Multimodal MRI biomarkers and machine learning in neuromyelitis optica spectrum disorder: a review

Abstract

Neuromyelitis optica spectrum disorder (NMOSD) is a relapsing inflammatory demyelinating disease of the central nervous system that can result in substantial cumulative neurological disability. Magnetic resonance imaging (MRI) plays a central role in lesion detection, differential diagnosis, and longitudinal assessment. However, conventional MRI may not fully capture subtle tissue damage and may show substantial overlap with related inflammatory demyelinating disorders in clinically ambiguous cases. Reliable imaging biomarkers reflecting disease burden, progression, and prognosis remain limited. This narrative review summarizes current evidence on conventional MRI features and advanced quantitative MRI biomarkers in NMOSD. We found that advanced techniques, including diffusion tensor imaging, morphometric analyses of T1 imaging, and resting-state functional MRI have revealed structural, microstructural, and functional abnormalities beyond lesions visible on conventional MRI. Furthermore, imaging-based machine learning applications have shown promising performance in differential diagnosis, particularly in distinguishing NMOSD from multiple sclerosis. However, findings across studies vary in the distribution, magnitude, and clinical relevance of reported abnormalities, likely reflecting differences in disease stage, acquisition protocols, and analytical methods. Additionally, most available studies remain limited by small cohorts, methodological heterogeneity, and insufficient external validation. Evidence for relapse prediction and treatment monitoring remains comparatively limited. Future research should prioritize multicenter prospective validation, harmonized imaging protocols, clearer separation of disease subgroups, and integration of imaging with clinical, serological, cerebrospinal fluid, and ophthalmic biomarkers before routine clinical implementation.