基于卷积神经网络的单路PPG信号的连续动脉血压测量算法

CONTINUOUS ARTERIAL BLOOD PRESSURE MEASUREMENT ALGORITHM BASED ON A SINGLE PPG SIGNAL OF CONVOLUTIONAL NEURAL NETWORK

利用脉搏波特征参数法估测动脉血压(ABP)时,需要进行人工特征设计和筛选以提取出与动脉血压相关性高的特征参数,具有局限性。对此,提出基于卷积神经网络(CNN)的单路光电容积脉搏波信号(PPG)的连续动脉血压测量算法。对PPG信号依次进行去噪平滑、单周期分割和插值处理后,将其输入到CNN中进行特征提取,以建立其与动脉血压间的关系。根据美国医疗器械促进协会(AAMI)的规定,医疗用途的血压测量器械的平均误差应不大于5±8 mmHg。实验结果表明,该算法的测量误差满足AAMI要求。

When using pulse wave characteristic parameter method to estimate arterial blood pressure(ABP),artificial features design and filter are needed to extract characteristic parameters with high correlation with ABP,which has limitations.In view of this,a continuous arterial blood pressure measurement algorithm based on the single-way photoelectric volume pulse wave signal(PPG)of convolutional neural network(CNN)is proposed.After denoising,smoothing,single period segmentation and interpolation,the PPG signal was input into CNN for feature extraction to establish the relationship between PPG signal and arterial blood pressure.According to the American medical device promotion association(AAMI),the average error of blood pressure measurement devices for medical use should not be greater than 5±8 mmHg.The experimental results show that the measurement error of this algorithm meets the requirement made by AAMI.