EliSeg Framework for Report-Grounded Medical Abnormality Segmentation
August 10, 2026
EliSeg uses an actor-verify-revise framework to segment clinical abnormalities directly from unfiltered radiology reports. It resolves ambiguity in finding-to-mask correspondence by using a grammar-constrained actor and an independent text-only verifier to reconstruct eligible finding inventories.
HOW THIS AFFECTS YOU
●
researcherThis approach removes the need for spatial prompts by automating target construction from unstructured text.
●
healthThis improves the reliability of automated medical imaging analysis by grounding segmentation in clinical report findings.