Enhancement and Classification of Coronary Heart Disease using X-ray Angiography Images

Abstract

Diagnosing Coronary Artery Disease (CAD) detection using angiographic image is challenging due to non-uniform image quality, inconsistent noise, complex vascular structure. This paper proposes an Integrated AI Coronary Artery Disease (AI-CAD) detection framework. The AI-CAD framework combines data integrity validation, image normalisation, vessel enhancement, topology-preserved segmentation, and graph-based classification Initially proposed. Confident Learning with Extended CleanLab is used for dataset reliablity check removes annotation noise from 17.4% to 3.2% and raises the baseline model AUC to 0.947. Next image I processed with propose A Self-Supervised Radiographic Normaliser (SSRN) method improves image quality, results in an average improvement of 13 dB in SNR and 92% decrease in device variance. Further proposed Deep Multi-Scale Vessel Transformer (DMVT) algorithm enhances vessels region. The proposed Graph-Enhanced Annotation Mapper (GEAM) is used for obtaining masks ROIs, which provides an accuracy of bifurcation to 96%. TransUNet++ is used for vessel segmentation, and provides a Dice score of 94.1%. Finally, Graph-Enhanced Coronary Classifier (GECC) is used to classify CAD severity. GECC provides AUC values up to 0.98 and sensitivity of 97.3% for multi-vessel disease.