Heart Sound Segmentation toward Automated Heart Murmur Classification in Pediatric Patents

Abstract

A computerized heart murmur classification requires localization of two primary heart sounds, S1 and S2. In this paper, we propose an automatic segmentation method for pediatric heart sounds. After identifying S1 and S2 candidates from the envelope signal, we identify the best S1 and S2 pairs from all possible combinations of candidates by examining the signal correlation and cardiac cycle information. The performance of the algorithm was evaluated on normal heart sounds, innocent murmurs, and pathological murmurs. The proposed algorithm yields an overall sensitivity of 96.7% and positive predictive value of 98.0%.