AI Demonstrates Excellent Understanding and Analysis of Eye Problems

Large language models (LLMs) are showing promise in ophthalmology, as they are nearing expert-level knowledge and reasoning skills, a recent study suggests. The study, published in PLOS Digital Health, evaluated the clinical potential of advanced LLMs in the field of ophthalmology.

Researchers, led by Arun James Thirunavukarasu from the University of Oxford, compared the responses of different LLMs – including GPT-3.5, GPT-4, PaLM 2, and LLaMA – to those of expert ophthalmologists and doctors in training in response to 87 questions. The findings revealed that GPT-4 outperformed the other models, with a 69 percent accuracy rate, compared to 48 percent for GPT-3.5, 32 percent for LLaMA, and 56 percent for PaLM 2. GPT-4’s performance also compared favorably with expert ophthalmologists.

The study highlighted that while there were differences in knowledge and reasoning between the LLMs and human doctors, there was overall consistency in responses. Grading ophthalmologists indicated a preference for GPT-4 responses due to higher accuracy and relevance.

The authors of the study believe that LLMs could be valuable in providing eye-related advice in areas where access to healthcare professionals is limited. However, they emphasize the need for further research to explore the potential clinical applications of these models.

One author involved in the study disclosed a patent related to a deep learning system for detecting retinal disease.

Overall, the study underscores the potential of LLMs in advancing the field of ophthalmology and the importance of continued research in this area.