Analyzing clinical variables indicative of uveal melanoma to determine how they affect decisions made by an artificial intelligence classifier
Bibliographic record
Abstract
OBJECTIVE: The "black box" nature of many artificial intelligence (AI) models has limited their adoption in real-world ophthalmologic practices. Our lab developed an AI model for detecting the presence of a choroidal melanocytic lesion (CML) in colour fundus images. The purpose of this article is to investigate whether there are known clinical features of CMLs that are associated with false-negative (FN) classifications from the model to aid in validation and increase its interpretability. METHODS: A retrospective cohort study of CML patients was performed. A total of 388 fundus images from 194 patients with (n = 194) and without (n = 194) CMLs collected through routine clinical assessment were used to train an AI model. The model's classification (lesion present/lesion absent) of the images with CMLs, as well as CML characteristics, demographics, and risk factors for uveal melanoma (UM) were extracted. Logistic regression models were used to test for associations between the FN classifications and these characteristics. RESULTS: The AI model returned 150 true-positive classifications and 44 FN classifications (23%) for CML eyes. Thinner lesions were more likely to be missed by the model (p = 0.026), resulting in a FN classification. The presence of imaging risk factors for UM was not shown to have any statistically significant relationships with a FN classification. CONCLUSIONS: The results from this study demonstrate that the FN classifications for CML fundus image classifications from our AI model are not associated with the presence of imaging risk factors for UM but are influenced by thinness of the lesion.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.021 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 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 source (direct Gemma or distilled Codex), 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".