Time-Scale-Based Segmentation for Degraded PCG Signals Using NMF

Abstract

This article deals with the challenging problem of segmenting narrowly spaced cardiac events (S1 and S2) in noisy phonocardiogram (PCG) signals by using a novel application of NMF based on time-scale approach. A novel energy-based method is proposed for the segmentation of noisy PCG signals in order to detect cardiac events, which could be closely spaced and separated by noisy gaps. The method is based on time-scale transform as well as nonnegative matrix factorization (NMF) and the segmentation problem is formulated using the paradigm of binary statistical hypothesis testing. The energy of the Morlet wavelet transform and NMF output is employed as a test statistics for segmentation where the number of scales are selected based on the preferences calculated along the time-scales. The simulation results using real recorded noisy PCG data that provide promising performance with high overall accuracy on the segmentation of narrowly separated, high noisy signals by our proposed method.

Localization and classification of heart beats in phonocardiography signals —a comprehensive review

“…There is a growing trend towards using wavelet in combination with other operators like Teager energy operator (TEO) and non-negative matrix factorization (NMF). Sattar et al [74] used NMF for PCG segmentation. Ramovic et al proposed a system for human authentication based on wavelet and TEO [75] while fetal heart sound detection is carried out by Koutsiana using wavelet and fractal dimensions [76].…”