Heart sound segmentation based on SMGU-RNN

Conference: BIBE 2019 - The Third International Conference on Biological Information and Biomedical Engineering
06/20/2019 - 06/22/2019 at Hangzhou, China

Proceedings: BIBE 2019

Pages: 7Language: englishTyp: PDF

Personal VDE Members are entitled to a 10% discount on this title

Xu, Chundong; Zhou, Jing; Li, Lan; Li, Qinglin (Faculty of Information Engineering, Jiangxi University of Science and Technology, Ganzhou, China)
Wang, Jing (School of Information and Electronics, Beijing Institute of Technology, Beijing, China)
Ying, Dongwen (Faculty of Information Engineering, Jiangxi University of Science and Technology, Ganzhou, China & Key Laboratory of Speech Acoustics and Content Understanding, Institute of Acoustics, Chinese Academy of Sciences, Beijing, China)

Heart sound segmentation is one of the difficulties in heart sound analysis. The works of how to effectively segment heart sounds was studied based on deep learning. The correlation between the front and back frames of the heart sounds is helpful to improve the state recognition accuracy of the current frame. The recurrent neural network (RNN) based on long short-term memory (LSTM) unit can effectively combine this through the gate unit. A simpler minimum gated unit (SMGU) was suggested based on the minimum gated unit (MGU) in this study. The heart sound database was constructed by open source data sets and self-collected, the effectiveness of the segmentation method was verified by comparison with MGU-based, LSTM-based, convolutional neural network based, deep neural network based, RNN-based, auto-encoder-based, machine learning and threshold-based classifiers. The experimental results showed that the SMGU-RNN achieves great results in segmentation (Accuracy-88.5%), and the time complexity was significantly reduced.