基于PPG信号的LSTM网络同步动脉血压预测被引量：10

Synchronous Blood Pressure Prediction Based on PPG Signals by LSTM Network

在线阅读下载PDF

导出

摘要为通过光电容积脉搏波信号获取动脉血压参数,并将其作为判断个人健康状况的依据,基于Tensorflow框架训练LSTM网络模型与传统RNN模型,使用625000条光电容积脉搏波数据序列通过忘记、选择记忆、输出阶段得出符合生理规律的血压参数,将两种模型放在125000条样本的测试集中进行有效性验证。实验结果表明,训练后的LSTM模型对血压的预测比传统RNN模型更准确,LSTM预测评价指标MAE、RMSE、STD和R2_score分别为4.05、8.78、7.42和0.89,且预测结果符合美国医疗仪器促进协会标准(MAE<5mmHg,STD<8mmHg),而传统RNN模型则为11.58、17.03、14.54和0.73。LSTM模型能较好地预测血压参数,在生物医学领域有较高的应用价值,其效果优于传统RNN模型。 In order to obtain the parameters of arterial blood pressure through the photoplethysmography signal,which is the basis of judging personal health status,the LSTM network model and the traditional RNN model were trained based on tensorflow framework,and 625000 photoplethysmography data sequences were used to obtain the blood pressure parameters that conform to the physiological rules through forgetting,selecting memory and outputting stages,and the validity of the two models was verified in 125000 sample test sets.The experimental results showed that the prediction of blood pressure of the trained LSTM model was more accurate than that of the traditional RNN model.The MAE,RMSE,STD and R2_score of LSTM were 4.05,8.78,7.42 and 0.89 respectively,and the pre⁃diction results were in line with the standards(MAE<5mmhg,STD<8mmhg)proposed by the American Association for the promo⁃tion of medical instruments,while the traditional RNN model was 11.58,17.03,14.54 and 0.73.The results show that LSTM model can predict the parameters of blood pressure better and has higher application value in biomedical field,and the effect is better than the traditional RNN model.

作者李帆程云章边俊杰耿晓斌 LI Fan;CHENG Yun-zhang;BIAN Jun-jie;GENG Xiao-bin(Shanghai Interventional Medical Device Engineering Technology Research Center,University of Shanghai for Science and Technology,Shanghai 200093,China;Zhejiang Shanshi Medical Equipment Co.,Ltd,Hangzhou 311100,China)

机构地区上海理工大学上海介入医疗器械工程技术研究中心浙江善时生物药械有限公司

出处《软件导刊》 2020年第8期44-48,共5页 Software Guide

基金上海工程技术研究中心资助项目(18DZ2250900)。

关键词同步血压预测 PPG信号 LSTM网络循环神经网络深度学习 synchronous blood pressure prediction photoplethysmography signal LSTM network recurrent neural network deep learning

分类号 TP312 [自动化与计算机技术—计算机软件与理论]

作者简介李帆(1996-),男,上海理工大学上海介入医疗器械工程技术研究中心硕士研究生,研究方向为无创血压实时监测技术;通讯作者:程云章(1964-),男,博士,上海理工大学上海介入医疗器械工程技术研究中心教授、博士生导师,研究方向为医疗器械大数据应用、血流动力学及其临床应用、精准医疗科学与工程;边俊杰(1978-),男,硕士,浙江善时生物药械有限公司董事长,研究方向为无创实时血压检测技术;耿晓斌(1994-),男,上海理工大学上海介入医疗器械工程技术研究中心硕士研究生,研究方向为无创血压实时监测技术。