Turn on cookies to sign in
Signing in saves your progress to your Curie account. We can only do that with cookies on — turn them on to continue.
AI tools for multiple sclerosis are advancing fast, with deep learning models pulling richer data from routine MRI scans than ever before. But at MSToronto 2026, experts warned that most published MS AI models have only been tested at the centers that built them. The gap between research promise and real-world clinical use remains wide.
AI is making waves in multiple sclerosis (MS) research, with new deep learning tools that can segment lesions, flag chronic inflammation markers, and extract prognostic signals from routine MRI scans. At MSToronto 2026, researchers showcased impressive advances — including a model that predicted disability worsening from standard FLAIR scans better than volume measures alone, validated in an external cohort of 504 patients. Clinicians also highlighted potential uses beyond imaging, from organizing treatment guidelines in rare neuroimmunologic diseases to screening drug libraries for repurposing candidates.
But the enthusiasm came with a reality check. Experts noted that most published MS AI models sit at the lowest rungs of the validation ladder — tested only within the center that built them, on small datasets, and rarely with full code or data sharing. The labeling problems run deep too: 40% of new lesions were flagged by just 2 or fewer of 4 experts, and disability progression has been defined 48 different ways across registries.
Key Takeaways:
Why it matters: AI holds real promise for improving MS diagnosis and monitoring, but without rigorous external validation and shared datasets, clinicians risk adopting tools that don't generalize beyond the research lab — potentially misleading care decisions for millions of patients.
