Classification of heart sound short records using bispectrum analysis approach images and deep learning

AbstractThe diagnosis of cardiac disorders using heart sounds is one of the hottest topics in recent years. In general, diagnosing in the early stage is usually performed using routine auscultation examination using a stethoscope which requires human interpretation. Recording of heart sounds using an electronic microphone embedded inside the stethoscope provides a digital recording which is known as a phonocardiogram (PCG). This PCG signal carries very informative data about the status of the heart and its valves. Recently, several machines and deep learning techniques employed signal processing to classify heart disorders using PCG. Based on the used datasets, heart sound can be exploited to classify five types of heart sounds, one is normal, and the others are abnormal and two classes of heart sound, normal and abnormal. This research used a modified version of previously proposed convolutional neural network (CNN) which is AOCTNet architecture for automatic diagnosis of heart valves conditions based on higher order spectral estimation using bispectrum of heart sounds recordings. The results show that the proposed system has a comparable performance comparing to other methods. The methodology proposed in this paper can detect heart valves disorders using PCG signals with an overall accuracy of 98.70 and 97.10% using full bispectrum images and contour bispectrum images, respectively, for five classes dataset and overall accuracy of 99.47 and 98.74% using full bispectrum imag...
Source: Network Modeling Analysis in Health Informatics and Bioinformatics - Category: Bioinformatics Source Type: research