The Research Of The Correlation Between BP And Multiple Physiological Parameters And Non-invasive Continuous BP Measurement Models

Blood pressure,including systolic blood pressure and diastolic blood pressure,is an important physiological parameter of the human body.In addition to its high and low levels,its dynamic changes are also a vital basis for the evaluation of the status of heart and vascular functions.And it has been demonstrated that blood pressure variability is closely related to cardiovascular diseases.As a result,monitoring and management of blood pressure is of great siginificance for clinical diagnosis and individual prevention of hypertension.Currently,the existing cuff based blood pressure measurement devices can only obtain instantaneous blood pressure intermittently,but can not get dynamic values,while the cuff-less continuous blood pressure measurement methods are commonly built based on one or more of the signals,called Electrocardiogram(ECG),Photoplethysmographic(PPG)and Ballistogram(BCG).However,there are still some problems,during which is the most important one that the accuracy is not high enough.Therefore,this study explores the correlation between blood pressure and multiple physiological parameters and studies the prediction model based on the characteristics,which is closely related to blood pressure.Following is the carried out aspects of this paper:1)Research on the correlation between multiple physiological parameters and blood pressure.ECG,PPG,Impedance Cardiography(ICG),Respiratory and continuous blood pressure signals from 55 healthy volunteers under five different conditions were collected simultaneously.Firstly,the signals were processed,including characteristic points detection and segmentation.Then,features that can reflect pulse wave velocity,energy and stability of the signals were extracted.Consequently,six methods were used to analyze their correlation between blood pressure and the features were sorted to find out the factors closely related to blood pressure.2)Research on the prediction models in single state.Four methods including Linear Regression,Support Vector Regression,Gradient Boosting Regression and Random Forest Regression were used to establish single state blood pressure prediction models.Firstly,the performance of the models built based on different types of features were compared and it was proved that the newly proposed features in this paper were effective.Consequently,Aiming at the drawback of large number of attributes,less important ones were removed according to the results of feature ranking.Then,Random Forest Rgression combined with Genetic Algorithm was adopted to obtain the optimal feature set establishing prediction model.Finally,analyzing the feature set acquired to obtain the characteristic variables that contribute more to blood pressure prediction.3)Research on the prediction models in united state.Considering blood pressure monitoring in daily life,unified models were established for the five states.Similar to the study in single state,Support Vector Regression and Random Forest Regression were used to establish prediction models.And Mean Absolute Error(MAE),Standard Error(SDE),Root Mean Square Error(RMSE)and R2(R-squared)were utilized to evaluate the models.As a result,we got MAE=2.97,SDE=4.14,RMSE=5.10,R2=0.95 for systolic blood pressure and MAE=2.02,SDE=2.37,RMSE=3.15,R2=0.96 for diastolic blood pressure on the test set.The results show that the performance of the model meets the standards of American Association for the Advancement of Medical Devices(AAMI).This work can provide reference for the use of cuff-less continuous blood pressure measurement method in daily life,and further promote the development of wearable continuous blood pressure measurement devices.