Retinal microvascular differences between multiple sclerosis and neuromyelitis optica spectrum disorder: a...

Retinal alterations in multiple sclerosis (MS) and neuromyelitis optica spectrum disorder (NMOSD) remain unclear, especially the specific patterns and extent of microvascular change. This study aimed to compare these alterations and to develop logistic regression and machine learning models using combined retinal microvascular and structural metrics for disease differentiation.

Patients with MS or NMOSD in clinically stable phases underwent swept-source optical coherence tomography (OCT) and OCT angiography (OCTA). Quantified OCTA metrics included density, perfusion, and microdensity of superficial and deep vascular complex (SVC, DVC), choroidal vascular, and stromal volumes (CVV, CSV). Structural OCT metrics included retinal nerve fiber layer (RNFL) and ganglion cell-inner plexiform layer (GCIPL) thickness. Logistic regression and machine learning models were developed for classification.

The study enrolled 658 participants (167 MS patients, 221 NMOSD patients, and 270 age- and sex-matched healthy controls) with 1277 eyes. In optic neuritis (ON) eyes, NMOSD showed significantly lower SVC metrics than MS, whereas in non-ON eyes, MS exhibited more severe microvascular loss (most p < 0.017); correspondingly, EDSS-related SVC decline was steeper in ON eyes of NMOSD but steeper in non-ON eyes of MS (interaction p < 0.05 for most comparisons). The logistic regression model with microvascular metrics achieved an AUC of 0.900. Among machine learning classifiers, support vector machine performed best (AUC 0.912, accuracy 84.5%).

Distinct retinal microvascular patterns differentiate NMOSD from MS and correlate with disability severity, especially considering ON history. OCTA-based models provide accurate, non-invasive differential diagnosis tools, highlighting microvascular integrity as a critical biomarker.