Ensemble of deep convolutional neural networks is more accurate and reliable than board-certified ophthalmologists at detecting multiple diseases in retinal fundus photographs
Bibliographic record
Abstract
AIMS: To develop an algorithm to classify multiple retinal pathologies accurately and reliably from fundus photographs and to validate its performance against human experts. METHODS: We trained a deep convolutional ensemble (DCE), an ensemble of five convolutional neural networks (CNNs), to classify retinal fundus photographs into diabetic retinopathy (DR), glaucoma, age-related macular degeneration (AMD) and normal eyes. The CNN architecture was based on the InceptionV3 model, and initial weights were pretrained on the ImageNet dataset. We used 43 055 fundus images from 12 public datasets. Five trained ensembles were then tested on an 'unseen' set of 100 images. Seven board-certified ophthalmologists were asked to classify these test images. RESULTS: Board-certified ophthalmologists achieved a mean accuracy of 72.7% over all classes, while the DCE achieved a mean accuracy of 79.2% (p=0.03). The DCE had a statistically significant higher mean F1-score for DR classification compared with the ophthalmologists (76.8% vs 57.5%; p=0.01) and greater but statistically non-significant mean F1-scores for glaucoma (83.9% vs 75.7%; p=0.10), AMD (85.9% vs 85.2%; p=0.69) and normal eyes (73.0% vs 70.5%; p=0.39). The DCE had a greater mean agreement between accuracy and confident of 81.6% vs 70.3% (p<0.001). DISCUSSION: We developed a deep learning model and found that it could more accurately and reliably classify four categories of fundus images compared with board-certified ophthalmologists. This work provides proof-of-principle that an algorithm is capable of accurate and reliable recognition of multiple retinal diseases using only fundus photographs.
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 | 0.003 | 0.006 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.000 | 0.001 |
| 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".