Automatic stenosis detection using SVM from CTA projection images

Identification of stenosis in computed tomography angiography (CTA) image of a heart is a challenging task. In this paper, we propose an automated support vector machine (SVM) based approach that detects the branches and stenosis in 2D projection images obtained from different rotation angles of CTA image of a heart. Coronary arteries are segmented from the projection images, centerlines of the arteries are obtained and the presence of stenosis is detected by tracking the arteries along the vessel direction. Tracking is done by sliding overlapping windows in the estimated vessel direction obtained by combining geometric and intensity directions of the vessel. Different SVM models have been built for branch and stenosis detections using geometric and shape based features obtained from the sliding window regions. The proposed system was evaluated in terms of Precision and Recall using CTA images obtained from Billroth Hospitals, Chennai, India, and the results are encouraging.