OpenAI-based GPT-4 can match, and in some cases surpass, human ophthalmologists in diagnosing and treating patients with glaucoma and retinal disease, according to research.
A study published in JAMA Ophthalmology suggests that advanced tools such as artificial intelligence (AI)-based large language models (LLMs), which are trained on vast amounts of data, text and images, could play an important role in providing decision-making. providing support to ophthalmologists in the diagnosis and treatment of cases involving glaucoma and retinal disorders that afflict millions of patients.
“The performance of GPT-4 in our study was quite impressive,” said lead author Andy Huang, an ophthalmology resident at the New York Eye and Ear Infirmary of Mount Sinai Hospital, USA.
“We realized the huge potential of this AI system from the moment we started testing it, and were fascinated to observe that GPT-4 could not only assist, but in some cases match or surpass the expertise of experienced ophthalmology specialists,” added Huang.
The researchers compared the knowledge of the GPT-4 (Generative Preschool Model 4) with 12 attending specialists and three senior fellows from the Department of Ophthalmology at the Icahn School of Medicine at Mount Sinai.
A core set of 20 questions (10 each for glaucoma and retina) was randomly selected from the American Academy of Ophthalmology list of frequently asked patient questions along with 20 de-identified patient cases collected from Mount Sinai-affiliated eye clinics.
Also read: Roche’s Xolair reduces reactions to severe food allergies in study
The results showed that the AI matched or outperformed human specialists in both the accuracy and completeness of its medical advice and assessments.
More specifically, the AI showed superior performance in responding to glaucoma questions and case management advice, reflecting a more balanced result on retinal questions, where AI matched humans in accuracy but outperformed them in completeness.
“The AI was particularly surprising in its expertise in dealing with both glaucoma cases and retinal patients, matching the accuracy and completeness of diagnoses and treatment suggestions made by human doctors in a clinical note format,” said Louis R. Pasquale, vice chair for ophthalmology research for Eye Clinics.
The findings “could serve as a reliable aid to eye specialists by providing them with diagnostic support and potentially easing their workload, especially in complex cases or in areas with high patient volume,” said Dr. Huang.
“For patients, the integration of artificial intelligence into routine ophthalmology practice could lead to faster access to expert advice coupled with more informed decisions about the management of their treatment.”
