基于深度学习的ECG/PPG血压测量方法

ECG/PPG blood pressure measurement method based on deep learning

近年来,基于ECG/PPG信号的血压测量方法已经在某些可穿戴设备上实现。但此类方法的检测精度尚未达到相关国际标准。本研究利用深度神经网络模型,对基于ECG/PPG信号的血压测量方法进行了深入研究,提高了该类方法的检测精度。首先,采用基于小波包的模态分解技术,从PPG信号中提取出心脏信号和呼吸信号,并将其与ECG信号同步。然后,采用卷积神经网络(convolutional neural network,CNN)基于上述信号建立血压检测模型。通过选用从MIMIC-Ⅲ数据集中筛选出的5776条数据作为实验数据,结果显示,当使用ECG/呼吸/心脏信号测量血压时,CNN模型的收缩压检测精度为(4.6852±6.0730)mmHg,舒张压的检测精度为(2.5340±3.9860)mmHg,均达到美国医疗器械促进协会(AAMI)标准和英国高血压协会(BHS)标准的最高级。当使用呼吸/心脏信号测量血压时,CNN模型的舒张压检测精度达到AAMI标准和BHS标准的最高级,收缩压检测精度未达到AAMI标准。结果表明,模态分解技术与ECG信号结合后,可以有效提高对血压的检测精度。

In recent years,blood pressure measurement methods based on ECG and PPG signals have been implemented on wearable devices.However,the accuracy has not yet reached the relevant international standard.We used deep neural network model to conduct in-depth research on the blood pressure measurement based on ECG and PPG signals,which improved the detection accuracy.First,the mode decomposition technique based on wavelet packet was used to extract the heart and the respiratory signals from the PPG signal,and synchronize them with the ECG signals.Then,a convolutional neural network(CNN)was constructed to establish a blood pressure detection model based on the above-mentioned signals.By selecting 5776 pieces of data from the MIMIC-Ⅲdataset,the results showed that when ECG/respiration/heart signals were used to measure blood pressure,the systolic blood pressure detection accuracy was(4.6852±6.0730)mmHg,and the diastolic blood pressure detection accuracy was(2.5340±3.9860)mmHg,both of them reached the AAMI standard and the highest level of BHS standard.When using respiration/heart signal to measure blood pressure,the detection accuracy of the diastolic blood pressure of the CNN model reached the AAMI standard and the highest level of BHS standard,and the detection accuracy of the systolic blood pressure did not meet the AAMI standard.The results show that the mode decomposition technique combined with ECG signals can effectively improve the accuracy of blood pressure measuring.