Non-Contact Blood Pressure Measurement Technology Based On Video

Most patients with hypertension have an insidious onset,progress slowly,and have no specific clinical manifestations.Regular follow-up monitoring of blood pressure can significantly reduce the risk of developing hypertension and related cardiovascular diseases.The cuffed electronic sphygmomanometer is the main device for household blood pressure measurement at present,but it cannot meet the demand for comfort,convenience and economy for daily monitoring of blood pressure.Using photoplethysmography(PPG)technology to collect surface pulse wave signals and combining with machine learning methods is the main way to solve the problem of blood pressure measurement.However,it requires special hardware acquisition devices,which increases the cost of users,and the measurement accuracy is susceptible to motion interference.The non-contact blood pressure measurement method based on Imaging Photoplethysmography(IPPG)technology is a new scheme proposed in recent years,which can solve the problems of existing blood pressure measurement methods.However,the current research has generally high requirements for imaging equipment and shooting environment,and the generalization ability of the model is poor in practical use.In this paper,we capture the skin color image of human palm area by mobile phone camera,acquire the pulse wave signal from it.Based on the deep learning technology,the dual-channel feature fusion(DCFF)blood pressure prediction model is established to realize the non-contact measurement of blood pressure.The main contents of this paper are as follows:(1)Compare the original pulse wave signal quality of RGB and YIQ color space under different lighting conditions.The Region of Interest(ROI)was analyzed,and the IPPG signal quality of the whole face,forehead,cheek and palm was compared.The signal quality of the palm area is much higher than that of the other three regions.(2)The denoising effects of three denoising methods,wavelet decomposition,empirical modal decomposition and ensemble empirical modal decomposition,on the original IPPG signal are compared respectively.The denoised pulse wave signal was subjected to singleperiod segmentation and waveform screening to complete the non-feature selection pulse wave pre-processing process.(3)A DCFF deep neural network blood pressure prediction model based on contact pulse wave signal is designed,which completes the automatic extraction of waveform features through deep feedforward network and deep convolution network.The prediction accuracy of the test set selected in the MIMIC1.0 database was 1.64±3.73 mm Hg for systolic blood pressure and 1.43±2.43 mm Hg for diastolic blood pressure.(4)Using IPPG data to transfer training on the basis of DCFF model,a blood pressure prediction model based on IPPG signal is obtained.The prediction accuracy on the self-built video blood pressure test set was 3.37±4.46 mm Hg for systolic blood pressure and 2.78±3.43 mm Hg for diastolic blood pressure.Compared with related work,the method in this paper can effectively improve the accuracy of non-contact blood pressure measurement,and the blood pressure measurement method is more suitable for practical application.The blood pressure measurement method designed in this paper only needs the mobile phone camera to obtain the video of the palm area to complete the accurate measurement of blood pressure.The measurement accuracy can meet the requirements of the American Association for the Advancement of Medical Instrumentation(AAMI)and reach the A level specified by the British Hypertension Society(BHS)standard.It has the potential to become a more convenient blood pressure measurement method.