Study On Multi-index Analysis Method Of Serum Bilirubin With Vis-NIR Spectroscopy

Serum bilirubin includes three parameters:indirect,direct,and total bilirubin?IBil,DBil,TBil?.They are important clinical indicators of hepatobiliary disease and have clinical reference value for diabetes and cardiovascular disease.Exsiting clinical analysis methods require sample preparation,biochemical reagents,and professional laboratory operations,which are tedious and time-consuming,and are not suitable for the rapid detection requirements of the diagnosis and treatment process.Visible-near-infrared?Vis-NIR?spectroscopy is an indirect analysis technique.Calibration can be performed using known samples'spectra and clinical actual values of indicators to achieve rapid reagent-free analysis of unknown sample indicators.Although IBil and DBil are both low levels products of the hepatobiliary metabolic pathway,they are structural correlation with several macromolecules?e.g.hemoglobin,albumin?,and the latters have significant Vis-NIR absorption.Therefore,serum Vis-NIR spectroscopy has the possibility of indirect quantitative analysis of IBil and DBil.This is the main research goal of this paper.At the same time,the research and integration of new spectral preprocessing and wavelength model optimization methods,which are carried out to improve the accuracy of serum Vis-NIR spectra analysis.The main content and results are as follows:1.Methodological research:1)Based on the partial least squares?PLS?method,a calibration-prediction multi-division modeling system was established,and parameters were optimized based on the multi-division comprehensive prediction error?SEP⁺?,that was to avoid data over-fitting,and ensure parameter stability.Independent validation used samples that did not participate in modeling made the results objective.2)Established a spectral preprocessing algorithm platform based on Norris derivative filter to overcome noise interference such as spectral baseline drift and tilt,and achieve large range of optimization for the parameters?derivative order d,number of smoothing points s,and number of differential gaps g?of moving average smoothing and differential derivation.3)Established a wavelength model optimization based on equidistant combination PLS?EC-PLS?.The initial wavelength?I?,number of wavelengths?N?,and number of wavelength gaps?G?were used to achieve a large range of optimization of the equidistant combination wavelength model.This method takes the advantages of both continuous and discrete wavelength models into account,and strictly properly included the classical moving window PLS?MW-PLS?method,it is a strict extension from an algorithm point of view.4)Further integrated the wavelength step-by-step phase-out PLS?WSP-PLS?method to realize the secondary optimization of the wavelength model.2.Modelling and validating of simultaneous analysis of serum bilirubin indicators?IBil,DBil and TBil?of Vis-NIR spectroscopy:1)Respectively established calibration prediction models for IBil and DBil,and determined the optimal Norris parameters?d,s,g?of IBil and DBil were?2,15,9?and?2,13,9?.Based on the Norris derivative spectra,the EC-PLS method was used to optimize the wavelength model,and the WSP-PLS method further was uesd for the secondary optimisation of the top 10 equidistant combination wavelength models in the front row to determine the optimal EC-WSP-PLS model.For IBil and DBil,the optimal wavelength combinations were 11 and 18 wavelengths,respectively,which located at the visible-near infrared octave region.After independent validation,for IBil and DBil,the predicted root mean square error?SEP?of the predicted values and the clinical actual values were 0.90 and 0.71(?mol L^-1),respectively;the predicted correlation coefficients?R_P?reached 0.975 and 0.955,respectively;the ratios of performance to deviation?RPD?were 4.4 and 3.3,respectively.The results showed high correlation and low error between predicted and measured values.2)Total bilirubin TBil used the independent spectral modelling and summation methods?TBil=IBil+DBil?.The results showed that the latter achieved significantly better prediction effects,and the validated SEP,R_P and RPD were 0.82?mol L^-1,0.990 and 7.1,respectively.Therefore,the summation of IBil and DBil spectral prediction values was more in accordance with the mechanism of TBil analysis.3)The Bland-Altman analysis was used to evaluate the consistency of the spectral predicted and clinical actual values of IBil,DBil and TBil.For the three indicators,95.0%,95.0%and 96.3%of the points,respectively,were within the agreement limit.The results indicated small random errors of NIR analysis method,and the prediction effect was good.The results confirmed the feasibility of the simultaneous analysis of serum bilirubin multiple indicators through reagent-free Vis-NIR spectroscopy.The technology is simple and novel,and has the potential for application in the rapid detection of health screening,diagnosis and treatment processes.The proposed multi-parameter integrated optimization method has novelty and application value.The relevant wavelength model can provide a reference for special instrument design and clinical applications.