => "15"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Use of the instantaneous wave-free ratio or fractional flow reserve in PCI"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "J&#46;E&#46; Davies"
                            1 => "S&#46; Sen"
                            2 => "H&#46;-M&#46; Dehbi"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1056/NEJMoa1700445"
                      "Revista" => array:6 [
                        "tituloSerie" => "N Engl J Med"
                        "fecha" => "2017"
                        "volumen" => "376"
                        "paginaInicial" => "1824"
                        "paginaFinal" => "1834"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/28317458"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            15 => array:3 [
              "identificador" => "bib0200"
              "etiqueta" => "16"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Instantaneous wave-free ratio versus fractional flow reserve to guide PCI"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "M&#46; G&#246;tberg"
                            1 => "E&#46;H&#46; Christiansen"
                            2 => "I&#46;J&#46; Gudmundsdottir"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1056/NEJMoa1616540"
                      "Revista" => array:6 [
                        "tituloSerie" => "N Engl J Med"
                        "fecha" => "2017"
                        "volumen" => "376"
                        "paginaInicial" => "1813"
                        "paginaFinal" => "1823"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/28317438"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            16 => array:3 [
              "identificador" => "bib0205"
              "etiqueta" => "17"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Coronary angiography image segmentation based on PSPNet"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "X&#46; Zhu"
                            1 => "Z&#46; Cheng"
                            2 => "S&#46; Wang"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1016/j.cmpb.2020.105897"
                      "Revista" => array:5 [
                        "tituloSerie" => "Comput Methods Programs Biomed"
                        "fecha" => "2021"
                        "volumen" => "200"
                        "paginaInicial" => "105897"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/33317873"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            17 => array:3 [
              "identificador" => "bib0210"
              "etiqueta" => "18"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "et&#58; Nested encoder&#8211;decoder architecture for the main vessel segmentation in coronary angiography"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "T&#46;J&#46; Jun"
                            1 => "J&#46; Kweon"
                            2 => "Y&#46;H&#46; Kim"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1016/j.neunet.2020.05.002"
                      "Revista" => array:6 [
                        "tituloSerie" => "Neural Networks"
                        "fecha" => "2020"
                        "volumen" => "128"
                        "paginaInicial" => "216"
                        "paginaFinal" => "233"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/32447265"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            18 => array:3 [
              "identificador" => "bib0215"
              "etiqueta" => "19"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Treatment of very small de novo coronary artery disease with 2&#46;0 mm drug-coated balloons showed 1-year clinical outcome comparable with 2&#46;0 mm drug-eluting stents &#91;Internet&#93;"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "H&#46;W&#46; Sim"
                            1 => "R&#46; Ananthakrishna"
                            2 => "S&#46;P&#46; Chan"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "Revista" => array:6 [
                        "tituloSerie" => "J Invasive Cardiol"
                        "fecha" => "2018"
                        "volumen" => "30"
                        "paginaInicial" => "256"
                        "paginaFinal" => "261"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/29656281"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            19 => array:3 [
              "identificador" => "bib0220"
              "etiqueta" => "20"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Small-vessel treatment with contemporary newer-generation drug-eluting coronary stents in all-comers&#58; Insights from 2-year DUTCH PEERS &#40;TWENTE II&#41; randomized trial"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "L&#46;C&#46;H&#46; van der"
                            1 => "K&#46; MM"
                            2 => "D&#46; PW"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:2 [
                      "doi" => "10.1016/j.ahj.2016.02.020"
                      "Revista" => array:6 [
                        "tituloSerie" => "Am Heart J"
                        "fecha" => "2016"
                        "volumen" => "176"
                        "paginaInicial" => "28"
                        "paginaFinal" => "35"
                        "link" => array:1 [
                          0 => array:2 [
                            "url" => "https://www.ncbi.nlm.nih.gov/pubmed/27264217"
                            "web" => "Medline"
                          ]
                        ]
                      ]
                    ]
                  ]
                ]
              ]
            ]
            20 => array:3 [
              "identificador" => "bib0225"
              "etiqueta" => "21"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "U-Net&#58; convolutional networks for biomedical image segmentation"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => false
                          "autores" => array:3 [
                            0 => "O&#46; Ronneberger"
                            1 => "P&#46; Fischer"
                            2 => "T&#46; Brox"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "Revista" => array:5 [
                        "tituloSerie" => "Lect Notes Comput Sci &#40;including Subser Lect Notes Artif Intell Lect Notes Bioinformatics&#41;"
                        "fecha" => "2015"
                        "volumen" => "9351"
                        "paginaInicial" => "234"
                        "paginaFinal" => "241"
                      ]
                    ]
                  ]
                ]
              ]
            ]
            21 => array:3 [
              "identificador" => "bib0230"
              "etiqueta" => "22"
              "referencia" => array:1 [
                0 => array:1 [
                  "referenciaCompleta" => "Silva JL&#44; Nobre Menezes M&#44; Rodrigues T&#44; et al&#46; Encoder-decoder architectures for clinically relevant coronary artery segmentation&#59; 2021 Jun 21&#44; arXiv&#58;210611447 &#91;eessIV&#93;&#46;"
                ]
              ]
            ]
            22 => array:3 [
              "identificador" => "bib0235"
              "etiqueta" => "23"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "EfficientNet&#58; rethinking model scaling for convolutional neural networks"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => false
                          "autores" => array:2 [
                            0 => "M&#46; Tan"
                            1 => "Q&#46;V&#46; Le"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "LibroEditado" => array:5 [
                        "titulo" => "36th Int Conf Mach Learn ICML 2019"
                        "paginaInicial" => "10691"
                        "paginaFinal" => "10700"
                        "conferencia" => "International Machine Learning Society &#40;IMLS&#41;&#59; 2019 May 28"
                        "serieFecha" => "2019 June"
                      ]
                    ]
                  ]
                ]
              ]
            ]
            23 => array:3 [
              "identificador" => "bib0240"
              "etiqueta" => "24"
              "referencia" => array:1 [
                0 => array:2 [
                  "contribucion" => array:1 [
                    0 => array:2 [
                      "titulo" => "Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations"
                      "autores" => array:1 [
                        0 => array:2 [
                          "etal" => true
                          "autores" => array:3 [
                            0 => "C&#46;H&#46; Sudre"
                            1 => "W&#46; Li"
                            2 => "T&#46; Vercauteren"
                          ]
                        ]
                      ]
                    ]
                  ]
                  "host" => array:1 [
                    0 => array:1 [
                      "LibroEditado" => array:3 [
                        "titulo" => "Deep Learn Med image Anal multimodal Learn Clin Decis Support Third Int Work DLMIA 2017&#44; 7th Int Work ML-CDS 2017&#44; held conjunction with MICCAI 2017 Quebec City&#44; QC&#44; vol&#46; 2017"
                        "paginaInicial" => "240"
                        "serieFecha" => "2017"
                      ]
                    ]
                  ]
                ]
              ]
            ]
          ]
        ]
      ]
    ]
    "agradecimientos" => array:1 [
      0 => array:4 [
        "identificador" => "xack643450"
        "titulo" => "Acknowledgments"
        "texto" => "<p id="par0460" class="elsevierStylePara elsevierViewall">MNM was responsible for conceptualization&#44; data gathering&#44; processing and analysis&#44; interpretating the results and paper drafting&#46; JLS was responsible for technical and AI tasks&#44; data and image processing&#44; model implementation and training&#46; TR and BS were responsible for data gathering&#44; processing and analysis&#46; ARF and PCF were responsible for data analysis and results interpretation&#46; ALO was responsible for supervising the work of JLS&#44; and participated directly in the same tasks&#46; FJP was responsible for supervising the work of MNM&#44; and participated directly in the same tasks&#46;</p><p id="par0465" class="elsevierStylePara elsevierViewall">All authors revised the paper critically for important intellectual content&#44; gave final approval for its publication and agreed to be accountable for all respects of its accuracy and integrity&#46;</p>"
        "vista" => "all"
      ]
    ]
  ]
  "idiomaDefecto" => "en"
  "url" => "/08702551/0000004100000012/v2_202212230532/S0870255122001354/v2_202212230532/en/main.assets"
  "Apartado" => array:4 [
    "identificador" => "93359"
    "tipo" => "SECCION"
    "pt" => array:2 [
      "titulo" => "Original articles"
      "idiomaDefecto" => true
    ]
    "idiomaDefecto" => "pt"
  ]
  "PDF" => "https://static.elsevier.es/multimedia/08702551/0000004100000012/v2_202212230532/S0870255122001354/v2_202212230532/en/main.pdf?idApp=UINPBA00004E&text.app=https://revportcardiol.org/"
  "EPUB" => "https://multimedia.elsevier.es/PublicationsMultimediaV1/item/epub/S0870255122001354?idApp=UINPBA00004E"
]

Informação do artigo

ISSN: 08702551
Idioma original: Inglês

DOI:10.1016/j.repc.2022.04.001

Dados atualizados diariamente

Ano/Mês	Html	Pdf	Total

2024 Outubro	38	43	81
2024 Setembro	47	44	91
2024 Agosto	66	30	96
2024 Julho	90	38	128
2024 Junho	52	29	81
2024 Maio	42	25	67
2024 Abril	56	48	104
2024 Mar�o	55	25	80
2024 Fevereiro	40	31	71
2024 Janeiro	39	48	87
2023 Dezembro	71	47	118
2023 Novembro	70	74	144
2023 Outubro	36	34	70
2023 Setembro	44	25	69
2023 Agosto	38	18	56
2023 Julho	47	34	81
2023 Junho	47	23	70
2023 Maio	42	25	67
2023 Abril	32	11	43
2023 Mar�o	67	35	102
2023 Fevereiro	46	34	80
2023 Janeiro	55	29	84
2022 Dezembro	151	92	243
2022 Novembro	75	29	104
2022 Outubro	60	34	94
2022 Setembro	78	73	151
2022 Agosto	46	65	111
2022 Julho	70	42	112
2022 Junho	41	36	77
2022 Maio	16	16	32

Mostrar tudo

Indexada em:

Siga-nos:

Estatísticas

Indexada em:

Siga-nos:

Estatísticas

Receba a nossa Newsletter