Near-infrared Discrete Combination Model And Its Application For Hemoglobin Analysis

The hemoglobin?Hb?content is an important and common clinical indicator.It is one of the most common clinical indicators,and plays a major role in the diagnosis of anemic diseases and also has reference values in the diagnosis of other important diseases.It is one of diagnostic indicators of anemia diseases such as thalassemia and iron deficiency anemia,and it also has important reference values in the diagnosis of other related diseases.During surgery or recovery procedures,real-time,non-invasive hemoglobin monitoring is often required,but current clinical assays cannot meet the needs.The NIR spectrum has been used in vivo or in vitro hemoglobin analyses,and it has elicited widespread attention due to its green?non-invasive or reagent-less?detection methods At the same time,there are also methodological challenges.Due to reason that the weak target signals,many interference factors,and low detection signal to noise ratio,the current detection accuracy is far from the level of clinical application.In terms of high-precision analysis of the whole blood sample and wavelength model optimization,the basic research still needs to be carried out deeply.The moving-window partial least squares?MW-PLS?method achieve a global optimization of the continuous wavelength model using two variable parameters.Our team adopted three variable parameters,proposed an Equidistant combination-partial least squares?EC-PLS?method and achieved global optimization of continuous and quasi-continuous wavelength models.It covers MW-PLS in the algorithm.In order to eliminated the redundant wavelengths in the equidistant combination model,this study further proposed a promotion:the cluster analysis is performed on the equidistant combination model;by changing the number of allowable models that floated and the number of wavelengths of the frequency priority combination,then the discrete combination partial least squares?DC-PLS?method was proposed and achieved a wide range preference of discrete wavelength models.In this study,the blood routine samples and thalassaemia screening samples were used to perform hemoglobin analysis experiments,and the proposed DC-PLS method was tested.The EC-PLS method was compared with the other two discrete methods to examine the predictive power of the new method,and research methods for near-infrared hemoglobin analysis was advanced.The two discrete methods are successive projections algorithm PLS?SPA-PLS?and competitive adaptive reweighted sanoling PLS?CARS-PLS?methods.The main research is as follows:?1?Method strategy of DC-PLS model:1)achieve a wide range of optimization of the EC-PLS model by using the starting point wavelength,the number of wavelengths,and the number of wavelength intervals as the variation parameters.2)The combination of equidistant wavelength is sorted according to the prediction effect,and the model which is in front and equivalent to the optimal model is taken as the allowable model.3)The wavelengths of the multiple allowed models are collected,ranked according to the frequency and combined preferentially.Then the first round of discrete model optimization is achieved by varying the combined number of wavelengths.4)Finally,the number of allowed models is traversed,and the latter round of discrete wavelength model optimization is obtained.?2?Blood routine experiment:The optimal DC-PLS model only contains 42wavelengths?N=42?.Independent validation was performed using the samples which were not involved in modeling.The predicted root-mean-square errors?RMSEP?of the NIR-predicted and clinically measured Hb values was 2.86 g?L^-1,and the correlation coefficient?R_P?,the relative root-mean-square errors?RRMSEP?and the ratio of standard error of performance to standard deviation?RPD?were 0.983,2.3%,and 5.3,respectively.The prediction effects were obviously better than the EC-PLS,SPA-PLS and CARS-PLS models which were used for comparison.?3?Thalassemia screening experiment:the selected optimal DC-PLS model contained50 wavelengths?N=50?.The RMSEP,R_P,RRMSEP,and RPD values of the validation samples were 3.20 g?L^-1,0.989,2.8%,and 6.6,respectively.They were also superior to the EC-PLS,SPA-PLS,and CARS-PLS models significantly.The results shown that the proposed DC-PLS method was successfully applied for the quantitative analysis of hemoglobin in human peripheral blood.In the view of the methodological point,the promotion from continuous traversal to discrete traversal is realized here.It improves the spectral prediction performance and reduces the model complexity by overcoming the redundant wavelength data that must be included in the equal interval model and making the model more anti-jamming capability.The approach of near-infrared spectroscopy for the analysis of complex samples is expanded.It is worth noting that the relative root-mean-square errors for the NIR-predicted and clinically measured Hb values?RRMSEP?in both samples are less than 3%,which is expected for clinical application.The proposed wavelength model can be utilized to design dedicated spectrometers for Hb and provides a valuable reference for further non-invasive Hb detection.