Explainable Artificial Intelligence to Predict Neurocognitive Disorder Progression in Multiple Sclerosis Using MRI...

Area of Science:

• Neurology
• Artificial Intelligence
• Neuroimaging

Background:

• Cognitive impairment is prevalent in multiple sclerosis (MS), but diagnostic frameworks for Neurocognitive Disorders (NCDs) are underutilized.
• Integrating multimodal data with artificial intelligence (AI) for predicting cognitive outcomes in MS is an underexplored area.

Purpose of the Study:

• To characterize NCDs in MS patients.
• To predict cognitive worsening in MS using an explainable deep learning model.
• To integrate MRI and clinical data for cognitive outcome prediction.

Main Methods:

• 224 MS patients and 115 healthy controls underwent MRI and clinical assessments.
• Neuropsychological testing and cognitive reserve estimation were performed at baseline and follow-up.
• A deep learning model was trained on multimodal data (MRI, demographics, clinical, volumetric) to predict cognitive decline.

Main Results:

• At baseline, 15% of MS patients met criteria for Mild or Major NCD.
• The deep learning model achieved 90% accuracy in predicting follow-up cognitive status.
• Key predictors identified included cortical gray matter volume, age, thalamic/hippocampal volumes, T2 lesion volume, and cognitive reserve.

Conclusions:

• A multimodal AI approach shows strong performance in predicting cognitive worsening in MS.
• The model highlights critical brain regions and factors associated with cognitive decline.
• This AI strategy holds potential for personalized cognitive assessment and monitoring in MS.