Research On Non-invasive Continuous Blood Pressure Monitoring Models Cluster Based On Pulse Wave

Blood pressure is an extremely important hemodynamic parameter.An accurate and real-time blood pressure value is greatly important for the diagnosis,prevention,and treatment of hypertension-related diseases.However,blood pressure is susceptible to physiological and environmental factors.The results of single or intermittent blood pressure measurement vary widely.The pulse wave contains abundant blood pressure information,and the convenience and non-invasivity of its measurement make it ideal for noninvasive continuous monitoring of blood pressure.Based on the combined pulse wave and electrocardiogram signals,this study aimed to extract the waveform information,introduce individual characteristics,and construct systolic and diastolic blood pressure estimation models using the error back-propagation(BP)neural network.In the process of models construction,correlation analysis and the mean impact value(MIV)method were employed to reduce feature redundancy.In order to realize the self-correction of the models,the self-organizing feature mapping(SOFM)neural network was used to complete the classification of approximate attribute samples,and the blood pressure models based on the BP neural network were constructed according to the category to form the blood pressure monitoring models cluster.Moreover,the BP neural network was optimized by multiple population genetic algorithm(MPGA)to determine its initial weights and thresholds of the network.Finally,the personalized parameters were optimized by MPGA to obtain the final individual continuous blood pressure monitoring model.The results showed that the predicted values of the models in this paper were highly correlated with the measured values of the electronic sphygmomanometer,and the estimated error of the models met the Association for the Advancement of Medical Instrumentation criteria and the Grade A British Hypertension Society criteria.The models in this paper were expected to be applied to the long-term non-invasive continuous blood pressure monitoring equipment.For this paper,the main research work and achievements include:(1)In order to satisfy the construction and verification of the models,the experiments were designed to collect blood pressure data in this paper.In order to increase the fluctuation of blood pressure data,static and dynamic experiments were set up.In the process of collecting long-term blood pressure data,the fluctuation of human blood pressure was reasonably used.(2)For the collected ECG and pulse wave,the corresponding denoising algorithms and feature point recognition algorithms were proposed.(3)The MPGA-MIV-BP blood pressure monitoring models cluster was constructed.Correlation analysis and mean influence value method were used to screen feature parameters.Experimental samples were classified by self-organizing feature mapping neural network.The corresponding models were established according to sample categories based on BP neural network to form the blood pressure monitoring models cluster.And finally personalized parameters are optimized to determine the final individual blood pressure monitoring model.In the process of models construction,MPGA was used to optimize network parameters,as well as personalized parameters.(4)The performance of the MPGA-MIV-BP blood pressure monitoring models cluster was evaluated.To ensure objectivity,based on the same data samples,the BP neural network model based on PWTT,and the partial least squares regression model were introduced as comparison models.Through comparative analysis,the MPGA-MIV-BP blood pressure monitoring models cluster performed well during the long period of blood pressure monitoring.