Impact of hospital-specific domain adaptation on BERT-based models to classify neuroradiology reports
Question BERT-based models can classify radiology reports, but it is unclear if there is any incremental benefit from additional hospital-specific domain adaptation. Findings Hospital-specific domain adaptation resulted in the highest BERT-based model accuracies and performance scaled log-linearly with the number of reports. Clinical relevance BERT-based models after hospital-specific domain adaptation achieve the best classification results provided sufficient high-quality training labels. When labelled data is scarce, LLMs such as Llama-3.0 70B are a viable alternative provided there are sufficient computational resources.
Authors
- James H. Cole (ORCID: https://orcid.org/0000-0003-1908-5588)
- Thomas C. Booth (ORCID: https://orcid.org/0000-0003-0984-3998)
- David Michael Wood (ORCID: https://orcid.org/0000-0002-7826-7237)
- Marc Modat (ORCID: https://orcid.org/0000-0002-5277-8530)
- Ayisha Al Busaidi (ORCID: https://orcid.org/0000-0002-3539-3743)
- Peter David Sasieni (ORCID: https://orcid.org/0000-0003-1509-8744)
- Jeremy Lynch (ORCID: https://orcid.org/0000-0002-1476-0612)
- Emily Guilhem (ORCID: https://orcid.org/0000-0001-9727-1090)
- Siddharth Agarwal (ORCID: https://orcid.org/0000-0002-5581-5464)
- Sina Kafiabadi (ORCID: https://orcid.org/0000-0003-2755-8875)
- Benjamin Murray (ORCID: https://orcid.org/0000-0002-2413-923X)
- Gareth J. Barker (ORCID: https://orcid.org/0000-0002-5214-7421)
- Matthew Townend (ORCID: https://orcid.org/0000-0002-4282-5524)
- Sébastien Ourselin (ORCID: https://orcid.org/0000-0002-5694-5340)
- Yiran Wei (ORCID: https://orcid.org/0000-0003-1791-6050)
- Asif Mazumder (ORCID: https://orcid.org/0000-0002-1451-8770)
Institutions
- King's College London (GB)
- Guy's and St Thomas' NHS Foundation Trust (GB)
- King's College Hospital NHS Foundation Trust (GB)
- King's College School (GB)
- University College London (GB)
Publication Details
- Journal
- European Radiology
- Published
- 2025-03-17
- DOI
- https://doi.org/10.1007/s00330-025-11500-9
- Citations
- 2
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- article
- Field-Weighted Citation Impact
- 1.31
Funders
- Wellcome Trust
- King's College London
- Medical Research Council
- Engineering and Physical Sciences Research Council
- Centre For Medical Engineering, King’s College London