PCG Classification Using Scalogram And CNN With DAG Architecture

Abstract

Cardiovascular diseases (CVDs) are the most leading causes of death every year in the world. The threat of CVDs can be decreased and controlled with early diagnoses. Therefore, interpreting heart sounds is considered as one of the common ways to diagnose the cardiovascular system. Heart sound signal as known as phonocardiogram (PCG) provides useful information about the heart condition, which can be used in the diagnostic, and helps the physicians in the detection of several cardiovascular abnormalities. The technology development helped in the appearance of new diagnosis techniques, which combines new advanced signal processing techniques and deep learning algorithms. Thus, the heart sound classification is becoming a crucial task in the modern healthcare field. In this work a deep learning-based classification method was proposed. Using PCG database which contains five different classes taken from different cases of heart valve defects. Scalogram of heart sound signals was used as time-frequency representation to create a scalogram image database extracted from the PCG database. A convolutional neural network with Direct Acyclic Graph structure (DAG CNN) was used in the classification of the scalogram image database. The evaluation of the classification performance indicated that the accuracy was about 99,6%. A comparative results manifest that the proposed method had a better performance compared to other previous works in which the same database was used.