Development Of A Non-invasive Blood Pressure And Blood Glucose Monitoring System Based On Neural Network

Chronic diseases,such as hypertension and diabetes,are becoming a major threat to the health of our population because of their high incidence,low early diagnosis,and low control rate.For the prevention and treatment of chronic diseases,early detection and treatment are still one of the most effective means of prevention and treatment.Therefore,it is of great clinical significance to prevent the occurrence of major diseases through the rapid and accurate detection of blood pressure and blood glucose in a non-invasive manner.Photoplethysmography(PPG)generated by human blood circulation can be collected from the surface of the human body using photoelectric detection,and the PPG signal carries information on the state of each physiological system.In the study of PPG-based blood pressure and blood glucose non-invasive detection models,the core research content is divided into the following parts:1.pre-processing of PPG signals;2.characterization of PPG signals;3.construction of blood pressure and blood glucose detection models;4.deployment of detection models.To this end,the research work in this dissertations as follows:1.Preprocess the original PPG signal by wavelet threshold denoising,cubic spline interpolation,and cluster analysis to remove the interference of noise and baseline drift and select the optimal single-cycle waveform;2.Use the PFS feature analysis method proposed in this paper to analyze the combination of each feature parameter obtained on the PPG signal to obtain the optimal feature subset for modeling;3.The BP neural network is used to establish two blood pressure non-invasive detection models of systolic blood pressure and diastolic blood pressure,and the one-dimensional convolutional neural network(1DCNN)is used to establish a blood glucose non-invasive detection model;4.Based on Bootstrap+Django+MySQL architecture,the data cloud service platform of this paper was developed to realize the data management and model update of the noninvasive testing system of this paper;5.The experimental validation of the non-invasive testing system developed in this research showed that the mean absolute error of systolic blood pressure detection was 4.14 mmHg with a standard deviation of 7.42 mmHg;the mean absolute error of diastolic blood pressure was 3.62 mmHg with a standard deviation of 6.37 mmHg;the errors of blood pressure detection were all following the error standard of blood pressure measuring instruments set by AAMI.The accuracy of blood glucose detection was analyzed by Clark’s grid,and all the observed data points were located within the acceptable area,the proportion of points in area A was 83%,and the proportion of points in area B was 17%,and the blood glucose detection results met the error accuracy requirements of GB/T 19634-2005 blood glucose meter standard.