Research On Blood Pressure Measurement Of Multi-spectral Photoplethysmography Based On Deep Learning

According to the World Health Organization,there are currently at least 1.28 billion adults suffering from hypertension worldwide.Hypertension has become a common chronic disease,and its incidence is increasing year by year,becoming a major cause of premature death around the world.Currently,there are significant shortcomings in the diagnosis of hypertension.First,only severe symptoms such as headaches,blurred vision,and chest pain can cause a warning,and mild patients do not have symptoms,so most patients cannot timely detect changes in their own physical condition.Secondly,traditional blood pressure measurement techniques cannot achieve continuous monitoring,thus limiting their application in clinical and daily life.In recent years,the applications of multi-wavelength photoplethysmography(PPG)in noninvasive continuous blood pressure prediction has gradually become a research hotspot.Based on the PPG technology,imaging photoplethysmography(i PPG)greatly improves its accuracy and reliability in detecting physiological parameters such as heart rate and blood pressure by its multi-points measurement and ability to obtain more signal details.However,multi-spectral i PPG technology is still in the research stage,and there are many difficulties in blood pressure measurement,such as difficulties in signal measurement,poor signal quality,inability to determine wavelengths,and low prediction accuracy.In response to the various problems of multi-spectral i PPG technology in noninvasive blood pressure measurement,this paper conducted research on related issues of obtaining multi-spectral signals,image and signal processing,and blood pressure measurement,and proposes solutions and methods for some of the difficulties in the field.The specific researches is as follows:(1)Based on the physiological structure characteristics of the fingertip nail bed,a multi-spectral transmittance measurement device was designed and built.Designed the optical path for the system based on system indicators,analyzed and selected system components,and developed calibration algorithms tailored to system characteristics.Successfully obtained multi-spectral transmittance curves and i PPG signals within the570-970 nm wavelength range.(2)Proposed a multi-spectral signal weighted fusion algorithm to enhance signal quality by using correlation analysis techniques to perform weighted fusion on i PPG signals of different wavelengths.This algorithm can reduce signal noise and interference,improve signal quality and stability.After analyzing the main sources of i PPG signal noise,a cascade denoising method was used to eliminate singular mutation noise,high-frequency noise,and baseline drift noise.Optimization through multichannel fusion using correlation analysis increased final signal quality by 22.4%compared to the original full-band signal.(3)According to the Hilbert-Huang transform and convolutional neural networks(CNNs),a blood pressure level prediction method was proposed.The potential for PPG and its derivative signals in blood pressure prediction was verified by designing three two-classification experiments.Two self-made image datasets were used to validate the proposed method,which showed that the inclusion of first and second derivatives of PPG significantly improved blood pressure level prediction accuracy.(4)Constructed a blood pressure prediction model based on a one-dimensional convolutional neural network(1DCNN)and long short-term memory(LSTM).Using sequential signal data sets based on i PPG signals and their first and second derivatives,a 1DCNN network structure based on Alex Net was created.This was combined with a two-layer LSTM model to create the 1DCNN-LSTM network model.Blood pressure prediction was performed on optimized combinations of multi-spectral i PPG signals,achieving mean absolute error of 5.59 mm Hg for systolic blood pressure(SBP)and5.24 mm Hg for diastolic blood pressure(DBP).