Heart Sound Classification Algorithm Based on Neural Network

Rui Yu1, Guangyu Bin1, Ziyang Xu2, Fengya Liu1, Zhenbo Han1
1Beijing University of Technology, 2Johns Hopkins University


Background and objective: Analysis of the location, period, nature, intensity, and conduction direction of heart murmurs is of great significance for the diagnosis of valvular heart disease and some congenital heart diseases. Method: We first segment the heart sound signal, and obtain the temporal characteristics of the heart sound according to the heartbeat cycle, such as the heart sound interval. Then the heart sound signal is transformed into the frequency domain to obtain the frequency signals of its different sub-bands. For each signal, a characteristic heart sound of a single heartbeat is generated according to the segmentation result, and it is input into the convolutional neural network to obtain its morphological features. Finally, all feature vectors are input into the neural network for discrimination. Result: Under the scoring rules of Physionet 2022, our final score is 1653, proving the effectiveness of our method.