A Comprehensive Overview of Heart Sound Analysis Using Machine Learning Methods

Cardiovascular diseases (CVDs) are a prevalent cause of mortality worldwide, and the traditional cardiovascular disease diagnosis has relied heavily on stethoscope-based heart sound auscultations. Hence, the development of medical systems has led to several machine learning (ML)-related studies for analysing heart sounds using phonocardiograms (PCGs). Although this process enables the detection of additional mechanical activity information about the heart muscle, various crucial gaps (background noise, unbalanced datasets, and irrelevant extracted features) are still observed. Therefore, this review examined the advancements in diagnosing heart sounds using ML-based PCG signal analysis. The evaluation process involved three datasets in the past five years (2019–2024): the PhysioNet/Computing in Cardiology (CinC) Challenge (2016–2022) and the Yaseen Khan 2018 datasets. This review also comprehensively explained the importance of heart sounds in cardiovascular disease diagnosis. Moreover, these studies demonstrated an overview of the most effective methods in handling datasets, pre-processing of PCG signals (filtering, segmentation, and normalisation), PCG signal processing techniques (feature extraction and feature selection), and machine learning models (classification). Several directions concerning heart sound diagnosis for future studies are then presented, which can serve as a reference point for diagnosing cardiovascular diseases.