Automated Detection of Heart Murmurs From the PCG Signal Using Stationary Wavelet Transform

Abstract

Heart sound (HS) murmur detection in a phonocar-diogram (PCG) signal is essential for primary cardiac healthcare units. This work proposes a stationary wavelet transform (SWT) to automatically detect four HS murmurs, namely mitral valve prolapse, mitral regurgitation, aortic stenosis, mitral stenosis, and regular HS. First, the PCG signal is preprocessed and segmented into cardiac cycles. Second, SWT-based wavelet decomposition is carried out on each cardiac cycle to get different wavelet subband signals. Four novel features, relative wavelet subband energy, log-entropy energy, multi-scale kurtosis, and maximum absolute deviation, are extracted from each subband. Finally, these features are fed to a supervised random forest (RF) classifier for murmur detection. The efficacy of the proposed approach is verified using a multiclass heart murmur database. The proposed method shows a better performance result by achieving an overall accuracy (OA) of 99.28%, 99.27% precision, 99.27% recall, and 99.27% F1-score. The proposed method is competitive enough with the existing state-of-the-art methods. The performance results show that the proposed work is suitable for the preliminary detection of heart murmurs.