Machine learning for multiple sclerosis classification and disability prediction using clinical and MRI data.

Area of Science:

• Neuroimaging
• Machine Learning in Medicine
• Neurology

Background:

• Multiple sclerosis (MS) presents diverse clinical patterns, necessitating accurate classification and severity prediction for personalized treatment.
• Distinguishing MS patients from healthy controls (HC) and identifying disease phenotypes are critical clinical challenges.

Purpose of the Study:

• To apply machine learning (ML) models to demographic, clinical, and MRI data for MS patient classification and disability prediction.
• To differentiate between relapsing and progressive MS phenotypes.
• To predict disease severity using the Expanded Disability Status Scale (EDSS).

Main Methods:

• Utilized data from 1,554 MS patients and 520 HC, including neurological assessments and brain MRI scans.
• Extracted MRI features such as T2 lesion volumes (LV) and grey matter (GM) volumes.
• Trained various ML models (SVM, MLP, Random Forest, Gradient Boosting) for classification and prediction tasks.

Main Results:

• ML models achieved 89-96% accuracy in distinguishing MS from HC, primarily using T2 LV and brainstem/cerebellar GM volumes.
• Relapsing vs. progressive MS was classified with 92% accuracy, with EDSS, age, and thalamic/cortical GM volumes as key predictors.
• EDSS prediction showed moderate to good correlation (0.56-0.76), with T2 LV, sex, and GM volumes being significant contributors.

Conclusions:

• Machine learning models demonstrate high efficacy in MS detection, phenotype differentiation, and disability prediction.
• Integrating demographic, clinical, and MRI data offers a powerful strategy for patient classification and disease severity assessment in MS.
• This approach supports personalized medicine by improving the understanding and management of multiple sclerosis.