Study On Nondestructive Diagnosis Of Hyperglycemia And Hyperuricemia By Infrared Spectroscopy Based On Human Urine

Hyperglycemia and hyperuricemia are common human metabolic disorders.The traditional diagnostic methods for hyperglycemia and hyperuricemia are both invasive tests,which can bring risks to patients such as mental stress,bacterial infection,and secondary injury.Therefore,there is an urgent need for a non-invasive technique to help patients diagnose hyperglycemia and hyperuricemia easily and non-destructively.In this study,the infrared spectrum data of human urine was obtained,and the spectrum was filtered by the band selection algorithm.The data fusion strategy and stack generalization algorithm were combined to establish a diagnosis with high accuracy,strong robustness and good generalization.Non-destructive diagnostic model for hyperglycemia and hyperuricemia.The specific content is as follows:（1）Study on diagnosis model of hyperglycemia and hyperuricemia based on characteristic wavelength screening algorithmThe experiment collected NIR and FTMIR spectra of morning urine of 159 non-hyperglycemia,106 hyperglycemia,93 non-hyperuricemia and 102 hyperuricemia volunteers.Secondly,the first derivative,second derivative,standard normal transformation,and multivariate scattering correction are used to preprocess the original spectral data.Then,GA,CARS,and BGWO are used to filter the two spectral sources with characteristic bands.And preliminarily established the diagnosis and prediction model of hyperglycemia and hyperuricemia.The results show that the prediction accuracy of the PLS-DA model established after wavelength screening is significantly higher than that of the PLS-DA model established for the full spectrum.Optimal hyperglycemia diagnosis model:（1）NIR model:After the second derivative is processed,the data modeling effect after filtering through the CARS algorithm band is the best,and the accuracy of the model prediction result is 0.8750.The selected characteristic wavelengths are 1030 nm,1732 nm,and 1980～2180 nm.（2）FTMIR model:using the original spectrum,the data modeled by the CARS algorithm is the best,and the accuracy of the model prediction result is 0.9048.The selected characteristic wavelengths are around 1000 cm^-1and 1500 cm^-1.Optimal hyperuricemia diagnosis model:（1）NIR model:After standard normal transformation processing,the data modeling effect after screening through the BGWO algorithm band is the best,and the accuracy of the model prediction result is 0.7778.The selected characteristic wavelengths are 1030 nm,1540 nm,1732 nm,1980～2180 nm and 2350 nm.（2）FTMIR model:After the second derivative is processed,the data modeled by the CARS algorithm has the best modeling effect,and the accuracy of the model prediction result is 0.8491.The selected characteristic wavelengths are around 1000 cm^-1 and 1500 cm^-1.（2）Study on the diagnostic model of hyperglycemia and hyperuricemia based on spectral data fusionThe experiment obtains feature variables through the selected optimal band screening algorithm（CARS,BGWO）and the PCA/ICA method for feature extraction.After fusing the extracted feature variable data,it is input into the PLS-DA model for modeling analysis.The results show that the PLS-DA analysis model established by the fusion of the two spectral data is better than the hyperglycemia and hyperuricemia diagnostic model established by a single spectral data.Among them,the model established by the band screening algorithm combined with intermediate data fusion has the best effect,and the accuracy of the prediction result of the hyperglycemia diagnosis model is 0.9574.The accuracy of the prediction result of the hyperuricemia diagnostic model was 0.9057.（3）Study on the diagnostic model of hyperglycemia and hyperuricemia based on stacking generalizationThe experiment uses the fusion data of the urine infrared spectrum to establish four machine learning classifications.On this basis,the ensemble learning idea is introduced,and the four classifiers are integrated into one strong classifier using stacking generalization.The meta-learner layer of stacking generalization uses logistic regression and weighted voting methods to model and compare.The results show that the performance of the model after applying stacked generalization is significantly improved in all aspects compared to a single classifier.Among them,weighted voting used as the meta-learner is better.The accuracy of the prediction results of the hyperglycemia diagnostic model,F1,and the sensitivity are 1,1,1,respectively,and the accuracy of the prediction results of the hyperuricemia diagnostic model,F1,Sensitivity is 0.9434,0.9473,0.9310 respectively.Therefore,it is feasible to diagnose and differentiate hyperglycemia and hyperuricemia through infrared spectroscopy of human urine.This method is fast,simple,and non-destructive.The results of this study have positive practical significance for the diagnosis of hyperglycemia and hyperuricemia and the daily detection and control of the patient’s condition.