AGMN: Association Graph-based Graph Matching Network for Coronary Artery Semantic Labeling on Invasive Coronary Angiograms

Semantic labeling of coronary arterial segments in invasive angiography (ICA) is important for automated assessment and report generation artery stenosis computer-aided disease (CAD) diagnosis. However, separating identifying individual challenging because morphological similarities different branches on the tree human-to-human variabilities exist. Inspired by training procedure interventional cardiologists interpreting structure arteries, we propose an association graph-based graph matching network (AGMN) semantic labeling. We first extract vascular from convert it into multiple graphs. Then, constructed two graphs where each vertex represents relationship between segments. Thus, segment task a classification task; ultimately, becomes equivalent to artery-to-artery correspondence More specifically, AGMN extracts features embedding module using graph, aggregates adjacent vertices edges convolution network, decodes generate mappings arteries. By learning mapping graphs, unlabeled are classified labeled achieve A dataset containing 263 ICAs was employed train validate proposed model, five-fold cross-validation scheme performed. Our model achieved average accuracy 0.8264, precision 0.8276, recall F1-score 0.8262, which significantly outperformed existing methods. In conclusion, have developed validated new algorithm with high accuracy, interpretability, robustness ICAs.