Analysis of PCG signals for Automated Diagnosis of Cardiovascular Diseases (P-0298) (MFN 6066)

Abstract

Cardiac disorders are spreading all over the world very rapidly. As per World Health Organization (WHO), due to cardiovascular diseases (CVDs) approximately 17.5 million people die each year. So, the development of automated diagnosis of cardiovascular diseases through analysis of PCG signals is an innovative, cost effective, and time efficient solution to diagnose cardiovascular abnormalities. Due to lack of knowledge and resources available in rural areas, people cannot monitor their heart beats regularly and the main problem occurs in this disease because before a major loss has been done, no significant pain occurs. CVDs are being detected from electrocardiogram (ECG) signals by many researchers. The essential cause to design this system is limited number of cardiologists and necessity to diagnose CVD at right time. This paper presents a fully automated clinical decision support system that can identify cardiovascular pathology by analyzing heart sounds from PCG signals. First of all, the input signal is loaded into the proposed system which is de noised using Savitzky-Golay filter. Then the candidate signal is decomposed into multiple bands to identify S1 and S2 heart sounds through multi resolution wavelet analysis. Afterwards, the proposed system extracts five distinct features from the candidate signal which are passed to the proposed classification system to determine the cardiac pathology. The classification in our proposed system is based on an ensemble of supervised K-Nearest Neighbors (KNN), Naïve Bayes and Support Vector Machines (SVM), trained on publicly available pattern analysis, statistical modeling and computational learning (PASCAL) data set. The proposed system was tested on 55 PCG signals from which 23 samples contained healthy heart sounds and 32 samples contained abnormal heart sounds. The proposed system correctly classified healthy and diseased samples with the accuracy and negative predictive value (NPV) of 87.3%, 96.7% and 94.6% respectively