Accuracy of large language models in generating differential diagnosis from clinical presentation and imaging findings in pediatric cases
Commercial LLMs performed similarly on pediatric radiology cases in providing top 1 accuracy and top 3 differential accuracy when only a text-based image description was used. Adding clinical presentation significantly improved top 1 accuracy for ChatGPT-4 V and Claude 3.5 Sonnet, with Claude showing the largest improvement. Claude 3.5 Sonnet outperformed both ChatGPT-4 V and Gemini 1.5 Pro in top 1 accuracy when both image and clinical data were provided. No significant differences were found in top 3 differential accuracy across models in any condition.
Authors
- Roozbeh Houshyar (ORCID: https://orcid.org/0000-0003-4263-5533)
- Shawn Sun (ORCID: https://orcid.org/0000-0002-6170-1949)
- Kasha Chen
- Nathan W. Chan (ORCID: https://orcid.org/0000-0003-0909-8393)
- Jinho Jung (ORCID: https://orcid.org/0000-0002-9786-5507)
- Erwin Ho (ORCID: https://orcid.org/0009-0001-3773-8856)
- Bryant Tran (ORCID: https://orcid.org/0000-0002-1231-5611)
- Michael Phillipi (ORCID: https://orcid.org/0009-0007-2657-2989)
Institutions
- University of California, Irvine (US)
- University of California, Irvine Medical Center (US)
- Stony Brook University (US)
Publication Details
- Journal
- Pediatric Radiology
- Published
- 2025-07-12
- DOI
- https://doi.org/10.1007/s00247-025-06317-z
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- article
- Field-Weighted Citation Impact
- 0.00