Fast Abnormal Heart Sound Detection Method Based on Multi-struct Neural Network

Abstract

Cardiovascular Disease is the most fatal disease in human history and the number one cause of human death. Many techniques are used to investigate the heart health but the most effective one is based on heart sound signals which contains significant information that reflect heart health. Studying heart sound signals is helpful for early diagnosis and prevention of cardiovascular diseases. In order to make full use of the global information of heart sounds, this paper proposes an end-to-end abnormal heart sound classification neural network based on LSTM and 1D-CNN, which can upgrade, rectify, and map time-frequency features. This model fully detects the hyperparameter space, so that it can effectively provide the fitting ability. A Performance test is executed on the PhysioNet dataset and this model shown the accuracy is near to 100. Besides, a data compression algorithm based on 1D-CNN is used, which can reduce the amount of calculation by 60%. Compared with the existing methods, the HSC-LNet shows excellent robust and speed in the classification of heart sound signals.