Leveraging Hand-Crafted Radiomics on Multicenter FLAIR MRI for Predicting Disability Progression in People with Multiple Sclerosis

Multiple sclerosis (MS) is a chronic autoimmune disease of the central nervous system that results in varying degrees of functional impairment. Conventional tools, such as the Expanded Disability Status Scale (EDSS), lack sensitivity to subtle changes in disease progression. Radiomics offers a quantitative imaging approach to address this limitation. This study used machine learning (ML) and radiomics features derived from T2-weighted Fluid-Attenuated Inversion Recovery (FLAIR) magnetic resonance images (MRI) to predict disability progression in people with MS (PwMS). A retrospective analysis was performed on real-world data from 247 PwMS across two centers. Disability progression was defined using EDSS changes over two years. FLAIR MRIs were preprocessed using bias-field correction, intensity normalisation, and super-resolution reconstruction for low-resolution images. White matter lesions (WML) were segmented using the Lesion Segmentation Toolbox (LST), and MRI tissue segmentation was performed using sequence Adaptive Multimodal SEGmentation. Radiomics features from WML and normal-appearing white matter (NAWM) were extracted using PyRadiomics, harmonised with Longitudinal ComBat, and reduced via Spearman correlation and recursive feature elimination. Elastic Net, Balanced Random Forest (BRFC), and Light Gradient-Boosting Machine (LGBM) models were evaluated on validation data and subsequently tested on unseen data. The LGBM model with harmonised radiomics and clinical features outperformed the clinical only model by achieving a test performance of PR AUC of 0.20 and a ROC AUC of 0.64. Key predictive features, among others, included GLCM maximum probability (WML), GLDM dependence non uniformity (NAWM). Short-term changes (longitudinal imaging approach) showed limited predictive power by achieving a PR AUC of 0.11 and a ROC AUC of 0.69. These findings support the use of ML models trained on radiomics features integrated with clinical data for predicting disability progression in PwMS. Future studies should validate these findings in larger, balanced datasets and explore advanced approaches, such as deep learning and foundation models, to enhance predictive performance.