Fully automated identification of heart sounds for the analysis of cardiovascular pathology

Abstract

Cardiac disorders are spreading rapidly all over the world, and as per the World Health Organization (WHO), 17.5 million people die each year due to cardiovascular diseases (CVD). So there is a dire need to develop cost-effective, time-efficient, and fully automated solutions to diagnose cardiovascular abnormalities. Many researchers have worked on detecting CVD from electrocardiogram (ECG) signals. ECG signals give reliable information about cardiac pathology; however phonocardiogram (PCG) signal provides an easy, cost-effective, objective, and comprehensive information about cardiovascular abnormalities by measuring heartbeats. This paper presents a fully automated robust clinical decision support system that can identify cardiovascular pathology by analyzing heart sounds from PCG signals. The proposed system was tested on 55 PCG signals from which 24 samples contained healthy and 31 samples contained abnormal heart sounds. The proposed system correctly classified healthy and diseased samples with the accuracy, sensitivity, and negative predictive value (NPV) of 87.2%, 96.7%, and 94.7%, respectively.

Keywords: Phonocardiogram (PCG), Savitzky-Golay filter, Naïve Bayes, Support vector machines (SVM)