Related Experiment Video
Updated: May 5, 2026
07:40
Experimental Autoimmune Uveitis: An Intraocular Inflammatory Mouse Model
Published on: January 12, 2022
4.7K
Artificial intelligence derived large language model in decision-making process in uveitis.
Inès Schumacher1, Virginie Manuela Marie Bühler1, Damian Jaggi1
1Department of Ophthalmology, Inselspital, University Hospital of Bern, Bern, Switzerland.
International Journal of Retina and Vitreous
|September 11, 2024
Summary
Large language models (LLMs) show potential in uveitis care but require extensive training. While LLMs can assist in ophthalmology, their generated references are frequently inaccurate, necessitating careful validation.
Area of Science:
- Ophthalmology
- Artificial Intelligence in Medicine
Background:
- Uveitis involves intraocular inflammatory diseases.
- Large language models (LLMs) like ChatGPT are increasingly relevant in medicine.
Purpose of the Study:
- To explore the strengths and weaknesses of LLM applicability in the subfield of uveitis.
- To assess the accuracy and sufficiency of LLM-generated responses for clinical uveitis cases.
Main Methods:
- Clinically relevant questions on uveitis cases were posed to an LLM three times.
- LLM answers were classified for accuracy and sufficiency.
- Statistical analysis included descriptive statistics and reliability tests (Cohen's and Fleiss' kappa).
- Generated references were verified for accuracy in medical databases.
Main Results:
- LLM responses showed moderate agreement across attempts (average Cohen's kappa = 0.4577).
- A significant portion of generated references were inaccurate (42.3% unlocatable, 15.4% misinterpreted/incorrectly cited).
- Only 42.3% of references were accurate and correctly cited.
Conclusions:
- LLMs demonstrate significant potential for uveitis management and ophthalmology.
- Rigorous training and testing are essential before implementing LLMs in specific medical tasks.
- LLM-generated references require careful validation due to frequent inaccuracies.
