Fundus photograph interpretation of common retinal disorders by artificial intelligence chatbots
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Purpose While previous studies have examined the ability of artificial intelligence (AI) chatbots to interpret optical coherence tomography scans, their performance in interpreting fundus photographs of retinal disorders without text-based context remains unexplored. This study aims to evaluate the ability of three widely used AI chatbots to accurately diagnose common retinal disorders from fundus photographs in the absence of text-based context. Design Cross-section study. Methods We prompted ChatGPT-4, Gemini, and Copilot, with a set of 50 fundus photographs from the American Society of Retina Specialists Retina Image Bank® in March 2024, comprising of age-related macular degeneration, diabetic retinopathy, epiretinal membrane, retinal vein occlusion, and retinal detachment. Chatbots were re-prompted four times using the same images throughout June 2024. The primary endpoint was the proportion of each chatbot’s correct diagnoses. No text-based guidance was provided. Results In March 2024, Gemini provided a correct diagnosis for 17 (34%, 95% CI: 21%-49%) fundus images, ChatGPT-4 for 16 (32%, 95% CI: 20%-47%), and Copilot for 9 (18%, 95% CI: 9%-31%) (p>0.05). In June 2024, Gemini provided a correct diagnosis for 122 (61%, 95% CI: 53%-67%) images, ChatGPT-4 for 101 (51%, 95% CI: 43%-58%), and Copilot for 57 (29%, 95% CI: 22%-35%). Conclusion No AI chatbot use in this study was sufficiently accurate for the diagnosis of common retinal disorders from fundus photographs. AI chatbots should not currently be utilized in any clinical setting involving fundus images, given concerns for accuracy and bioethical considerations.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it