Posted inInteresting Papers
ChatGPT-4 Shows Promise in Interpreting Ophthalmic Images, But Needs Improvement
This study evaluated the new ChatGPT-4's ability to understand ophthalmic images and answer related questions. The chatbot achieved 70% accuracy on multiple-choice questions based on combined image and text data from eye cases, performing better with retinal images (77%) than neuro-ophthalmology cases (58%). It also answered non-image-based questions more accurately (82%) than those requiring image interpretation (65%). The findings suggest ChatGPT-4 has promise as a tool for understanding ophthalmic images, especially in retinal diseases, but needs further refinement, particularly for neuro-ophthalmology and questions heavily reliant on image analysis. The researchers emphasize the need for more research, consideration of ethical implications, and regulatory frameworks before such AI can be reliably used in clinical practice, noting that the study used a public dataset which may not fully represent real-world complexity.