A recent study highlights the potential of ChatGPT-4o in accurately answering questions related to retinal diseases, covering diagnostic criteria, treatment guidelines, and management strategies. The AI's performance was evaluated by vitreoretinal specialists, focusing on its alignment with the American Academy of Ophthalmology’s Preferred Practice Pattern (PPP) guidelines.
The AI generated highly relevant and accurate responses for conditions like diabetic retinopathy and retinal vein occlusion, with an impressive alignment score of 4.91 out of 5. However, it struggled with surgical retinal conditions, particularly retinal tears, showing lower accuracy.
While ChatGPT-4o cannot replace a retina specialist, it shows potential as a tool for education and practice monitoring in ophthalmology. Its performance in medical retina conditions was strong, though surgical cases revealed areas needing improvement.
1. Diagnostic Criteria: What are the key diagnostic criteria for nonproliferative diabetic retinopathy?
ChatGPT-4.0 Response: Nonproliferative diabetic retinopathy (NPDR) is diagnosed based on the presence of microaneurysms, retinal hemorrhages, hard exudates, and macular edema. The severity is classified into mild, moderate, or severe NPDR depending on the extent of these findings and the presence of venous beading or intraretinal microvascular abnormalities (IRMA).
2. Treatment Guidelines: What are the current treatment options for proliferative diabetic retinopathy?
ChatGPT-4.0 Response: Proliferative diabetic retinopathy (PDR) is primarily treated with pan-retinal photocoagulation (PRP) laser therapy. In cases of high-risk PDR or vitreous hemorrhage, vitrectomy may be indicated. Anti-VEGF injections are also used to reduce neovascularization and macular edema.
3. Management Strategies: How should a patient with a history of uveitis be monitored during treatment for axial spondyloarthritis?
ChatGPT-4.0 Response: Patients with a history of uveitis should have regular ophthalmologic evaluations during treatment for axial spondyloarthritis. Monitoring includes checking for signs of active inflammation, visual acuity tests, and slit-lamp examinations to detect early recurrence or complications.
Despite the promising results, the AI’s responses sometimes lacked critical clinical details, and a small percentage included outdated or incorrect information. Furthermore, the readability of AI-generated responses was noted to be high, necessitating advanced education for proper comprehension.
Source: Balas M, et al. Artificial intelligence as a retina specialist: A cross-sectional study. July 17-20, 2024