Prediction Of Unenhanced Lesions Evolution In Multiple Sclerosis Using Radiomics-based Models:A Machine Learning Approach

Objective:The volume change of multiple sclerosis(MS)unenhanced lesion is related to its activity and can be used to assess disease progression.Therefore,the purpose of this study was to develop radiomics models to predicting lesion volume changes by using different kinds of machine learning algorithms,thus investigating the chronic activity of unenhanced MS lesions and explore which of them presents the best performance.Materials and Methods:In this study,follow-up MR images obtained in 36 patients with MS(mean age 32.53 years±10.91;23 women,13 men)were evaluated.The unenhanced lesion will be defined as chronic active and chronic inactive lesion,respectively,based on the percentage of enlargement or reduction of the lesion >20% in the follow-up MR images.We extracted radiomic features of lesions on T2-FLAIR images,and used recursive feature elimination(RFE),Relief F algorithm and least absolute shrinkage and selection operator(LASSO)for feature selection,then three classification models including logistic regression,random forest and support vector machine(SVM)were used to build predictive models.The consistency and effectiveness of model performance were compared and verified by different combinations of feature selection and machine learning methods in different K-fold cross-validation strategies where K ranges from 5 to 10,thus demonstrating the stability and robustness.The performance of the models was evaluated based on the accuracy,sensitivity,specificity and receiver operating characteristic curve(ROC)curves analyses.Results:There were 135 chronic inactive lesions and 110 chronic active lesions included in our study.A total of 972 radiomics features were extracted,of which 265 were robust.SVM classifier with Relief F algorithm had the best prediction performance with an average accuracy of 0.928,sensitivity of0.915,specificity of 0.937,and AUC of 0.944 in the training set,and an average accuracy of 0.827,sensitivity of 0.809,specificity of 0.841,and AUC of 0.857 in the validation set.The most suitable model selected a total of 11 radiomics features,consisting of 6 texture features and 5 higher-order statistical features,among which LDHGLE in the higher-order statistical features,SAHGLE and HGLZE in the texture feature-GLSZM are the more important features.Conclusion:The radiomics model based on T2-FLAIR is of high value in predicting the chronic activity of unenhanced MS lesions.