Semi-supervised Deep Learning of Vessel Segmentation in Coronary Angiography

Abstract

Vessel segmentation and stenosis classification in coronary angiography have guiding significance for doctors in the diagnosis of coronary artery disease. We used a total of 120 coronary angiography images collected by ourselves and labeled 70 images, which include 50 images for supervised training, 10 images for verification and 10 images for testing. For the vessel segmentation task, we used a traditional threshold segmentation method such as the Otsu method, a supervised deep learning method such as the UNet network. In order to make full use of the other 50 unlabeled images, we propose a semi-supervised deep learning method based on the semi-supervised adversarial network. The segmentation results of the semi-supervised deep learning method are better than the results of the supervised deep learning method, whose segmentation results are better than the results of the traditional method.