A Disease Activity Evaluation Model Of Rheumatoid Arthritis Was Constructed Based On Raman Spectrum Combined With Machine Learning

Objective: Serum spectral characteristics of Rheumatoid arthritis(RA)in different disease states were obtained based on Raman spectroscopy.Combining support vector machine(SVM)and Linear Discriminant Analysis(LDA)algorithms,a fast,convenient and accurate model for evaluating RA disease activity was established.The aim is to use this model to assist the clinic to complete RA diagnosis and treatment more efficiently.Method: A total of 116 patients admitted to the rheumatology and immunology department of Xinjiang Uygur Autonomous Region People’s Hospital from January 2022 to December 2022 who were diagnosed With RA were selected.According to Disease Activity Score With 28-Joint Counts,Patients included in the study were divided into low group(DAS28≤3.2)and high group(DAS28 > 3.2).44 serum samples from low group and 72 serum samples from high group were detected by Raman spectroscopy and spectral data were obtained.After noise reduction,dimension reduction and feature extraction of the original spectral data,difference spectral maps were obtained.After processing the data,SVM,LDA and other algorithms were used to establish the disease activity evaluation model.Results: 1.The titers of joint swelling,joint tenderness,erythrocyte sedimentation,C-reactive protein,rheumatoid factor and anti-CCP antibody in high group were higher than those in low group(P < 0.05).2.In the difference spectrogram,after independent t test,the Raman spectral peak value at 642,853,934,1127,1186,1326,1445,1657cm-1 displacement of RA patients in the high group was lower than that in the low group(P < 0.05).3.SVM and LDA algorithms were used to model the data before and after dimensionality reduction by Principal Component Analysis(PCA),and four models were obtained: SVM,LDA,PCA-SVM and PCA-LDA.By analyzing the accuracy,sensitivity,specificity,and Area Under Curve(AUC)of the four models,PCA-LDA was the best in distinguishing the low group from the high group(accuracy 90%;Sensitivity is100%;Specificity 71.4%;AUC0.82).Conclusion: In this study,SVM and LDA algorithms were combined to model the data before and after dimension reduction,and the obtained four groups of machine learning models could effectively discriminate the disease activity of RA,among which the PCA-LDA model had the best discrimination effect.