The Clinical Study Of Radiomics Based On MRI T2WI Single Sequence For Differentiating Rheumatoid Arthritis From Gouty Arthritis

Objective:To evaluate the feasibility of radiomics based on MRI T2 WI single sequence for differentiating rheumatoid arthritis(RA)from gouty arthritis(GA).The differential diagnostic efficiency of the radiomics models developed by the common feature selection methods were compared to determine the feature selection method with better identification efficiency and sift the important features of the radiomics.Methods:The MRI imaging data of 81 joints with RA and 61 joints with GA were retrospectively analyzed.For each patient,three ROIs in the most clearly proliferative synovium were manually delineated on T2WI/FS images referred to the contrast images and a total of 972 features were extracted from ROIs.Study cohort was randomly divided into 70% of the samples for the training cohort and the remaining 30% of the samples for the training cohort.The intra-class correlation coefficient(ICC)was calculated to evaluate the consistency of the features extracted by the two doctors,and the features with good consistency(ICC>0.75)were selected using principal component analysis(PCA),least absolute shrinkage selection operator(LASSO)and Relief F algorithm to filter and reduce the features.The sifted features were applied to support vector machine(SVM)machine learning algorithm for developing radiomics models.The diagnostic performance of these models were estimated and compared using the area under the receiver operating characteristic(ROC)curve(AUC)and accuracy(ACC).Results:94 features showed good stability and reproducibility(ICC>0.75).The16 features selected by the Relief F feature selection method showed the best discrimination performance in the training cohort and training cohort with AUCs of 0.9213 and 0.8812,ACCs of 0.8592 and 0.8239,respectively.The minimum,skewness and size zone nonstability were the important radiomics features.The radiomics model developed by Lasso feature selection algorithm showed second discrimination performance in the training cohort and training cohort with AUCs of 0.9125 and 0.8735,ACCs of 0.8380 and 0.8239,respectively.The radiomics model developed by PCA feature selection algorithm displayed the worst discrimination performance,achieved AUCs of 0.8907 and 0.8579,ACCs of 0.8028 and 0.7958 in the training cohort and training cohort,respectively.Conclusion:Radiomics based on MRI T2 WI single sequence could effectively distinguish RA from GA,and the radiomics model developed by the Relief F feature selection method displayed the best discrimination efficiency.Minimum,skewness and size zone nonstability were the important radiomics features.