Classification of Cardiovascular Disease Using 2D-Image Representations of Phonocardiogram Signals

Abstract

Auscultation of the heart is a process that refers to listening to heart sounds to detect abnormalities. It is a good way to diagnose multiple cardiac problems. However, this process requires experienced professionals. The waveforms of a phonocardiogram (PCG) are useful in identifying disorders like these. This work employs self-acquired PCG recordings to develop an intelligent classification model that will serve as an objective diagnostic tool for physicians diagnosing heart disorders based on sound. Our model uses spectrograms and scalograms representations of PCG signals and uses a convolutional neural network to learn suitable features for PCG classification into normal and abnormal recordings. Performance comparison with similar studies reflects the efficacy of our model.