Deep Learning Model Differentiates Neuroinflammatory Disorders

A deep learning model was able to determine the presence or absence of multiple sclerosis (MS), myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD), and neuromyelitis optica spectrum disorders (NMOSD), according to study results presented at the European Committee for Treatment and Research in Multiple Sclerosis (ECTRIMS) Congress 2025, held in Barcelona, Spain from September 24 to 26, 2025.

Patients with MS, MOGAD, and NMOSD have overlapping clinical symptoms despite distinct pathophysiology. Patients with autoimmune neuroinflammatory disorders can be distinguished from controls by retinal thickness measured by optical coherence tomography (OCT), but it remains unclear whether OCT data may be used to distinguish between disorder subtypes.

Researchers used optic nerve and macular volumetric OCT images from 623 patients and controls in the US, Germany, and India. Data collected at 2 of the study sites were used to train a Residual Deep Convolutional Neural Network with a Variational AutoEncoding model, which was tested with data from all 3 sites. Cross-validation analysis was conducted by randomly splitting the data into training (70%), validation (15%), and test (15%) sets.

The study included patients with MS (n=345), MOGAD (n=92), NMOSD (n=69), primary ocular diseases (n=16), and controls (n=101).

The deep learning model identified each autoimmune neuroinflammatory disorder with an area under the receiver operating characteristic curve ranging between 0.81 and 0.85 using all data, and between 0.87 and 0.93 using the training, validation, and test data sets. The model was robust and generalizable across the study sites and imaging conditions.

The study authors concluded, “The model’s strong performance across diverse datasets suggests its potential for clinical integration, aiding in early and accurate diagnosis, especially in the settings with limited access to subspecialty clinical expertise.”

Disclosures: Some study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.