Noninvasive detection of coronary artery disease based on heart sounds

In this study, four channel heart sounds of normal persons and patients with coronary artery disease (CAD) were detected by a highly sensitive sound sensor array placed in special positions on the thorax. The acquired signals were analyzed using wavelet transform and neural networks. The wavelet transform is a highly applicable method to separate complicated heart sounds into various frequency segments and can also give time localization of the event in the cardiac cycle. Each and every channel's four average power ratios of whole cycle heart sound to diastolic period heart sound at four frequency segments based on WT coefficients were calculated. The sixteen power ratios of four channels decomposed into sixteen parameters input pattern of radial basis function (RBF) neural networks. After training, the neural networks can diagnose the CAD automatically.