Performance of ChatGPT in French language analysis of multimodal retinal cases
Bibliographic record
Abstract
PURPOSE: Prior literature has suggested a reduced performance of large language models (LLMs) in non-English analyses, including Arabic and French. However, there are no current studies testing the multimodal performance of ChatGPT in French ophthalmology cases, and comparing this to the results observed in the English literature. We compared the performance of ChatGPT-4 in French and English on open-ended prompts using multimodal input data from retinal cases. METHODS: GPT-4 was prompted in English and French using a public dataset containing 67 retinal cases from the ophthalmology education website OCTCases.com. The clinical case and accompanying ophthalmic images comprised the prompt, along with the open-ended question: "What is the most likely diagnosis?" Systematic prompting was used to identify and compare relevant factor(s) contributing to correct and incorrect responses. Diagnostic accuracy was the primary outcome, defined as the proportion of correctly diagnosed cases in French and English. Diagnoses were compared with the answer key on OCTCases to confirm correct or incorrect responses. Clinically relevant factors reported by the LLM as contributory to its decision-making were secondary endpoints. RESULTS: , P=0.36). Imaging findings were reported as most influential for correct diagnosis in English (37.5%) and French (42.1%) (P=0.76). In incorrectly diagnosed cases, imaging findings were primarily implicated in English (35.6%) and French (33.3%) (P=0.81). In incorrectly diagnosed cases, the differential diagnosis list contained the correct diagnosis in 39.5% of English cases and 41.7% of French cases (P=0.83). CONCLUSION: Our results suggest that GPT-4 performed similarly in English and French on all quantitative performance metrics measured. Ophthalmic images were identified in both languages as critical for correct diagnosis. Future research should assess LLM comprehension through the clarity, grammatical, cultural, and idiomatic accuracy of its responses.
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.
How this classification was reachedexpand
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 | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".