Grouping Modeling Method For Blood HCT Analysis With Vis-NIR Spectroscopy

Hematocrit（HCT）refers to the volume ratio of red blood cells in a certain volume of whole blood,and is often used for the diagnosis and classification of anemia.HCT is related to blood viscosity and has reference value for cardiovascular and cerebrovascular diseases.It is one of the important indicators included in blood routine testing.Existing detection methods are complicated to operate,which are not suitable for large population screening and real-time monitoring during surgery.In recent years,related scholars have explored the feasibility of applying near-infrared spectroscopy to quantitative analysis of blood hematocrit,but there are few related studies on further model optimization and grouping modeling strategies.In this paper,visible-near infrared（Vis-NIR）spectroscopy combined with partial least squares（PLS）method was used to establish a quantitative model for rapid detection of HCT in blood without reagents.Considering that the homogeneity of samples can help improve data correlation,reduce interference,and improve prediction effect,a grouping modeling strategy was proposed,and spectral analysis models for male,female and mixed groups were established respectively,and the parameters were optimized,and a comparative analysis of modeling effects was given.The main contents and results are as follows:1.Research on relevant spectral modeling methods:Based on the multi-division of calibration-prediction samples and the independence of inspection samples,a rigorous calibration-prediction-inspection experimental system was established,to improve model stability and objectivity;combined with the PLS method,a Norris derivative filtering（NDF）a multi-parameter optimization platform was established.According to the modeling effect,selected appropriate spectral preprocessing parameters（derivative order d,smoothing point number s,difference interval number g）;equidistant combination-PLS（EC-PLS）and wavelength step-by-step phase-out-PLS（WSP-PLS）methods,were used for the first and second stages of wavelength model optimization,respectively.Based on the Top 10 EC-PLS models,the globally optimal EC-WSP-PLS model was determined;the well-executed competitive adaptive reweighted sampling-PLS（CARS-PLS）method was also used for modeling as a comparison;finally,independent samples not involved in the modeling were used to test the preferred model to ensure the objectivity of the results.2.Modeling and testing effects:The optimal NDF parameters for male,female,and mixed groups were:s=27,g=16,d=1;s=29,g=3,d=2;s=11,g=22,d=2.The number of wavelengths（N）of the optimal EC-WSP-PLS model for male,female,and mixed groups were 20,22,and 12,respectively;the average prediction root mean square error of 10 divisions modeling(SEP_Ave)were 1.22%,0.85%,and 1.76%,respectively.The mixed model and the grouping model were used for testing,respectively.The root mean square error（SEP）of the test prediction in the male group were 1.56%and 1.26%,respectively,and the latter decreased by 18.8%;the SEP of the female group were 1.49%,0.95%,the latter decreased by 36.6%;the SEP of the mixed group were 1.52%,1.12%,and the latter decreased by 26.2%.The above results were better than the results obtained by the CARS-PLS method.Using the grouping model,the predicted and clinically measured values of all validation samples had a high prediction correlation coefficient（R_P=0.94）and a low relative prediction error（R-SEP=2.5%）,thus indicating the potential for clinical application.The results show that it is feasible to use Vis-NIR spectroscopy combined with EC-WSP-PLS method for rapid reagent-free analysis of blood HCT,and grouping modeling can greatly reduce the error of quantitative analysis（more than 18.8%）.The proposed spectral modeling strategy and analysis wavelength model can provide an important reference for the corresponding clinical applications and the design of small special-purpose spectrometers,and can also be applied to other analytical objects.