文献类型：期刊文献

英文题名：Automated Segmentation of the Systolic and Diastolic Phases in Wrist Pulse Signal Using Long Short-Term Memory Network

作者：Huang, Lin[1];Yan, Jianjun[1];Cai, Shiyu[1];Guo, Rui[2];Yan, Haixia[2];Wang, Yiqin[2]

机构：[1]East China Univ Sci & Technol, Inst Intelligent Percept & Diag, Sch Mech & Power Engn, Shanghai 200237, Peoples R China;[2]Shanghai Univ Tradit Chinese Med, Shanghai Key Lab Hlth Identificat & Assessment, Comprehens Lab Tradit Chinese Med Four Diagnost In, Shanghai 201203, Peoples R China

年份：2022

卷号：2022

外文期刊名：BIOMED RESEARCH INTERNATIONAL

收录：;WOS:【SCI-EXPANDED(收录号:WOS:000848414000013)】；

基金：This study was supported by the National Natural Science Foundation of China (No. 82074332, No. 81673880), Shanghai Science and Technology Committee Funding (No. 19441901100), and Shanghai Science and Technology Funding (No. 21DZ2271000).

语种：英文

摘要：Purpose. Single-period segmentation is one of the important steps in time-domain analysis of pulse signals, which is the basis of time-domain feature extraction. The existing single-period segmentation methods have the disadvantages of generalization, reliability, and robustness. Method. This paper proposed a period segmentation method of pulse signals based on long short-term memory (LSTM) network. The preprocessing was performed to remove noises and baseline drift of pulse signals. Thus, LabelMe was used to label each period of the pulse signals into two parts according to the location of the starting point of main wave and the dicrotic notch, and the dataset of the pulse signal period segmentation was established. Consequently, the labeled dataset was input into the LSTM for training and testing, and the results were compared with sum slope function method. Result. The remarkable result with the whole period segmentation accuracy of 92.8% was achieved for the segmentation of seven types of pulse signals. And the segmentation accuracies of the systolic phase, diastolic phase, and whole period using this method were higher than those of the sum slope function method. Conclusion. LSTM-based pulse signal segmentation method can achieve outstanding, robust, and reliable segmentation effects of the systolic phase, diastolic phase, and whole period of pulse signals. The research provides a new idea and method for the segmentation of pulse signals.