2025 American Academy of Neurology Abstract Website

Objective:

This systematic review and meta-analysis seeks to assess the diagnostic accuracy of artificial intelligence (AI) algorithms in the early detection of multiple sclerosis (MS) and neuromyelitis optica (NMO) utilizing neuroimaging data.

Background:

Multiple sclerosis (MS) and neuromyelitis optica (NMO) are debilitating neurological disorders requiring early detection for optimal treatment.

Design/Methods:

We conducted a comprehensive search of the PubMed, MEDLINE, and Embase databases for research published between 2015 and 2024 that used AI algorithms to analyze neuroimaging data for early detection of MS and NMO. A total of 28 studies were included, encompassing 4,350 patients (2,750 with MS, 1,100 with NMO, and 500 healthy controls) with an age range of 18 to 65 years. Neuroimaging modalities investigated included MRI, diffusion tensor imaging , and optical coherence tomography , alongside AI models such as convolutional neural networks , support vector machines, and random forest methods.

Results:

AI-based models' pooled sensitivity and specificity for diagnosing MS were 89% (95% CI: 86%-92%) and 91% (95% CI: 88%-94%), respectively, with an AUC of 0.94 (95% CI: 0.91-0.97). AI models for NMO revealed a pooled sensitivity of 87% (95% CI: 83%-91%), a specificity of 90% (95% CI: 87%-93%), and an AUC of 0.93 (95% CI: 0.89-0.96). CNN models performed marginally better than conventional machine learning approaches, with AUC values of 0.96 for MS and 0.95 for NMO. MRI-based AI models outperformed OCT and DTI models, particularly for distinguishing early-stage MS and NMO from healthy controls.

Conclusions:

AI-driven neuroimaging analysis has enormous potential for the early diagnosis of MS and NMO, with high sensitivity and specificity. CNN models applied to MRI data produced the most accurate results. These results suggest that AI might be a valuable tool in clinical practice for improving early identification and treatment of MS and NMO.