Models of Cardiac Artery Segmentation From Coronary Angiographic Images

Abstract

Cardiovascular diseases remain the leading cause of death. Mortality can be reduced and diagnostic accuracy can be improved by developing artificial intelligence-based solutions. The most important task here is segmentation of cardiac arteries. Accurate segmentation of coronary vessels on angiographic images is critical for detecting stenosis and planning interventional procedures. However, automated segmentation methods face a number of problems associated with difficulty in recognizing cardiac arteries: uneven distribution of contrast agent, motion artifacts, and superposition of shadows from anatomical structures. This paper presents a vessel segmentation model based on a modified U-Net architecture, including residual blocks and attention mechanisms (SCSE) with pre-training of the encoder on an artificial dataset for extracting vascular features. Particular attention is paid to the model's resistance to noise, a key problem of angiographic data that many existing methods cannot cope with. The model was trained on a mixed dataset of 1285 2D coro-nary angiogram images annotated by experts. The proposed approach achieved an IoU of 0.54 and an F1-score of 0.79 on the test set, demonstrating robustness to noise and artifacts. Additional post-processing with adaptive filtering improved the quality of binary masks, eliminating false positives from catheters and metal objects. External evaluation on 50 images from an independent clinical dataset showed an IoU of 0.50 and an F1-score of 0.75, which outperforms baseline segmentation methods such as the classical U-Net (IoU 0.42). The results highlight the effectiveness of the proposed model for vessel seg-mentation in real-world angiograms and demonstrate the potential of the approach as a basis for subsequent 3D reconstruction of the vascular network, which may improve diagnosis and treatment planning for coronary artery stenosis.