ELAMI @ MICCAI 2026 · Oral
MICCAI Open Access ↗ · Paper PDF ↗ · arXiv ↗
A medical vision-language model can propose an action, while the local service retains authority to accept or replace it. AuthEval distinguishes proposal quality from selected-action quality. Its logging and evaluation framework scores both under the same declared local criteria where the records permit, and reports the resulting authority gap rather than treating model-only evidence as workflow evidence.
