Study of effectiveness of stockwell transform for detection of coronary artery disease from heart sounds

Recent studies show that in the case of coronary artery disease (CAD) patients, the diastolic part of cardiac cycle contains very weak murmurs caused by turbulent flow in the narrowed coronary arteries. This paper aims to analyze the diastolic period of heart sounds recorded with electronic stethoscope for detection of coronary artery disease. 3M Littmann 3200 model is used for recording of heart sounds of CAD and non CAD category. Cardiac sounds were analyzed with spectral and spectro-temporal features. Thirteen features including three spectral and ten spectro-temporal were extracted from each sound. Spectral features used for the study were spectral centroid, spectral roll off and spectral flatness. Spectro-temporal features include Stockwell transform with band wise kurtosis. It is observed that along with Spectral features, spectro-temporal feature provides significant discrimination between CAD and non-CAD heart sounds with sensitivity of 98.18% and specificity of 93.10%.