Hierarchical Auto-labeling of Coronary Arteries on CT Coronary Angiography Images

Abstract

The auto-labeling of coronary artery segments plays an important role in the diagnosis of cardiovascular diseases. Due to the high degree of complexity and diversity in coronary artery structures, it is still a very challenging task after many years of exploration and study. In this paper, we propose a hierarchical scheme based on PointNet++ models and new topological structural features for automatic labeling of coronary artery segments. The inputs are 3D coronary artery centerline points extracted from CTCA images, and the outputs are the correspondent label indexes. The auto-labeling scheme include two stages: first stage is to identify the three main branches, LAD(LM), LCX and RCA. After that, utilizing the topological connectivity relationship with the three main branches, the indexes of sub-branches are identified in the second stage. We evaluated our method on a private clinical dataset. Experimental results show that the proposed method has achieved a satisfactory accuracy for clinical use.