Researchers release ChestPheNoT for auditable radiology report analysis
Researchers introduced ChestPheNoT, a compact 0.5–3B language model designed to extract structured medical findings, status labels (present/absent/uncertain), and verbatim evidence spans from radiology reports. The model avoids external APIs by running locally, addressing concerns about data leaving institutional networks. It uses hybrid silver supervision (CheXbert + 72B models) and GRPO…
Key points
- ChestPheNoT extracts findings, status labels, and evidence spans from radiology reports locally, not via APIs
- Outperforms CheXbert on cross-institution detection by +2.0 F1 and matches larger models on detection tasks
- Achieves 99% evidence span locatability and 47.5 auditable-F1, surpassing Qwen2.5-7B by 7.6 points
ChestPheNoT achieves 99% evidence span locatability and an auditable-F1 of 47.5, surpassing Qwen2.5-7B by 7.6 points. The team claims it enables deployable, explainable radiology analytics without relying on cloud-based inference. Code and prompts will be open-sourced on GitHub.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Anthropic's Claude begins 'dreaming' to reorganize memory in tests · 1 src
- John Langford quits Microsoft, declares AI Frontiers over · 1 src
- Anthropic releases study on Claude usage in scientific research · 3 src
- MIT and BASF release CheMeleon for low-data molecular prediction · 2 src
- Anthropic tests Claude agents in EVE Online trading and sabotage detection · 1 src
Comments
via GitHub Discussions