Research On Rheumatoid Arthritis Diagnosis Model Based On Raman Spectroscopy Combined With Machine Learning

Objective:The Raman spectroscopy combined with machine learning technique was applied to classify Rheumatoid Arthritis（RA）by serum detection and establish a diagnostic model of rheumatoid arthritis based on Raman spectroscopy combined with machine learning.Methods:A total of 142 patients diagnosed with RA in the rheumatism immunology department of the Xinjiang Uyghur People’s Hospital between January 2021 and December 2022 were selected as the experimental group,while 271 patients in the control group（92 patients with ankylosing spondylitis,104 patients with osteoarthritis,and 75 patients in the health checkup group）were collected for Raman spectroscopy.The measured spectral dataset is analyzed by adaptive iterative reweighted penalized least squares（air PLS）to remove the fluorescence background and correct the baseline,and then principal component feature extraction is carried out by Principal Component Analysis（PCA）.Then we use Support Vector Machine（SVM）algorithm,linear discriminant analysis（LDA）algorithm to establish diagnostic model,and use 8:2 segmentation method to establish training set and verification set.According to the classification results,we calculate accuracy,specificity,sensitivity,draw receiver operating characteristic curve（ROC）and obtained area under curve（AUC）to evaluate the classification efficiency of the model.Results:There were significant differences between RA group and control group in the shifts of 524cm^-1、573cm^-1、640cm^-1、717cm^-1、776cm^-1、850cm^-1、1053cm^-1、1120cm^-1、1180cm^-1、1250cm^-1、1324cm^-1、1450cm^-1、1657cm^-1（P<0.05）.The changes of these spectral peaks correspond to different biomolecules and metabolites,which may be beneficial to the recognition and screening of RA by serum Raman spectroscopy.The PCA method is used for the extraction of spectral feature information by dimensionality reduction of spectral data.The feature principal component can distinguish the two groups of samples in a two-dimensional plane.Further,PCA-SVM and PCA-LDA models were established to classify RA.The results showed that PCA-SVM model had a classification sensitivity of 89.7%,a specificity of 92.7%（AUC=0.91）,a classification sensitivity of 87.1%,a classification specificity of 89.8%（AUC=0.85）,and PCA-LDA model had a better classification effect.Conclusion:The study showed that the diagnosis model based on Raman spectroscopy combined with machine learning can classify and identify RA,and Raman spectroscopy technology can provide new ideas and technical support for the diagnosis and screening of RA.