Research On The Detection Method Of Hemoglobin Concentration Based On PPG Dual Wavelength

Nowadays,the concentration of human hemoglobin is determined by invasive methods.Although invasive methods can accurately detect the concentration of human hemoglobin,these methods are intrusive to human beings.The cost of these invasive methods is very high,which are unavailable for people in daily life.In this paper,based on Photoplethymography(PPG)and the revised Lambert Beer's law,a non-invasive hemoglobin concentration detection system is designed by using the dual-wavelength reflectance method,which can realize the non-invasive detection of hemoglobin.The system consists of a lower computer and a host computer.The lower computer includes a dual-wavelength PPG signal acquisition module,a STM32F103 system module,a serial communication module and an OLED display module.The host computer adopts the LABVIEW software to design relevant functional modules,which can save the collected data of the system and a hemoglobin concentration prediction algorithm.MAX30102 is a dual-wavelength PPG signal acquisition module.Firstly,the wavelet transform is used to eliminate baseline drift signals in pulse wave original signals.Then 22 characteristic parameters of PPG's time domain and 3 characteristic parameters of human body are extracted from the dual-wavelength PPG signal.Finally,10 key feature parameters are selected as the input values among the 25 characteristic parameters by using the RELIEF-PCA combination algorithm.The BP neural network prediction model was constructed by using these 10 key eigenvalues.The values of the Mission HB hemoglobin analyzer were used as comparison values.A total of 115 pieces of experimental data were collected.Among the experimental data,randomly,80 pieces of data were selected as the training set of the BP neural network,and 35 pieces of data were used as the test set.Compared with the detection value of the Mission HB hemoglobin analyzer and the experimental data of the system,the correlation coefficient is 0.805 and the mean square error was 0.71g/dL.It demonstrates that the hemoglobin prediction model based on BP neural network can effectively measure the hemoglobin concentration.This method provides an important approach for non-invasive measurement of hemoglobin concentration.It also provides a new way for remote monitoring of hemoglobin cloud services and mobile medical services.