Coronary Full artery segmentation using U-Net neural network architecture

Coronary artery disease (CAD) ranks among the most widespread heart conditions today. The World Health Organization's recent reports show cardiovascular diseases are now claiming lives faster than any other cause globally. What happens is that plaque slowly builds up inside blood vessels condition known as atherosclerosis, causing them to narrow and stiffen over time. This progression subsequently causes ischemic changes in tissues or organs, increasing the risk of angina, myocardial infarction, and other cardiovascular events This study presents an optimized U-Net architecture for coronary artery segmentation using the ARCADE dataset, achieving ۹۷.۰۸% pixel accuracy (loss: ۰.۰۸۱۰), ۰.۷۵۷۷ precision, and ۰.۵۷۰۲ recall. Three critical findings emerge from our analysis: ۱) Metric Discrepancy, ۲) Architectural Optimization, and ۳) Clinical Implications. These results emphasize the need for specialized evaluation protocols in medical image segmentation, where traditional computer vision metrics must be supplemented with clinically relevant measures. The study provides concrete guidance for future ARCADE benchmark development, particularly regarding annotation standardization for thin tubular structures and the adoption of compound loss functions to address class imbalance.

کلیدواژه ها:

Medical image segmentation ، U-Net ، evaluation metrics ، Angiography ، Coronary artery disease