【摘要】 对血压进行连续监测，能够有效预防心血管疾病，袖带式血压测量无法对血压进行连续的测量。提出一种基于深度学习的无袖带血压测量模型，使用卷积神经网络(Convolution neural network, CNN)和Croos former网络搭建无袖带血压测量模型，利用脉搏波信号(Photoplethysmograph, PPG)和心电信号(Electrocardiosignal, ECG)作为模型的输入，用于测量血压。该模型能够捕捉生理信号跨维度和跨时间的依赖关系，以此准确、连续的测量血压。模型使用BP-Net数据集进行训练、验证与测试，其中预测舒张压(Diastelic blood pressure, DBP)的平均绝对误差(Mean absolute error, MAE)为1.12 mmHg,标准差(Standard deviation, STD)为1.56 mmHg;收缩压(Systolic blood pressure, SBP)的MAE为2.02 mmHg, STD为2.79 mmHg。该模型的预测结果达到了美国医疗仪器促进协会(AAMI)的标准和英国高血压协会(BHS)的A级水平。更多还原

【Abstract】 Continuous monitoring of blood pressure can effectively prevent cardiovascular diseases. However, cuff-based blood pressure measurement is incapable of conducting continuous measurements. Therefore, a deep learning-based cuffless blood pressure measurement method is proposed, constructing a blood pressure measurement model using convolutional neural network and crossformer networks, with photoplethysmogram(PPG) and electrocardiogram(ECG) signals as model inputs for blood pressure estimation. This model can capture the cross-dimensional and cross-temporal dependencies of physiological signals, thereby enabling accurate and continuous blood pressure measurement. The model was trained, validated, and tested on the BP-Net dataset. The average absolute error(MAE) for predicting diastolic blood pressure(DBP) was 1.12 mmHg, with a standard deviation(STD) of 1.56 mmHg; the MAE for systolic blood pressure(SBP) was 2.02 mmHg, and the(STD) was 2.79 mmHg. The prediction results of this model have reached the standards of the Association for the Advancement of Medical Instrumentation(AAMI) in the United States and level A of the British Hypertension Society(BHS).更多还原