Non-invasive detection of total bilirubin based on multi-wavelength PPG signal

• Published in: Computer methods and programs in biomedicine Vol. 236; p. 107570
• Main Authors: Cheng, Leiyang, Li, Gang, Nawaz, Muhammad Zeehan, Fan, Meiling, Lin, Ling
• Format: Journal Article
• Language: English
• Published: Ireland Elsevier B.V 01.06.2023
• Subjects: Bilirubin; Fingers; Humans; Mass Screening; Modeling dataset partitioning; Non-invasive detection of bilirubin; Non-invasive extraction of blood absorption spectrum; Photoplethysmography - methods; Sample PPG deviation screening; Spectrum Analysis - methods; Non-invasive extraction of blood absorption spectrum; Sample PPG deviation screening; Modeling dataset partitioning; Non-invasive detection of bilirubin
• ISSN: 0169-2607; 1872-7565; 1872-7565
• DOI: 10.1016/j.cmpb.2023.107570

•Noninvasive detection of human total bilirubin concentration based on multi-wavelength PPG signal.•Propose a method to screen sample spectra based on the degree of PPG deviation.•Propose a multi-component concentration distribution selection sample set method to achieve analysis in complex solutions.•The detection error of the final model bilirubin is only 9.3%. Abnormal bilirubin metabolism can result in various liver function disorders. Current clinical practice for bilirubin level detection involves invasive blood collection from patients, which is time-consuming, painful, and poses infection risks. Thus, there is a pressing need for non-invasive bilirubin detection methods. This study aims to develop a non-invasive total serum bilirubin(TSB) detection method in humans based on multi-wavelength photoplethysmography (PPG) signals. The experimental instrument includes a light source and a spectrometer. PPG signals are collected from the subjects' fingers, and the samples are selected based on the PPG deviation degree screening method. The absorption spectrum of blood is extracted from the PPG signal using dynamic spectroscopy. Finally, locally developed software calculates the total bilirubin value. The instrument is modeled and validated according to the clinical-biochemical test values. The results of the prediction set (correlation coefficient is 0.91, RSMEP is 2.32 umol/L, average absolute error percentage is 9.3%) show that our method has a strong correlation with the detection results of clinical-biochemical analysis instruments. The Bland-Altman test showed that the device deviated from the data detected by biochemical methods in the clinic with a mean deviation of about 0.12 umol/L and a 95% confidence interval between -2.95 umol/L and 2.7 umol/L. This study's non-invasive bilirubin detection method has high accuracy, which can meet the needs of continuous non-invasive total bilirubin detection in clinical practice.