基于U-net++的心冲击图信号自动标注及心拍估计方法研究  

Method of automatic annotation of BCG signal and heart beat estimation based on U-net++

作　　者：代宜辰 侯晓敏 廖小丽[1] 刘航 李科[1] DAI Yichen;HOU Xiaomin;LIAO Xiaoli;LIU Hang;LI Ke(School of Life Science and Technology,University of Electronic and Technology of China,Chengdu 610054,Sichuan Province,China;The First Veterans Hospital of Sichuan Province)

机构地区：[1]电子科技大学生命科学与技术学院,成都610054 [2]四川省第一退役军人医院

出　　处：《中国数字医学》2023年第6期36-41,共6页China Digital Medicine

基　　金：四川省科技支撑计划项目(22ZDYF0376,21ZDYF2028)。

摘　　要：心冲击图信号(BCG)是通过压力传感器获取由心脏搏动引起的人体重力变化信号,适用于长时间的心功能实时监测。心跳检测是BCG信号分析研究的重要组成部分。本研究设计并验证了利用同步ECG信号自动标注及心拍检测的原理,提出一种基于一维U-net++模型从BCG信号自动估计心拍位置的方法。该方法利用一维U-net++网络中卷积块稠密的跳跃连接,有效提取并结合BCG信号中的高阶和低阶信息进行心拍预测。实验采用公共数据集的20名被试同步采集的BCG和ECG数据,建立了从BCG信号到心跳位置序列的端到端网络。模型的正确率和精确率分别达到99.97%和99.34%。实验结果表明,该方法可有效从BCG信号中提取心拍位置,完成准确的心跳节律估计。Ballistocardiogram(BCG)is the signal of human body gravity changes caused by heart beat obtained by pressure sensors,which is suitable for long-term real-time monitoring of cardiac function.Heartbeat detection is an important part of BCG signal analysis.In this paper,the principle of automatic labeling and heart beat detection using concurrently recorded ECG signals are designed and verified.And a method of automatically estimating the heartbeat position from BCG signals based on one-dimensional U-net++model is proposed.This method makes use of the dense skip connections of convolutional blocks in the one-dimensional U-net++network,effectively extracts and integrates the high-order and low-order information of BCG signal for beat prediction.The BCG and ECG data collected synchronously from 20 subjects in a public dataset are used in our experiment,to establish an end-to-end network from BCG signal to heartbeat position sequence.The accuracy and precision of the model reached 99.97%and 99.34%respectively.The experimental results show that the proposed method can effectively extract the beat position from BCG signals and accurately estimate of heartbeat rhythm.

关 键 词：心冲击图信号 深度学习 心率提取 

分 类 号：R319[医药卫生—基础医学]