基于TCN-SE神经网络模型的智能连续血压估计方法  

Intelligent and Continuous Estimation of Blood Pressure Method Based on TCN-SE Neural Network Model

作　　者：熊嘉豪 姜晨希 陈永毅 张丹[1] 尹武涛 XIONG Jiahao;JIANG Chenxi;CHEN Yongyi;ZHANG Dan;YIN Wutao(College of Information Engineering,Zhejiang University of Technology,Hangzhou Zhejiang 310023,China;Wuxi bozhixin Technology Co.,Ltd,Wuxi Jiangsu 214029,China)

机构地区：[1]浙江工业大学信息工程学院,浙江杭州310023 [2]无锡博智芯科技有限公司,江苏无锡214029

出　　处：《传感技术学报》2022年第11期1499-1505,共7页Chinese Journal of Sensors and Actuators

基　　金：国家自然科学基金(61873237);国家重点研发计划(2018YFE0206900);产学研项目(KYY-HX-20180742,KYY-HX-20200842)。

摘　　要：血压是反映人体心血管系统状况重要信息的四个生命体征之一。随着科技的进步,智能化血压测量逐渐融入人类的日常生活当中。针对当下血压测量方法复杂,测量精度不高等问题,提出了一种基于嵌入SE注意力机制的时域卷积网络(TCN-SE)模型。该网络有效解决了现有方法中模型过拟合的问题,且进一步加强了模型对于不同通道信息的关注度。在保证信息完整的前提下,该模型可有效增大感受野。在重症监护中的多参数智能监测(MIMIC-II)数据集进行实验测试,通过计算均方误差和平均绝对误差等指标,得出收缩压的误差为(5.09±7.04)mmHg,舒张压的误差为(2.96±4.23)mmHg,表明所提出的方法相比于现有方法误差损失更低,在血压测量领域具有广阔的应用前景。Blood pressure is one of the four vital signs that reflect important information about the state of the body’s cardiovascular system.With the progress of science and technology,intelligent blood pressure measurement gradually integrats into people’s daily life.Focusing on the fact that the blood pressure measurement method is complex and the accuracy of measurement is low,an SE attention mechanism embedded temporal convolutional network(TCN-SE)model is proposed.This network effectively solves the problem of model over-fitting in existing methods,and further strengthens the model’s attention on different channel information.It also effectively increases the receptive field on the premise of ensuring the integrity of information.Experimental tests are carried out on MIMIC-II data set.By calculating the mean square error and mean absolute error,the error of systolic blood pressure is(5.09±7.04)mmHg,and the error of diastolic blood pressure was(2.96±4.23)mmHg,which indicates that the proposed method has a lower error loss than those of existing methods and thus has a broad application prospect in the field of blood pressure measurement.

关 键 词：血压 深度学习 时域卷积网络 SE注意力机制 光电容积描记技术 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]