基于多参数融合的高分辨率连续血压测量方法研究

Continuous blood pressure measuring plays as an important role in early prevention and diagnosis of acute cardiovascular event. However, the impacts of individuality and time-varying factors in the current continuous blood pressure measuring method based on pulse wave transmit theory is hard to overcome. Thus the accuracy of measurement cannot meet clinical demands. The time variation of heart Pre-Ejection Period (PEP) is one of the typical factors. This project aims to achieve high resolution cuffless continuous blood pressure measurement through the integration of physiological mechanism analysis and statistical machine learning based on high quality clinical experiment. First, through synchronous acquisition, matching, and signal pattern analysis for color ultrasonic imaging of heart and surface multi-physiological signal, the accurate PEP measurement based on surface physiological signal can be achieved. Second, by conducting synchronous measuring of invasive continuous blood pressure, individual cardiovascular system status, and surface physiological signal under strictly clinical-controlled environment, data mining technology will be used to analyze the association among the three kinds of signals. Thus to illustrate the impact of individual and time-varying difference on continuous blood measuring model and its medical mechanism. And then, by integrating the rules above into the multi-feature, multi-model, and nonlinear integrated machine learning modeling process, and into the self-adapting calibration process, the model error caused by individual and time-varying difference can be eliminated by direct surface physiological signal measuring. Thus the high-precision cuffless blood pressure measurement with group universality and individual adaptability can be realized.

连续血压测量对心血管急性事件预防具有重要意义。现有基于脉搏波传导理论的连续血压测量方法难以克服个体性、时变性因素的影响而导致测量精确度难以达到临床要求。心脏射血前期时间的变异是上述因素的典型例子之一。本项目基于严格临床实验控制条件下获取的高质量实测数据，通过结合生理机制分析和统计机器学习，实现高分辨率的无袖带连续血压测量。首先，通过心脏彩超图像和体表多生理信号的同步采集、匹配及模式分析，实现基于体表信号的心脏射血前期的准确测量。进而运用数据挖掘技术分析有创连续血压、个体心血管系统生理状态、体表生理信号三者的同步关联规律，阐明心血管系统状态的个体差异、时变差异对连续血压测量模型造成的影响及其医学机制，将上述规律融入到多特征、多模型非线性的集成机器学习建模和自适应校准过程，达到通过体表生理参数直接测量消除个体性、时变性模型误差的目的，实现具有群体普适性和个体适应性的高精准无袖带连续血压测量。

研究高精准的连续血压测量方法和技术，对深入探究心血管疾病的病理机制和内在规律，对患者及特殊作业人群进行更全面的生命体征监护，以及更好的预防心血管危重疾病，都具有非常重要的理论研讨意义和临床应用价值。本项目开展了基于体表生理信号的连续血压建模理论研究。首先，通过大数据手段进行基于体表生理信号的血压关联特征挖掘，从而建立了体表生理信号与连续血压的关联关系，并根据Moens–Korteweg血流动力学方程及每搏输出量与脉搏波传导强度之间的正比关系，提出了三种新型血压强关联因子，克服了传统基于单一因素的血压模型精度及适用性受限的难题。针对血流动力学不稳定期人群（如心律失常患者），提出了一种时空特征融合的深度网络架构（ResNet+LSTM）用于连续血压测量，实现了心律失常下基于心电信号与脉搏波信号的逐拍血压准确测量。为提高血压模型的普适性，通过深度挖掘的可穿戴监测生理信号与血压变异之间的关联规律，提出了血流动力学模型与生理数据模型耦合的无袖带血压测量理论及方法，建立了面向健康人群、高血压人群、低血压人群、心律失常人群的高普适性血压计算模型，精度满足AAMI标准。最后基于多生理信号数据，通过基于一维残差网络对血压波形的高斯拟合建模，实现基于体表生理信号的无创连续血压波形的精确重建。相关研究成果共发表15篇SCI论文和7篇国际会议论文，申请15件发明专利，提交1项标准提案。