AI-Based Classification of Multiple Sclerosis Using OCT Retinal Layer Thickness Across Two Centers

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• The latest McDonald criteria include the optic nerve as a diagnostic site for multiple sclerosis (MS).
• Optical coherence tomography (OCT) provides imaging evidence for MS diagnosis.
• The optic nerve's role in MS diagnosis is increasingly recognized.

Purpose of the Study:

• To evaluate the accuracy of AI-based classification for MS patients using OCT data.
• To assess the performance of automated MS classification across different clinical centers.
• To determine the effectiveness of AI in analyzing OCT for MS diagnosis.

Main Methods:

• OCT data collected from two centers using standardized protocols.
• Analysis of retinal layer thicknesses (mean and STD) in multiple layers and regions.
• Application of a support vector machine classifier with feature selection (recursive feature elimination) and explainability analysis (Shapley additive explanations).

Main Results:

• A classifier achieved 0.8459 accuracy using 20 features.
• Mean and standard deviation metrics were similarly important.
• Key layers identified: ganglion cell layer, inner plexiform layer, inner retinal layer complex.
• Significant regions: papillomacular bundle and superior temporal perimacular area.

Conclusions:

• OCT data enables high-accuracy, multi-center MS diagnosis.
• AI-driven OCT analysis shows potential for automated MS classification.
• These findings underscore the optic nerve's critical role in diagnosing MS.