Coronary p-Graph: Automatic classification and localization of coronary artery stenosis from Cardiac CTA using DSA-based annotations.
Authors
Affiliations (6)
Affiliations (6)
- Laboratory of Image Science and Technology, School of Computer Science and Engineering, Southeast University, Nanjing, Jiangsu, China.
- Department of Radiology, Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing 210002, Jiangsu, China.
- Laboratory of Image Science and Technology, School of Computer Science and Engineering, Southeast University, Nanjing, Jiangsu, China; Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China; Jiangsu Provincial Joint International Research Laboratory of Medical Information Processing, Southeast University, Nanjing 210096, China; Centre de Recherche en Information Biomédicale Sino-français, 35042 Rennes, France.
- Centre de Recherche en Information Biomédicale Sino-français, 35042 Rennes, France; Laboratoire Traitement du Signal et de l'Image, Université de Rennes 1, 35000 Rennes, France; National Institute for Health and Medical Research, 35000 Rennes, France.
- Laboratory of Image Science and Technology, School of Computer Science and Engineering, Southeast University, Nanjing, Jiangsu, China; Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China; Jiangsu Provincial Joint International Research Laboratory of Medical Information Processing, Southeast University, Nanjing 210096, China; Centre de Recherche en Information Biomédicale Sino-français, 35042 Rennes, France. Electronic address: [email protected].
- Department of Radiology, Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing 210002, Jiangsu, China. Electronic address: [email protected].
Abstract
Coronary artery disease (CAD) is a prevalent cardiovascular condition with profound health implications. Digital subtraction angiography (DSA) remains the gold standard for diagnosing vascular disease, but its invasiveness and procedural demands underscore the need for alternative diagnostic approaches. Coronary computed tomography angiography (CCTA) has emerged as a promising non-invasive method for accurately classifying and localizing coronary artery stenosis. However, the complexity of CCTA images and their dependence on manual interpretation highlight the essential role of artificial intelligence in supporting clinicians in stenosis detection. This paper introduces a novel framework, Coronaryproposal-based Graph Convolutional Networks (Coronary p-Graph), designed for the automated detection of coronary stenosis from CCTA scans. The framework transforms CCTA data into curved multi-planar reformation (CMPR) images that delineate the coronary artery centerline. After aligning the CMPR volume along this centerline, the entire vasculature is analyzed using a convolutional neural network (CNN) for initial feature extraction. Based on predefined criteria informed by prior knowledge, the model generates candidate stenotic segments, termed "proposals," which serve as graph nodes. The spatial relationships between nodes are then modeled as edges, constructing a graph representation that is processed using a graph convolutional network (GCN) for precise classification and localization of stenotic segments. All CCTA images were rigorously annotated by three expert radiologists, using DSA reports as the reference standard. This novel methodology offers diagnostic performance equivalent to invasive DSA based solely on non-invasive CCTA, potentially reducing the need for invasive procedures. The proposed method was evaluated on a retrospective dataset comprising 259 cases, each with paired CCTA and corresponding DSA reports. Quantitative analyses demonstrated the superior performance of our approach compared to existing methods, with the following metrics: accuracy of 0.844, specificity of 0.910, area under the receiver operating characteristic curve (AUC) of 0.74, and mean absolute error (MAE) of 0.157.