Multiple sclerosis (MS) is a prevalent neurodegener- ative disease with significant visual pathway-related symptoms. Optical coherence tomography (OCT) has emerged as a valuable tool, and machine learning (ML) techniques hold promise for MS diagnosis. However, ex- isting studies often lack comprehensive feature exploita- tion and require interpretable model analysis to improve clinical insights and diagnostic criteria. This study evaluates machine learning models for classification of healthy controls and MS patients using a comprehensive set of macular and optic-disc parameters from OCT imaging. The study included a dataset of 77 MS eyes and 54 control eyes, obtained by ophthalmic examination and OCT measurements from Optic Disc and Macular Cube scan protocols of a Cirrus HD-OCT 5000 (Carl Zeiss, Meditec, Dublin, CA, USA). Our results identi- fied 19 features, validated by p-values (p < 0.001), as effective discriminators between MS patients and healthy controls. Patient-wise cross-validation is used to eval- uate the performance of five ML algorithms. Gaussian Naive Bayes achieved the best AUC (87.9% ± 7.7%), while SHAP analysis reinforced the alignment with clin- ical observations of MS-related visual pathway changes and ganglion cell layer degeneration, with minimum ganglion cell thickness being the feature with the highest impact on classification. These findings underscore the potential of OCT-ML for early diagnosis and personal- ized treatment of MS.
Pablo et al. (Mon,) studied this question.