Automatic phonocardiogram signal analysis in infants based on wavelet transforms and artificial neural networks

Abstract

Discusses infant related aspects of a computer based phonocardiogram analysis system. The evaluation is achieved in several stages. The first step is the segmentation of the heart sound signal in single cardiac cycles. For further analysis artefact free periods are regarded, which are automatically selected by a new algorithm. In the following features for the automatic classification are calculated using a wavelet transform. Finally a diagnostic proposal is determined utilising the calculated features by two artificial neural networks, that were trained with reference databases. The first network serves for murmur detection and the second for classification of the particular disease. The murmur detection yields about 93% correct classified signals. All cases used in this investigation have been counterchecked and verified by echocardiography.