Exploring factors driving the evolution of chronic lesions in multiple sclerosis using machine learning

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Multiple sclerosis (MS) is a chronic inflammatory disease affecting the central nervous system.
• The evolution of chronic MS lesions is variable and not fully understood.
• Predicting lesion changes is crucial for understanding disease progression.

Purpose of the Study:

• To develop a machine learning model to predict the volume changes of chronic lesions in MS.
• To identify key factors influencing the evolution of chronic MS lesions.

Main Methods:

• Longitudinal data from relapsing-remitting MS (RRMS) patients were analyzed.
• Quantitative susceptibility mapping (QSM) identified the 'iron rim' sign.
• T1/FLAIR ratios quantified microstructural damage.
• Machine learning models (SVM, RF, LR) were trained to predict lesion volume change.

Main Results:

• A support vector machine (SVM) model showed high predictive efficiency (AUC 0.90 training, 0.81 testing).
• The most important predictors for lesion volume change were the 'iron rim' sign, disease duration (DD), and T1/FLAIR ratios.
• The study analyzed data from 47 RRMS individuals and 833 chronic lesions.

Conclusions:

• A machine learning model can effectively predict chronic MS lesion volume outcomes.
• Chronic inflammation around lesions, disease duration, and microstructural damage are critical factors in lesion evolution.
• These findings enhance our understanding of MS lesion dynamics.