A Coronary Artery Segmentation Method Based on Mask RCNN Integrated with Geometric Features

Abstract

Coronary artery segmentation is a major step in coronary heart disease computer-aided diagnosis systems, which is used to ensure only the coronary artery region to be processed in subsequent steps. The coronary computed tomography angiography (CCTA) image has its own inherent characteristics, such as unclear boundary, complicated structure, and inconspicuous features, which make the CCTA image segmentation a difficult task. To address this problem, the fusion of geometric features of the image can enhance the feature learning ability of network, thereby improving the segmentation accuracy. Therefore, we proposed a coronary segmentation method that integrates the geometric features into Mask RCNN. Boundary extraction algorithm and fractal feature extraction algorithm are proposed to extract boundary and fractal features. Through the feature fusion method, the boundary features and fractal features are fused into the network. Experimentally, we evaluated the proposed method by using coronary artery dataset. The average accuracy (PA) and Dice coefficient of the proposed method reached 83% and (84.0±10.1)%, respectively. The test results demonstrate that the proposed method provides greater accuracy and robustness.

Keywords

• coronary artery segmentation
• mask rcnn
• geometric feature
• fractal dimension