An Interpretable Machine Learning Model to Predict Cortical Atrophy in Multiple Sclerosis.

Area of Science:

• Neuroimaging
• Neurology
• Radiology

Background:

• The relationship between white matter (WM) lesions, cortical gray matter atrophy, and multiple sclerosis (MS) progression is not fully understood.
• New radiological markers, including specific lesion types, are emerging for MS disease tracking.

Purpose of the Study:

• To investigate the interplay between cortical atrophy and distinct lesion patterns in multiple sclerosis.
• To identify key radiological and demographic factors predicting cortical thinning in MS patients.

Main Methods:

• Utilized a machine learning model to predict mean cortical thinning across 150 cortical regions.
• Employed ultrahigh field (7 Tesla) MRI to evaluate demographic and lesion characteristics.
• Analyzed the predictive power of white matter lesion volume, rimless lesions, patient age, and intracortical lesion volume.

Main Results:

• White matter lesion volume, rimless lesions, patient age, and intracortical lesion volume were the strongest predictors of cortical thinning.
• The relative importance of white matter versus cortical lesions shifted with disease progression.
• White matter lesions were more influential in early stages, while cortical lesions became more critical in later stages of atrophy progression.

Conclusions:

• MS pathology involves complex interactions between different lesion types and brain structures.
• Understanding these interactions necessitates advanced statistical and mechanistic modeling approaches.
• These findings underscore the need for comprehensive analysis of lesion patterns in MS progression.