Significance of frequency domain features of pcg records for murmur detection - An investigation

Abstract

Automated identification of valve ailments from heart sound is a popular method an is a skilled task in cardiology. However, the automated methods, mainly depend upon the features extracted from the heart signal. The analysis of Phonocardiogram (PCG) signals, supply significant data about the heart functioning. In this paper, the significance of frequency domain features of Phonocardiogram (PCG) records for murmur detection is inspected. Frequency domain features like Dominant Frequency (DF), Spectral Centroid (SC), Spectral Flux (SF), Spectral Role-off (SR) and Median Frequency (MF), may be used in Artificial Intelligence (AI) to replicate the physical aspects of signals. The first three features are detected directly from the preprocessed heart signal, and the last two features are calculated from the Power Spectral Density (PSD) by means of an analytical method. It has been noticed that, among the features, MF and SR are superior to detect the presence of murmur than others. Besides, they are statistically more significant than all other features. The MF and SR can identify the murmur with an accuracy of 87.35% (dataset1), 76.67% (dataset2) and 80.59% (dataset1), 78.33% (dataset2), respectively without using any classifiers.