A frontier machine intelligence for human neuroimaging
Summary
- Researchers introduce Neuro-VL, a vision-language model trained on nearly one million neuroimaging studies from routine clinical care.
- Neuro-VL interprets complete imaging examinations, engages in natural language conversations, and generates comprehensive radiology reports.
- The model outperformed frontier general-purpose and specialized medical models in diagnosing, localizing, and describing abnormalities across internal and international datasets.
- Neuro-VL successfully compared serial examinations for disease progression and resisted misleading clinical assumptions better than leading models from OpenAI and Google.
Why it matters
Neuro-VL demonstrates that training vision-language models on large-scale, private clinical health system data can yield specialized medical AI that outperforms general-purpose frontier models in complex diagnostic and longitudinal tasks.