Hemoglobin Measurement Based On Multi-Wavelength Photoplethysmography

Hemoglobin is one of the important physiological indicators of the human body,and hemoglobin testing is more predictive of the occurrence of anemia and other related diseases.Conscious hemoglobin measurement in daily life,early detection,and intervention of abnormal changes in hemoglobin concentration can reduce the risk of anemia and other related diseases.Therefore,this paper designs a non-invasive hemoglobin detection system based on multi-wavelength pulse wave,which extracts the characteristic parameters of multi-wavelength PPG signal by collecting photoplethysmography(PPG)from human fingertips,builds a machine learning-based hemoglobin algorithm model,and finally realizes non-invasive hemoglobin detection.The main elements are as follows:First,the hemoglobin detection algorithm is combined with specific practices to build a multi-wavelength noninvasive hemoglobin detection system,which is divided into two parts:hardware acquisition system and software display and storage system.The hardware part mainly contains five parts,including PPG signal acquisition sensor module and multi-mode sensor front-end chip module.The software part mainly includes five parts,such as ADPD4100 parameter setting function and real-time data plotting display function.Meanwhile,the hardware equipment was used to complete the PPG signal data acquisition of 58 experimental samples to provide reliable data for the subsequent hemoglobin study.Secondly,in order to improve the signal utilization and obtain the high-quality PPG signal,for the noise of PPG signal,Butterworth band-pass filter and type II Chebyshev band-pass filter were used to filter out the high frequency noise and baseline drift in the PPG signal to finally obtain the high-quality PPG signal.The morphological and time-domain feature parameters of the PPG signal at four wavelengths extracted from the high-quality signal were filtered by using three feature selection methods,namely Info Gain,Chi Square and Relief F,for the 160 obtained features,and finally 30 features highly correlated with hemoglobin were obtained.And based on this,the XGBoost-based regression hemoglobin prediction model was built and compared with other machine learning models for analysis.Finally,the experimental results show that the MAE result of the test set in the XGBoost model for hemoglobin concentration is 0.325 g/L,which is less error and better than the other two algorithm models,with R~2 result of 0.997 and RMSE result of 0.762 g/L.Then the consistency analysis is performed by Bland-Altman plot,and the experimental results show that most of the points are in the 95%The results showed that most of the points were within the 95%agreement interval,and the 95%agreement limit was(-1.504,1.486)g/L,indicating the validity and practical applicability of the non-invasive hemoglobin detection system based on multi-wavelength PPG signal for hemoglobin detection proposed in this paper.In this paper,the mechanism of noninvasive hemoglobin detection is thoroughly studied and discussed,and the constructed noninvasive hemoglobin detection system based on multi-wavelength PPG signal and the machine learning algorithm model provide new ideas and effective methods for the assessment of noninvasive hemoglobin level.