Open LLM-based actionable incidental finding extraction from [18F]fluorodeoxyglucose PET-CT radiology reports
We find that a QLoRA-adapted LLM utilising CoT prompting successfully extracts AIF information at both document- and sentence-level from both internal and external PET-CT reports. We believe this model can assist with short-term clinical challenges like clinical alerts and reminders, and long-term tasks like investigating comorbidities.
Authors
- Yulan He (ORCID: https://orcid.org/0000-0003-3948-5845)
- Gary Cook (ORCID: https://orcid.org/0000-0002-8732-8134)
- Anna Barnes (ORCID: https://orcid.org/0000-0002-6902-214X)
- Sugama Chicklore (ORCID: https://orcid.org/0000-0002-5250-2139)
- Stephen Barlow (ORCID: https://orcid.org/0000-0001-9059-9974)
- Sébastien Ourselin (ORCID: https://orcid.org/0000-0002-5694-5340)
- Thomas Wagner (ORCID: https://orcid.org/0000-0002-3867-6829)
Institutions
- St Thomas' Hospital (GB)
- King's College London (GB)
- The Royal Free Hospital (GB)
- University of Warwick (GB)
- The Alan Turing Institute (GB)
- University College London (GB)
Publication Details
- Journal
- Frontiers in Digital Health
- Published
- 2025-11-14
- DOI
- https://doi.org/10.3389/fdgth.2025.1702082
- Primary Topic
- Radiology practices and education
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Cancer Research UK
- King's College London
- Engineering and Physical Sciences Research Council
- Centre For Medical Engineering, King’s College London