NeuroVFM AI: Outperforms GPT-5 in Hospital Brain Scan Trial

A new specialist AI model, NeuroVFM, has outperformed GPT-5 in a real-world hospital trial. NeuroVFM beat GPT-5 by 21.4 percentage points in critical-findings triage. This study took place over one week in January 2026 across an entire health system. Here's the thing: general AI models like GPT-5 are largely "blind" to brain imaging. This is because brain scans, with identifiable facial features, are not typically found in public datasets used to train these AIs. NeuroVFM was trained on 5.24 million clinical MRI and CT scans from Michigan Medicine. This demonstrates that a hospital's own archived scans can be used to train specialized medical AI. This approach works without needing hand-labeled data or curated datasets. The bottom line: this shows a new way to develop medical AI using a hospital's own data, potentially improving diagnostic tools in healthcare.