摘要
目的 研究基于脉搏波和心电信号的无创连续血压预测方法。方法 从MIMIC-Ⅲ数据库中选取300个病例,用于构建血压预测模型、模型验证;另收集2022年1月至6月入住福建省立医院重症监护病房的121例患者,用于测试模型;采集患者动脉血压、光电容积脉搏波和心电图信号。构建两个血压预测模型,一个是以人工提取出的8种特征参数构建的人工特征参数模型,另一个是以8种特征参数加1种卷积神经网络提取的特征进行融合构建的特征融合模型。对两个预测模型进行验证、测试,评价指标采用平均绝对误差(MAE)、标准差(SD)、均方根误差(RMSE),根据国际公认的美国医疗器械促进协会(AAMI)规定的标准进行评价,对比两个模型预测能力。结果 用MIMIC-Ⅲ数据对两个模型进行评价,特征融合模型的MAE、SD符合AAMI标准,RMSE比人工特征参数模型低。用实际收集的重症患者数据对两个模型进行评价,特征融合模型收缩压的SD、舒张压的MAE和SD达到AAMI标准,RMSE也比人工特征参数模型低。结论 特征融合模型的预测能力比人工特征参数模型好。
Objective To study the non-invasive continuous blood pressure prediction method based on pulse wave and electrocardiosignal.Methods A total of 300 cases were selected from MIMIC-Ⅲ database to build blood prediction model and model verification.Meanwhile,121cases were collected which were hospitalized in the Intensive Care Unit of Fujian Provincial Hospital from January to June 2022 for test model.The arterial blood pressure,photoplethysmography,and electrocardiography signal of patients were collected.Two blood pressure prediction models were built.The first one was artificial feature parameter model that was built based on eight artificially collected feature parameters.One was feature fusion model that was fused and built based on the eight feature parameters and the other one feature collected from convolutional neural network.These two prediction models were verified and tested.The evaluation indexes applied mean absolute error(MAE),standard deviation(SD),and root mean square error(RMSE).Evaluation was proceeded according to the internationally recognized specified standard of(Association for the Advancement of Medical Instrumentation,AAMI) to compare the predictive ability of both models.Results MIMIC-Ⅲ data were applied to evaluate both models.The MAE and SD of feature fusion model were consistent with the standard of AAMI.RMSE was lower than it of artificial feature parameter model.The actual collected data of critical patients were applied to evaluate.The SD of systolic pressure,MAE,and SD of diastolic pressure of feature fusion model met the standard of AAMI.RMSE was also lower than that of artificial feature parameters.Conclusion The predictive ability of feature fusion model is better than artificial feature parameter models.
作者
张健春
王量弘
庄丽媛
张炜鑫
王新康
ZHANG Jianchun;WANG Lianghong;ZHUANG Liyuan;ZHANG Weixin;WANG Xinkang(Department of Electrocardiographic Diagnosis,Fujian Provincial Hospital,Fujian Province,Fuzhou350001,China;College of Physics and Information Engineering,Fuzhou University,Fujian Province,Fuzhou350108,China;Shengli Clinical Medical College,Fujian Medical University,Fujian Province,Fuzhou350001,China)
出处
《中国医药导报》
CAS
2024年第13期12-15,共4页
China Medical Herald
基金
国家自然科学基金资助项目(61971140)
福建省卫生教育联合攻关项目(2019-WJ-18)
福建医科大学启航基金项目(2020QH1187)。
关键词
光电容积脉搏波
心电图
融合特征
无创连续血压预测
可穿戴式血压设备
Photoplethysmography
Electrocardiogram
Fusion feature
Non-invasive continuous blood pressure prediction
Wearable blood pressure equipment
作者简介
张健春(1988.6-),女,硕士,研究方向:心电诊断、无创电生理、医工结合;通讯作者:王新康(1972.9-),男,硕士,主任医师,研究方向:心血管内科、无创电生理、医工结合。
