Computerized Classification Method for Significant Coronary Artery Stenosis in Whole-Heart Coronary Magnetic Resonance Angiography Images

The purpose of this study is to develop a computerized classification method for significant coronary artery stenosis in whole-heart coronary magnetic resonance angiography (WHCMRA) images using a three-dimensional convolutional neural network (CNN) enhanced with attention mechanisms. Our database included 951 segments from WHCMRA images (pixel size = 0.645 mm) obtained from 75 patients. Forty-two segments with significant stenosis (luminal diameter reduction ≥ 75%) were annotated on WHCMRA images by an experienced radiologist, whereas 909 segments without stenosis were annotated at representative sites. In the proposed method, high-resolution WHCMRA images (pixel size = 0.3225 mm) were generated from the original WHCMRA images using a CNN-based super-resolution model that extends conventional two-dimensional architecture into a three-dimensional framework. Volumes of interest, centered on annotated points, were extracted from these high-resolution WHCMRA images. The three-dimensional CNN for classifying coronary artery stenosis consists of two feature extractors, two attention mechanisms, and a classifier. The proposed method achieved an area under the receiver operating characteristic curve of 0.954, indicating a substantial impact on the interpretation of WHCMRA.