Environmental noise elimination of heart sound based on singular spectrum analysis

Automatic heart sound (HS) auscultation enjoys advantageous features in terms of high intelligence, accuracy, and efficiency over traditional way. Unfortunately, sensitivity to noise corruption exposes automatic auscultation to misdiagnosis risks since original pathological features are vulnerable to miscellaneous HS noise. Therefore, HS denoising is pivotal to obtain qualified HS signal for further analysis and precise diagnosis. Traditional wavelet shrinkage (TWS) method achieves good performance on eliminating Gaussian distributed noise, yet it is powerless against randomly distributed environmental noise. To tackle such a bottleneck problem, an environmental HS noise elimination method based on singular spectrum analysis (SSA) is proposed in this paper. With the aid of singular value decomposition (SVD), effective eigenvalues related to the principle components (PC) of pure HS signal are selected to reconstruct HS signal while eliminating environmental noise efficiently. Validated using both normal and pathological HS signals with diversified environmental noises, the proposed method exhibits better denoising performance than TWS in most cases. As such, this work provides an attractive alternative for HS environmental HS noise denoising.