Reproduce “CLARIFID: Improving Radiology Report Generation by Reinforcing Clinically Accurate Impressions and Enforcing Detailed Findings” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kyeongkyu Lee, Seonghwan Yoon, Hongki Lim
A new framework, CLARIFID, generates radiology reports by mirroring the two-step radiologist workflow, leveraging multi-view chest X-rays, and using reinforcement learning optimized for Impression section accuracy. The approach demonstrates superior clinical efficacy on the MIMIC-CXR dataset and robust generalization on IU X-ray, enhancing diagnostic reliability.
Key insights
- Optimizing for the clinical accuracy of the Impression section through reinforcement learning effectively guides the model to produce higher-quality, factually aligned Findings.
- Explicitly mirroring the radiologist's two-step workflow, where detailed Findings precede the Impression, improves the clinical coherence and reliability of generated reports.
- Leveraging multi-view chest X-rays through a flexible fusion module provides essential complementary information, substantially enhancing diagnostic accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.