Accurate vessel segmentation and plaque detection in coronary angiography (CAG) images are crucial for the intelligent diagnosis of coronary artery disease. To address the performance bottlenecks of existing deep learning methods in handling complex vascular topology, small branches, and low-contrast plaques, this study proposes a deep learning-assisted diagnostic system based on an enhanced Pyramid Scene Parsing Network (PSPNet). By incorporating a Multi-scale Feature Fusion (MSF) module, a Convolutional Block Attention Module (CBAM), and a Wavelet Feature Enhancement Module (WFEM), the system significantly improves the perception of multi-scale vascular structures and the focus on minute plaque regions. Evaluated on a dataset comprising 1,200 clinical CAG images, the proposed system achieved Dice coefficients of 94.8% and 90.8% for main vessel and small branch segmentation, respectively. In the plaque detection task, it attained a precision of 89.1%, a recall of 86.5%, an F1-score of 87.8%, and an AUC value of 92.4%, outperforming current mainstream segmentation models. Ablation studies confirmed the effectiveness and synergistic contributions of each proposed module. This system enables a fully automated pipeline from image preprocessing to segmentation, detection, and visual analysis. It not only provides a high-precision and robust technical solution for the intelligent diagnosis and risk assessment of coronary artery disease but also advances the development of interpretable, end-to-end assisted diagnostic systems for complex medical imaging.
Li et al. (Tue,) studied this question.