A Robust Algorithm for Segmentation of Phonocardiography Signal Using Tunable Quality Wavelet Transform

Segmentation of the phonocardiography (PCG) signal is of importance for diagnosis of heart health and the same is recently reported to be done with tunable quality Wavelet transform (TQWT). In the reported method, parameters of TQWT are tuned to vary the frequency range of the approximation level such that its kurtosis gets maximized. Kurtosis will be larger for the fundamental heart sounds (FHS) as compared to the murmur because probability distribution of FHS is super-Gaussian and murmur is sub-Gaussian. However, this is observed in some pathological cases with sharp peak murmurs, murmur and FHS are indistinguishable using kurtosis. Since Fano factor gives a measure of relative variability, we propose to use Fano-factor for selection of the sub-band, in place of Kurtosis. The proposed method decomposes the signal up to twenty levels only once as compared to approximately 55 times in the reported method and then selects a sub-band considering the detailed levels only. This result into 25 times reduction in computational cost. Moreover, exclusion of the approximation level also helps in reduction of the real-life noise. Another contribution of the paper is the adaptive thresholding method used to suppress the noise level from the selected level that improves the segmentation accuracy in the presence of noise. Experiments are performed on PCG signals of various normal and pathological cases, with and without simulated noise using white Gaussian, pink, and red noise models, separately. The obtained results show the efficiency of the proposed method to segment the PCG signal in the presence of noise, murmurs, and both.