Assessing AI and Neurologist Diagnostic Reasoning Against Neuropathological Ground Truth
- Open access
DIAGNO, an AI system, achieves 67% top-3 accuracy, surpassing neurologists' 63%, in diagnosing complex neurological disorders confirmed by neuropathology.
- Why it matters: Accurate diagnosis of neurological diseases is difficult due to overlapping symptoms and diverse presentations, and current benchmarks lack biological ground truth, limiting AI evaluation.
- What they did: Researchers developed NeuroBench with 203 autopsy-confirmed cases and evaluated DIAGNO using retrieval-augmented generation, comparing its performance to neurologists and baseline models across multiple accuracy metrics.
- The result: DIAGNO matched or outperformed neurologists and baseline models, demonstrated high diagnostic agreement, and received positive clinician ratings, indicating its potential as a reliable clinical decision-support tool.