Heart sound classification using machine learning and phonocardiogram

This research pertains to classification of the heart sound using digital Phonocardiogram (PCG) signals targeted to screen for heart ailments. In this study, an existing variant of the decision tree, i.e. XgBoost has been used with unsegmented heart sound signal. The dataset provided by PhysioNet Computing in Cardiology (CinC) Challenge 2016 has been used to validate the technique proposed in this research work. The said dataset comprises six databases (A–F) having 3240 heart sound recordings in all with the duration lasting from 5–120 s. The approach proposed in this paper has been compared with 18 existing methodologies. The proposed method is accurate with the mean score of 92.9, while sensitivity and specificity scores are 94.5 and 91.3, respectively. The timely prediction of heart health will support specialists to attain useful risk stratification of patients and also assist clinicians in effective decision-making. These predictive facts may serve as a guide to provide improved quality of care to the patients by way of effective treatment planning and monitoring.