Carving Radiology Reports at Their Evidential Joints: Anatomical Region Sets as Generation Units

Key points

• Organizing radiology reports into independently generated units offers a way to generate fine-grained clinical content and bind it to visual evidence.1
• However, existing methods typically adopt generation-unit structures not directly specified in natural reports or their annotations, introducing additional modeling compromises.1
• RadCarve's planner uses composition-specialized query banks to jointly predict unit existence and region composition from the image, iteratively refining both with regional evidence aggregated according to current membership predictions.1
• Compared with strong baselines, RadCarve improves CXRFEScore by 9.4% in unit generation given region sets and achieves 2.4 times the end-to-end binding F1.1

Sources (1)

• [1]Carving Radiology Reports at Their Evidential Joints: Anatomical Region Sets as Generation Units
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 21, 03:55 AM

  We propose RadCarve, which instead defines generation units directly from the anatomical region sets in existing statement-region correspondences, unifying unit formation with visual evidence selection.

  Organizing radiology reports into independently generated units offers a way to generate fine-grained clinical content and bind it to visual evidence.

Extractive summary: sentences quoted from the sources.