Coronary Vessel Segmentation in X-ray Using U-Net

Cardiovascular disease remains a significant global health concern, causing substantial morbidity and mortality. This study addresses the limitations of subjective variability in traditional diagnostic modalities, particularly angiography, which assesses coronary artery blockages using X-ray images and radio opaque dye. Employing deep learning, the research utilizes a pre-trained U-Net model to efficiently recognize coronary artery structures in X-ray angiography images. Despite challenges like weak contrast and deformable vessel shapes, the model accurately performs image segmentation and calculates vessel stenosis. Achieving a mean F1 score of 0.921, precision of 0.938, and accuracy of 0.915, the model consistently outperforms benchmarks with lower standard deviations. Stenosis identification demonstrates significant prediction accuracy. The study's future implications include predicting stent placement positions and estimating post-stenting circulation success. This innovative approach offers an objective analysis, enhancing diagnostic accuracy in cardiovascular health.