Learning to Detect Coronary Artery Stenosis from Multi-Detector CT imaging

Recent advances in multi-detector CT (MDCT) are turning it into one of the leading technologies for detection of coronary artery disease. We propose here an automatic coronary stenosis detection mechanism that uses computerized tomography coronary angiography (CTCA) images extracted from the MDCT data. The artery is first isolated using image morphology and its boundaries then estimated using dynamic programming. A multi-scale descriptor is extracted for every artery point and a decision algorithm based on this descriptor is used. The decision function is learned by a support vector machine (SVM) from CTCA images, where arterial lesions were detected and quantified by a team of expert physicians. The system was tested and the results compared with those obtained through visual interpretation of the CTCA images and with quantitative coronary angiography. A very good agreement between the automatic and the human interpretation of CTCA