Single-timepoint low-dimensional characterization and classification of acute versus chronic multiple sclerosis...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• Multiple sclerosis (MS) is a chronic inflammatory and neurodegenerative disease.
• Distinguishing acute from chronic MS lesions is crucial for patient management and clinical trials.
• Current methods using post-gadolinium MRI may underestimate acute lesion activity.

Purpose of the Study:

• To improve the sensitivity of acute MS lesion detection in single-timepoint MRI scans.
• To develop a novel machine learning approach for automatic acute MS lesion detection.
• To utilize conventional non-contrast T1- and T2-weighted brain MRI for enhanced detection.

Main Methods:

• A convolutional neural network was used for image inpainting to generate "lesion-free" reconstructions.
• A multi-objective statistical ranking module evaluated radiomic features from lesion sites.
• An ensemble classifier was optimized using a recursive loop for feature and model selection.

Main Results:

• The method identified a compact textural signature characterizing MS lesion phenotype.
• The approach achieved a balanced accuracy of 74.3-74.6% for acute versus chronic MS lesion classification.
• Validation was performed on fully external cohorts using a patch-level classification task.

Conclusions:

• The developed machine learning method shows promise for sensitive detection of acute MS lesions.
• This approach may improve the assessment of inflammatory disease activity in MS.
• The findings could support clinical decision-making and enhance clinical trial endpoints.