文章摘要

文章信息/Info

Title:: The research of blood pressure measurement based on mixed feature convolution neural network

作者:: 张佳骕1△; 顾林跃1; 姜少燕2; 1.浙江好络维医疗技术有限公司，浙江杭州 310012；2. 青岛大学附属心血管病医院，山东青岛 266071

Author(s):: ZHANG Jiasu1; GU Linyue1; JIANG Shaoyan2; 1.Zhejiang Helowin Medical Technology Co. Ltd, Hangzhou 310012, China；2. The Affiliated Hospital of Qingdao University, Qingdao 266071, China

Keywords:: Mixed feature; Pulse transit time (PTT); One-dimensional convolution neural network (1D-CNN); Pulse waveform feature extraction; Pulse wave decomposition

摘要:: 传统使用脉搏波测量血压的方法存在准确率较低和特征点难以提取的问题。针对这些问题，本研究首先提出脉搏波分解算法，将脉搏波分解为三个钟型波复合的形式，由此获取到准确的脉搏波传播时间；之后提出混合特征卷积神经网络模型ABP-net，该模型将脉搏波传播时间特征和使用一维卷积自动提取的脉搏波波形特征相结合对动脉血压进行预测。最后使用ABP-net对MIMIC III中15个患者的血压进行预测。实验结果表明，ABP-net能够有效地提取脉搏波波形特征而且对血压的预测精度更高。

Abstract:: The traditional method of using pulse wave to measure blood pressure has the problem of low accuracy and difficult extraction of feature points. For the above problems, we firstly proposed pulse wave decomposition algorithm (PWDA), which decomposed the pulse wave into three bell-wave complexes and derived the accurate pulse transit time. Then mixed feature convolution neural network ABP-net was also proposed. The model predicted arterial blood pressure by combining the pulse transit time with the waveform features automatically extracted using one-dimensional convolution. Finally, ABP-net was used to predict blood pressure in 15 patients in MIMIC III. The experimental results show that ABP-net can effectively extract the pulse waveform feature and has higher prediction accuracy.