Explainable machine learning on baseline MRI predicts multiple sclerosis trajectory descriptors.

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Multiple sclerosis (MS) presents diagnostic and management challenges.
• Artificial Intelligence (AI) offers potential for early detection and personalized treatment in MS.
• Predictive modeling for MS patient evolution is crucial for effective management.

Purpose of the Study:

• To propose novel Multiple Sclerosis (MS) trajectory descriptors for Machine Learning (ML) models.
• To assess the capability of ML models in identifying MS trajectory descriptors from baseline Magnetic Resonance Imaging (MRI) data.
• To evaluate the predictive performance of AI models in forecasting MS patient progression.

Main Methods:

• Utilized data from 446 MS patients with baseline MRI, Expanded Disability Status Scale (EDSS) measurements, and 1-year follow-up.
• Developed and evaluated regression and classification XGBoost models to relate MS trajectory descriptors (β1, β2, EDSS(t)) to baseline MRI parameters.
• Employed Shapley Additive Explanations (SHAP) analysis for model transparency and feature importance identification.

Main Results:

• AI models, utilizing proposed MS trajectory descriptors and baseline MRI, demonstrated superior prediction of MS progression compared to traditional Multiple Linear Regression (MLR).
• SHAP analysis successfully identified key features influencing MS progression predictions, enhancing model interpretability.
• The study confirmed the potential of AI in analyzing baseline MRI for predicting patient evolution in MS.

Conclusions:

• MS trajectory descriptors are vital for predicting disease progression.
• Integrating AI analysis with MRI assessments significantly advances predictive capabilities for MS.
• SHAP analysis provides crucial insights into feature importance, supporting clinical decision-making in MS management.