Diagnosis of multiple sclerosis using optical coherence tomography supported by explainable artificial intelligence.

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• Relapsing-remitting multiple sclerosis (RRMS) diagnosis relies on neuroimaging, but early detection remains challenging.
• Optical coherence tomography (OCT) offers a non-invasive method to assess retinal structure, a potential indicator of neurological changes.
• Explainable artificial intelligence (AI) is crucial for reliable clinical decision-making in diagnosing complex conditions like RRMS.

Purpose of the Study:

• To investigate the utility of OCT-derived retinal parameters for the early diagnosis of RRMS.
• To develop and validate an AI-based classification model using OCT data for RRMS detection.
• To identify specific retinal layers and zones most predictive of RRMS and functional disability.

Main Methods:

• A cohort of 79 recently diagnosed RRMS patients and 69 healthy controls underwent OCT imaging.
• Retinal layer thickness (Avg) and inter-eye difference (Diff) were analyzed in six zones using the posterior pole protocol.
• Support Vector Machine with Recursive Feature Elimination and Leave-One-Out Cross-Validation (SVM-RFE-LOOCV) was employed for feature selection and classifier optimization.

Main Results:

• The SVM-RFE-LOOCV approach identified the papillomacular bundle area as the most influential feature for RRMS diagnosis.
• A significant correlation was found between reduced retinal layer thickness, increased inter-eye asymmetry, and greater functional disability in RRMS patients.
• The developed classifier achieved a sensitivity of 0.86 and a specificity of 0.90.

Conclusions:

• The identified OCT features align with the known neuroanatomy of retinal nerve fibers and the optic nerve head in RRMS.
• This AI-assisted diagnostic approach using OCT demonstrates superior accuracy compared to current central nervous system MRI standards.
• The study provides novel insights into the neuroretinal impact of RRMS, supporting the development of advanced diagnostic systems.