Method Study On Near-infrared Spectroscopic Of Serum Urea Nitrogen

Serum urea nitrogen?SUN?is an important blood clinical index,which has clinical value for kidney function,liver function and other diseases.In the process of health screening and disease treatment,it is often necessary to monitor SUN level quickly and in real time.However,the current clinical measurement methods are cumbersome and time consuming,and cannot meet the above requirements.Near-infrared spectroscopy is an important application direction in the field of optical engineering.This method requires no reagents and is quick and easy.In recent years,near-infrared spectroscopy has been attempted to be applied to the rapid detection of serum urea nitrogen.Since blood is a complex system with multiple components,blood near-infrared analysis needs to solve many chemometric problems.In this paper,near-infrared spectroscopy analysis of serum urea nitrogen is taken as an example,focusing on the research of correlation spectral pretreatment and wavelength model optimization to mprove the prediction ability of blood spectral analysis.The main results of this study are as follows:?1?Methods research:1)Multi-partition modeling design with stability:Firstly,samples?75?that are not involved in modeling are randomly selected for independent validation,the remaining samples are further randomly divided into calibration?80?and prediction?55?samples for 10 times.Based on all partitions,the model is modeled and the parameters are optimized according to the average prediction effect,which makes the model stable and reliable.2)Norris derivative filter?NDF?and its multi-parameter optimization platform:Based on partial least squares?PLS?regression,a Norris parameter optimization platform for derivative order?d?,smooth points?s?and differential interval?g?cycles is constructed.The optimal Norris parameters of PLS model for serum urea nitrogen were determined.3)Wavelength screening method for equidistant combination?EC?and wavelength step-by-step phase-out?WSP?:Using the starting wavelength?I?,the number of wavelengths?N?and the number of intervals?G?as cycle parameters to achieve a wide range of EC-PLS models preferably.Further,the WSP method is used to improve the top 100 EC-PLS models with leading effects and to select them.Eliminate interference wavelengths in equidistant spaced wavelength combinations,simplify wavelength models and improve spectral prediction.4)Backward interval PLS?BiPLS?wavelength screening method and its improved algorithm:An improved moving window-backward interval PLS?MW-BiPLS?algorithm is proposed,which realizes two-parameter cycle of interval starting wavelength and wavelength number and achieves better modeling effect than BiPLS.?2?Experimental research:1)EC-PLS model:The optimal EC-PLS model for serum urea nitrogen analysis is I=1228nm,N=26,G=16 and LV=12.The deviation between the predicted value of the independent validation and the clinical measured value?SEP?,correlation coefficient?R_P?,sample deviation and predicted deviation ratio?RPD?were 1.03 mmol L^-1,0.992 and 7.6,respectively.2)EC-WSP-PLS model:The first 100 EC-PLS models were improved and selected by WSP method.The optimal WSP-EC-PLS model was N^*=19 and LV=12,the corresponding SEP,R_P,and RPD were 1.01 mmol L^-1,0.992 and 7.7,respectively.Both the model prediction accuracy and the wavelength model complexity are improved.3)BiPLS,MW-BiPLS models:The optimal BiPLS model obtained is N=22,LV=18 and the corresponding SEP,R_P and RPD are 1.12 mmol L^-1,0.985 and 7.0,respectively.The optimal MW-BiPLS model obtained is N=77,LV=15,the corresponding SEP,R_P and RPD are 1.10mmol L^-1,0.986 and 7.1,respectively.The improved MW-BiPLS method achieves better prediction results.In this paper,the spectral pretreatment and wavelength model optimization methods for the near-infrared analysis of serum urea nitrogen were systematically studied and good spectral prediction results were obtained.The proposed WSP-PLS is a quadratic optimization method for wavelength models.It can further optimize any wavelength model obtained through other strategies?continuous,discrete?for simpler and better predictions.The improved MW-BiPLS method also achieved a significantly better prediction than BiPLS.It provides a valuable reference for the improvement of blood spectroscopy methods and the clinical application of near-infrared spectroscopy.