Abstract
Neuromyelitis optica spectrum disorder (sNMOSD) poses significant diagnostic challenges due to its clinical and radiologic similarities with multiple sclerosis (MS), despite requiring distinct therapeutic approaches. Early, accurate differentiation is essential to prevent irreversible disability and avoid inappropriate treatment. Machine learning (ML) offers a powerful means to detect subtle imaging differences, such as peri-ependymal lesions and area postrema involvement, often missed in conventional assessment. Recent advances include radiomics-based logistic regression models achieving AUCs above 0.92, compressed 3D convolutional neural networks outperforming traditional models, and transformer-based fusion networks integrating brain and spinal cord MRI with AUCs exceeding 0.93 for NMOSD classification. Incorporating immunological testing, particularly aquaporin-4 (AQP4) antibody status, further improves discrimination between NMOSD, MS, and myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD). While promising, generalizability is limited by the scarcity of large, multi-center, diverse datasets. Combining advanced ML techniques with robust immunologic profiling offers a path toward precision diagnostics, enabling earlier intervention and improved outcomes for patients with NMOSD.