Comparative analysis of methods for identifying tree structures of coronary vessels

The paper presents a comparative analysis of the performance of machine learning models and one filter method used for semantic segmentation of the coronary vessels based on the source coronary angiographic image. Seven machine learning models were tested: UNet3+, AngioNet, Reg-SA-UNet++, EfficientUNet++ B5, SE-RegUNet 4GF, SE-RegUNet 16GF, FR-UNet and one filter method, which was implemented as part of the paper. Despite the impossibility of determining the exact hierarchy of model performance, based on the results of statistical tests, the model that presented the best results with accuracy of 97,7% was distinguished – FR-UNet and the model that showed the lowest quality – UNet3+.