Beyond the Surface: Assessing GPT-4's Accuracy in Detecting Melanoma and Suspicious Skin Lesions From Dermoscopic Images
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
Introduction: Self-examinations for skin cancer detection are limited by sensitivity. ChatGPT-4 has image recognition capabilities that can be a useful adjunct for screening cancers and tele-health applications. This study investigated the efficacy of ChatGPT-4 in identifying skin lesions. Methods: Dermoscopic images were retrospectively selected from the PH 2 dataset, categorized by clinical diagnosis, and uploaded to ChatGPT-4 with a predesigned prompt. Responses were compared against clinical diagnoses. Confidence intervals were calculated using the bootstrap method assessing precision and significance was calculated using McNemar's test. Analyses were performed using Jupyter Notebook and Python. Results: The GPT-4 model showed moderate performance in melanoma detection with 68.5% accuracy, 52.5% sensitivity, and 72.5% specificity, significantly differing from the clinical standard ( P = .002). For suspicious lesion detection, it performed better with 68.0% accuracy, 78.0% precision, and 70.0% F-measure, still not closely matching clinical diagnosis for atypical nevi and melanoma ( P = .0169). Conclusion: The statistical difference between ChatGPT-4 diagnosis of melanoma and suspicious lesions compared with clinical diagnoses and other AI models suggests the need for improvement in ChatGPT-4 algorithms. This study's limitations included the use of a secondary care database with a higher melanoma incidence, high-quality dermoscopic images that limit generalizability, a small sample size lacking diversity, and the need for larger datasets to validate findings in broader contexts.
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 | 0.004 |
| 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