Study On The Data-driven Noninvasive Continuous Blood Pressure Measurement Method

In order to address the global health problem that caused by the hypertension and cardiovascular disease(CVD),the diagnosis and treatment of hypertension has gradually changed from traditional passive drug therapy to disease prevention,early detection and blood pressure management.These changes require the support of a simple and convenient noninvasive continuous blood pressure measurement(NCBPM)method,while traditional blood pressure measurement techniques and instruments can not satisfy this requirement.Therefore,in order to control the CVD and guarantee the public health,it is of great significant to develop the NCBPM technology.This paper focused on the data-driven research of new NCBPM technology,and conducted a systematic study on the analysis of blood pressure measurement model,the construction of blood pressure measurement model and the transfer of blood pressure measurement model.The following is the main contents:1.In the analysis of blood pressure measurement model,the arterial baroreflex(ABR)neuromodulation mechanism was introduced for the first time to analyze the failure of the existing continuous blood pressure measurement model.An adaptive variational mode decomposition(AVMD)algorithm was proposed as a time-frequency analysis tool for the physiological signals.The algorithm can automatically decompose the non-stationary physiological signal into a set of sub-modal signals,which are independent of each other in the time-frequency domain and reflect the intrinsic characteristics of the original signal.The effects of ABR feedback neuromodulation mechanisms on existing continuous blood pressure measurement models were demonstrated by qualitative and quantitative experiments.These effects lead the failure of the existing continuous blood pressure measurement model in the measurement of low-frequency blood pressure changes,and is also one of the main reasons for the poor accuracy and stability of the existing continuous blood pressure measurement model.To improve the accuracy and stability of the continuous blood pressure measurement model,a blood pressure-related physiological signal characteristic under the ABR was proposed,which was named as heart rate reflectance(HRR).2.In the construction of blood pressure measurement model,employing the datadriven model construction method,a mapping relationship between 16 physiological signal features including HRR and blood pressure was constructed based on the gradient boosting regression tree(GBRT).This model enhanced the representational capacity of the continuous blood pressure measurement model through multiple physiological signal feature fusion,and improved the accuracy and stability of the blood pressure measure model by reducing the effects of ABR neuromodulation mechanism.3.The existing continuous blood pressure measurement model are individual variation,which should be separately calibrated for different objects and hard to be widely used.To address these problems,based on the theory of transfer learning,a twostage region transfer(TSRT)algorithm was proposed to train the model.The basic idea is to find the similar domains of physiological signal features for different individuals or different time periods,which can help to transfer the models between different individuals and different time periods,thus to simplify the calibration process and improve the applicability of the model.