Machine Learning-Based Classification of Heart Sound Using Hilbert Transform

Abstract

Phonocardiogram (PCG) or heart sound signal administers crucial information for the diagnosis of various cardiovascular affliction. The heart sound classification is a confronting task in the field of modern healthcare. This paper confers the heart sound classification using the Hilbert transform envelope technique. The major constituents in the classification stage are preprocessing of PCG signal, features (temporal, spectral, and statistical) extraction, machine learning, and features-based classification of PCG signal. The accuracy and firmness of the proposed method are evaluated using two different datasets with different classes. The heart sound signals are taken from the standard phonocardiogram databases, i.e., PASCAL and PhysioNet/CinC. Evaluation results manifest that the proposed method for PCG signal classification achieves an overall accuracy (A) of 97.7% for the PASCAL dataset and overall accuracy (A) of 98.8% for PhysioNet/CinC dataset. Comparative results manifest that the proposed method is capable of classification of the PCG signal. Further, the method permits extraction of appropriate features for the classification of the PCG signal.