Coronary vessel segmentation in coronary angiography with a multi-scale U-shaped transformer incorporating boundary aggregation and topology preservation
Guangpu Wang,Peng Zhou,Hui Gao,Zewei Qin,Shuo Wang,Jinglai Sun,Hui Yu
DOI: https://doi.org/10.1088/1361-6560/ad0b63
IF: 3.5
2024-01-12
Physics in Medicine and Biology
Abstract:Coronary vessel segmentation plays a pivotal role in automating the auxiliary diagnosis of coronary heart disease. The continuity and boundary accuracy of the segmented vessels directly affect the subsequent processing. Notably, during segmentation, vessels with severe stenosis can easily cause boundary errors and breakage, resulting in isolated islands. To address these issues, we propose a novel multi-scale U-shaped transformer with boundary aggregation and topology preservation (UT-BTNet) for coronary vessel segmentation in coronary angiography. Specifically, considering the characteristics of coronary vessels, we first develop the UT-BTNet for coronary vessels segmentation, which combines the advantages of a convolutional neural networks (CNN) and a transformer, and is able to effectively extract the local and global features of angiographic images. Secondly, we innovatively employ boundary loss and topological loss in two stages, in addition to the traditional losses. In the first stage, boundary loss is adopted, which has the effect of boundary aggregation. In the second stage, topological loss is applied to preserve the topology of the vessels, after the network converges. In the experiment, in addition to the two metrics of Dice and intersection over union (IoU), we specifically propose two metrics of boundary intersection over union (BIoU) and Betti error to evaluate boundary accuracy and the continuity of segmentation results. The results show that the Dice is 0.9291, the IoU is 0.8687, the BIoU is 0.5094, and the Betti error is 0.3400. Compared with the other state-of-the-art methods, UT-BTNet achieves better segmentation results, while ensuring the continuity and boundary accuracy of the vessels, indicating its potential clinical value.
engineering, biomedical,radiology, nuclear medicine & medical imaging