Automated framework for CTA coronary segmentation and quantitative validation

摘要

Based on the fact that coronary heart disease (CHD) is the leading cause of death among all cardiovascular abnormalities, clinicians are keen in early detection and continuous monitoring of arterial atherosclerosis. Conventional cardiac catheterization method yields 2D angiograms of cardiac vasculature with possibly missed abnormalities as well as its invasive nature involves risk to the patient. These limitations motivated the research community for the non-invasive imaging modalities for vascular diagnosis. A prominent example is the effective use of computed tomography angiography (CTA) phenomena in clinical practice. The high temporal and spatial resolution of latest scanners has made possible to acquire a real time 3D view of moving heart; however, it becomes difficult for clinicians to traverse the data cloud for anomaly detection. Consequently, accurate segmentation of coronary tree is first step towards effective diagnosis of coronary atherosclerosis. We proposed a simple yet efficient framework for the coronary segmentation in 3D CTA volume using level set formulation of localized region based Chan-Vese energy model. Moreover, the quantitative validation of the segmented coronary tree has always been challenging due to non-availability of ground truth reference in clinical practice. The proposed framework facilitates manual experts to establish the coronary lumen boundaries based upon cross sectional and curved planar reformation analysis. Consequently, the proposed model computes quantitative accuracy against manual ground truth using two different similarity measures of Jaccard index and the Dice coefficient. We believe that this framework can help research community in effective coronary analysis in terms of instant clinical ground truth annotations and effective validation of the segmented surface.