Myocardial territory segmentation on coronary computed tomography angiography images: Comparison between projection and non-projection methods in a pig model

Background Quantitative analysis on the segmentation of the myocardium territory supplied by each coronary artery (CA) may provide an important diagnostic and treatment tool for the identification of ischemic heart disease caused by coronary stenosis. In the previous study, CAs have been projected on the myocardial surface, and their supplying territories were segmented by the Voronoi diagram, which is still considered controversial due to lack of validation. Methods We proposed and compared two algorithms including coronary artery-based myocardial segmentation (CAMS) and the CAMS after projecting a CA onto the myocardium surface (PRJ_CAMS) to divide the left ventricle (LV) territory basdetection of regional anomalies of myocardial motion ed on supplying coronary arteries in coronary computed tomography angiography (CCTA) and validated their accuracy by pig models (N = 6) with a single coronary occlusion method. The gold standard of the territories was determined by the specimens confirmed by an expert cardiothoracic radiologist with more than 20 years’ experience. The dice similarity coefficient (DSC) was used as an evaluation metric. Results The CAMS method had significantly better accuracy in the aspect of DSC results of the two algorithms in CCTA images. The comparison of the DSC results for individual pigs and territories of the CAMS method shows 1.073, 1.100 times better in the left anterior descending (LAD) and right coronary artery (RCA), respectively, better than that of the PRJ_CAMS method, but the DSC value of left circumflex artery (LCX) is the same at 77%. The accuracy of the CAMS method shows 3.667, 4.545, 1.101 times better than that of the PRJ_CAMS method when predicting the LAD, the mid-region, RCA, and LCX territories, respectively. The overall DSC values were 82% for the CAMS method and 78% for the PRJ_CAMS method. In addition, the overall CAMS method accuracy was 61% and the PRJ_CAMS was 30%, more than double that of the PRJ_CAMS method. Conclusions The CAMS method is more accurate than the PRJ_CAM method in identifying myocardial territories based on the corresponding CA with a 3D Voronoi diagram.