A diffusion model-based self-explainable generative classifier for retinal image analysis
Bibliographic record
Abstract
Vision-influencing ocular conditions, such as diabetic retinopathy (DR), age-related macular degeneration (AMD), cataracts, glaucoma, and degenerative myopia, affect millions of patients globally. Early diagnosis and treatment can reduce the impact of those conditions, which can be identified in color fundus photographs (CFPs). However, manual interpretation of CFPs is a time-consuming process that requires visual assessment by an experienced ophthalmologist. Therefore, several deep learning models have been proposed to support automated retinal disease detection / classification from CFPs. While those discriminative classification models usually achieve a high performance, their black-box nature generally creates challenges when trying to understand their decision-making process, which is crucial to establish trust in their results in a clinical setup. To alleviate this problem, this paper introduces the first fully self-explainable generative classifier for the computer-aided diagnosis of multiple ocular conditions from CFPs. More precisely, we propose a generative diffusion model, which we then repurpose as a classifier. In contrast to discriminative models, our generative classifier is inherently self-explainable due to its ability to synthesize so-called counterfactual images that can be used to visually highlight alternate classification outcomes. We train and evaluate the proposed method on more than 25,000 images encompassing five different ocular conditions, including age-related macular degeneration and diabetic retinopathy, and achieve a competitive classification accuracy of 97.38%. Moreover, the counterfactual images generated by our model visually confirm that it has learned the hallmark features of the ocular conditions included in the data. We believe that our self-explainable generative classifier is an important step towards trustworthy AI in ophthalmology as it removes the explainability challenges surrounding standard discriminative classifiers.
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 | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| 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".