Ensemble Learning for Multiple Sclerosis Disability Estimation Using Brain Structural Connectivity.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Multiple sclerosis (MS) is a central nervous system autoimmune disease causing demyelination and neurodegeneration.
• Predicting MS disability is crucial for personalized treatment strategies.
• Current machine learning (ML) models for MS prognosis lack transparency.

Purpose of the Study:

• To develop an interpretable ML model for estimating MS patient disability using structural connectivity data.
• To compare the performance of ensemble ML models with conventional MRI measures.
• To identify critical white matter (WM) pathways associated with MS disability.

Main Methods:

• An ensemble of four boosting-based ML models (GBM, XGBoost, CatBoost, LightBoost) was employed using diffusion tensor imaging (DTI)-based structural connectivity.
• A conditional logistic regression model was used for interpretability, identifying key WM links.
• Counterfactual methods were applied to explain the model's predictions.

Main Results:

• The ensemble ML model achieved high performance in estimating disability (RMSE of 0.92 ± 0.28).
• DTI-based connectivity outperformed conventional MRI measures combined with patient data (age, gender, disease duration).
• The interpretable model highlighted the importance of left hemisphere WM links and the corpus callosum, with better accuracy for high disability classes.

Conclusions:

• Combining advanced ML models with DTI-based structural connectivity effectively estimates MS disability.
• This approach identifies crucial WM networks involved in MS-related disability.
• The findings support the clinical utility of integrating DTI, graph modeling, and ML for MS patient management.