Detection of Heart Murmurs for Imbalanced Dataset Using Adaptive Synthetic Sampling Approach

Abstract

Cardiovascular diseases (CVDs) are potentially detected using heart sound characteristics. The presence of pathological heart murmurs is associated with cardiac abnormalities. In this paper, a noble automatic detection framework that employs a cascaded variational mode decomposition (VMD) and acoustic-based features of the nonstationary phonocardiogram (PCG) signals is proposed. The main objective is to investigate whether a reliable detection of heart murmurs can still be achieved when the dataset is imbalanced. In light of this, the proposed work has been extended for the imbalanced dataset by incorporating Adaptive Synthetic (ADASYN) sampling approach that learns from the minority samples to detect heart murmurs as the performance of the conventional classifiers degrades when the dataset is imbalanced. The result analysis shows the promising outcome on both the balanced and imbalanced datasets in the correct recognition of heart murmurs thereby paving a way to assist cardiologist for the diagnosis of heart diseases.