摘要
目的:基于光电容积脉搏波描记法(PPG)和人工智能算法的智能血压监测设备预测血压的两种模式准确性比较。方法:使用智能血压监测设备(O2Ring指环)采集患者PPG信号后上传至深度卷积神经网络(DCNN)系统,用于提取患者PPG信号的自身特征,并构建回归网络连接其特征,通过不断采集临床数据和机器学习来优化算法。O2Ring指环预测血压有直接预测和标定预测两种算法模式,采集41例临床高血压患者446个时间点的血压数据,将两种模式的预测血压值与动态血压计实测的血压值进行比较,对比两种预测血压模式的准确性。结果:41例患者的收缩压O2Ring指环的直接预测值和标定预测值与动态血压计的实测值比较,差异无统计学意义(t=-0.237,t=1.738;P>0.05);舒张压的O2Ring指环的直接预测值与动态血压计的实测值比较,差异有统计学意义(t=-6.332,P<0.001),舒张压的O2Ring指环的标定预测值与动态血压计的实测值比较,差异无统计学意义(t=-1.371,P>0.05)。Pearson相关分析,收缩压的实测值与直接预测值和标定预测值均有显著相关性(r=0.5195,r=0.8828;P<0.001);舒张压的实测值与直接预测值和标定预测值亦都有显著相关性(r=0.4333,r=0.7544;P<0.001)。标定预测值与实测值的一致性亦好于直接预测。结论:基于PPG信号和人工智能算法的标定预测模式与直接预测模式相比,能够更准确预测血压,可为临床应用提供参考。
Objective:To compare the accuracies of two kinds of prediction models of artificial intelligence(AI)device of blood pressure monitoring in predicting blood pressure,which based on photoplethysmography(PPG)and artificial intelligence algorithm.Methods:Intelligent device of blood pressure monitoring,which was one kind of ring which name was O2Ring,was used to collect PPG signals of patients and upload them to the deep learning convolutional neural network(DCNN)system for extracting the self-characteristics of PPG signals of patients,and constructing a regression network to connect these characteristics.The algorithm was optimized through constantly collected clinical data and conducted machine learning.The O2Ring ring predicted blood pressure through two kinds of algorithms that included direct prediction and calibrated prediction.In this study,the blood pressure data of 41 patients with clinical hypertension were collected at 446 time points.The predicted values of blood pressure of the two kinds of models were compared with measured blood pressure values of dynamic sphygmomanometer,respectively,and the accuracies of the two kinds of prediction models were further compared.Results:There were no significant differences in systolic pressure of 41 patients between direct prediction and calibrated prediction of O2Ring ring and the actual measured value of dynamic sphygmomanometer(t=-0.237,t=1.738,P>0.05).There was statistically significant difference in diastolic pressure between the directly predictive values of O2Ring ring and the measured values of the dynamic sphygmomanometer(t=-6.332,P<0.001),and there was no statistically significant difference in that between the calibrated predictive values of O2Ring ring and the measured values of the dynamic sphygmomanometer(t=-1.371,P>0.05).The results of Pearson correlation analysis showed that the measured values of systolic pressure were significantly correlated with the directly predictive values,and with the calibrated predictive values(r=0.5195,r=0.8828,P<0.001),and the measured values of diastolic pressure were also significantly correlated with the directly predictive values,and with the calibrated predictive values(r=0.4333,r=0.7544,P<0.001).The results of consistency analysis indicated that the consistency between calibrated predictive value and measured value(Bland-Altman method)was also better than that between directly predictive value and measured value.Conclusion:Based on photoplethysmography signal and AI algorithm,the calibrated prediction model can more accurately predict blood pressure than directly prediction model.And it can provide references for clinical application.
作者
吴燕
汪奇
赛晓勇
汪晶晶
史纯纯
曹君
李瑞莱
张洪盼
韩宝石
WU Yan;WANG Qi;SAI Xiao-yong(Department of Cardiology,The Sixth Medical Center of Chinese PLA General Hospital,Beijing 100037,China;不详)
出处
《中国医学装备》
2022年第5期34-38,共5页
China Medical Equipment
基金
国家重点研发计划(2020YFC1512305)“后送舱系统性能验证及综合示范应用”
军队重点课题(2021-JCJQ-JJ-0528)“新冠疫情城市传播与预测模型构建方法与应激防控策略”
解放军总医院军事医学科研创新转化项目(CX19014)“危重伤员重要脏器功能的快速检测与高级生命支持系统研究”。
