A comparative performance analysis of regular expressions and a large language model-based approach to extract the BI-RADS score from radiological reports
For structured, standardized data like BI-RADS, traditional NLP techniques seem to be superior, though future work should explore hybrid approaches combining Regex precision for standardized elements with LLM contextual understanding for more complex information extraction tasks.
Authors
- Knud Nairz (ORCID: https://orcid.org/0000-0003-3337-110X)
- Nikola Čihorić (ORCID: https://orcid.org/0000-0003-1534-8520)
- Fabio Dennstädt (ORCID: https://orcid.org/0000-0002-5374-8720)
- Hendrik von Tengg‐Kobligk (ORCID: https://orcid.org/0000-0003-3207-3223)
- Janna Hastings (ORCID: https://orcid.org/0000-0002-3469-4923)
- Harald Bonél (ORCID: https://orcid.org/0000-0002-3033-6248)
- Grazia M. Cereghetti (ORCID: https://orcid.org/0009-0005-2228-8169)
- Daniel M. Aebersold (ORCID: https://orcid.org/0000-0002-9493-3834)
- Irina Filchenko (ORCID: https://orcid.org/0000-0002-9486-1017)
- Luc Lerch
- Florian Dammann
- Roberto Gaio
- Max Schmerder (ORCID: https://orcid.org/0009-0009-3663-4412)
Institutions
- University of Bern (CH)
- SIB Swiss Institute of Bioinformatics (CH)
- ZHAW Zurich University of Applied Sciences (CH)
- University of St.Gallen (CH)
- University Hospital of Bern (CH)
- Stiftung Lindenhof Bern (CH)
- Lindenhofspital (CH)
- University of Applied Sciences St. Gallen (CH)
Publication Details
- Journal
- JAMIA Open
- Published
- 2025-11-03
- DOI
- https://doi.org/10.1093/jamiaopen/ooaf128
- Primary Topic
- Topic Modeling
- Type
- article
- Field-Weighted Citation Impact
- 0.00