Segmentation and feature extraction of heart murmurs in newborns

Heart murmurs are typically detected in newborns, in many cases due to the patent ductus arteriosus condition. Hence, providing an automatic tool for heart sound analysis can be helpful to physicians in the diagnosis of heart murmurs in newborns. In this paper we propose a novel method that performs segmentation and feature extraction, aimed at separating heart sound signals in two parts: innocent and pathological murmurs. The segmentation of heart sounds into single cardiac cycles (Systole and Diastole murmurs) uses wavelet transform and k-mean clustering. Feature values are evaluated after segmentation by using time and frequency measures. Sound signal encoding is carried out using Shannon energy, bispectrum, and Wigner bispectrum. A comparative analysis has shown differences in their capability of preserving discriminant information. Experimental results highlight that all cited techniques perform well; however, the best results have been obtained by using Wigner bispectrum.