Predicting 10-Year Clinical Outcomes in MS with Radiomics-Based Machine Learning Models.
Mario Tranfa1,2, Maria Petracca3, Renato Cuocolo4
1From the Department of Advanced Biomedical Sciences (M.T., L.U., A.E., A.S., A.B., S.C., G.P.), University of Naples "Federico II," Naples, Italy.
AJNR. American Journal of Neuroradiology
|July 3, 2025
Summary
Machine learning models using routine brain MRI features can predict long-term disability progression and cognitive impairment in people with multiple sclerosis (pwMS). These prognostic models aid in stratifying patients for better clinical management.
Area of Science:
- Neuroimaging
- Machine Learning
- Neurology
Background:
- Identifying patients with multiple sclerosis (pwMS) at high risk of clinical progression is crucial for effective management.
- Routine brain MRI scans offer a rich source of data for predicting long-term outcomes.
Purpose of the Study:
- To develop and validate machine learning (ML) prognostic models for predicting long-term clinical outcomes in pwMS.
- To utilize volumetric, radiomic, and macrostructural disconnection features from routine brain MRI scans.
Main Methods:
- Retrospective analysis of 3T structural MRI scans from 177 pwMS with a ten-year follow-up.
- Classification of patients based on confirmed disability progression (CDP) and cognitive impairment (CI) using EDSS and BICAMS.
- Segmentation of MRI images to extract volumetric, disconnection, and radiomic features from 116 brain regions (AAL atlas).
- Development of predictive models using Extra Trees, Logistic Regression, and Support Vector Machine algorithms with feature selection and holdout validation.
Main Results:
- ML models achieved moderate accuracy in predicting long-term CDP (up to 0.71) and CI (up to 0.69) on the test set.
- No significant differences in accuracy were observed between the different ML models for either CDP or CI prediction.
- Feature subsets identified by ML algorithms provided predictive power for long-term clinical outcomes.
Conclusions:
- Quantitative features derived from conventional MRI scans can be used to build effective long-term prognostic models for pwMS.
- These models have the potential to inform patient stratification and guide clinical decision-making in multiple sclerosis management.
