Heart Sound Segmentation Based on Mel-Scaled Wavelet Transform

Abstract

The identification and segmentation of heart beats in a phonocardiogram signal is of great interest, with applications ranging from diagnosis to use as a timing source. This paper proposes a new algorithm for the identification of the first and second (S1 and S2) heart sounds and for segmentation of the signal. The proposed algorithm is a novel combination of documented techniques, leading to improved segmentation accuracy. The Shannon Energy is first used to find sounds of interest. The algorithm subsequently uses the Mel-Scaled WaveleTransform (MSWT) which is a modified Mel-Frequency Cepstral Coefficient (MFCC) algorithm with the Discrete Wavelet Transform (DWT) in order to reduce the impact of noise on the coefficients. The coefficients and sounds of interest are used to distinguish S1 from S2 and segment the signal accordingly. The algorithm is tested on real signals and is compared to a simpler Shannon Energy algorithm and to a traditional MFCC based algorithm. The new algorithm presents an improvement in accuracy especially when signals contain noise. It is therefore less susceptible to outside interference and could be used more accurately in a hospital setting.