Automated detection of diabetic retinopathy in digital retinal images: a tool for diabetic retinopathy screening
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
AIMS: To develop a system to detect automatically features of diabetic retinopathy in colour digital retinal images and to evaluate its potential in diabetic retinopathy screening. METHODS: Macular centred 45 degrees colour retinal images from 1273 patients in an inner city diabetic retinopathy screening programme. A system was used involving pre-processing to standardize colour and enhance contrast, segmentation to reveal possible lesions and classification of lesions using an artificial neural network. The system was trained using a subset of images from 500 patients and evaluated by comparing its performance with a human grader on a test set of images from 773 patients. RESULTS: Maximum sensitivity for detection of any retinopathy on a per patient basis was 95.1%, accompanied by specificity of 46.3%. Specificity could be increased as far as 78.9% but was accompanied by a fall in sensitivity to 70.8%. At a setting with 94.8% sensitivity and 52.8% specificity, no cases of sight-threatening retinopathy were missed (retinopathy warranting immediate ophthalmology referral or re-examination sooner than 1 year by National Institute for Clinical Excellence criteria). If the system was implemented at 94.8% sensitivity setting over half the images with no retinopathy would be correctly identified, reducing the need for a human grader to examine images in 1/3 of patients. CONCLUSION: This system could be used when screening for diabetic retinopathy. At 94.8% sensitivity setting the number of normal images requiring examination by a human grader could be halved.
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 | 0.002 | 0.010 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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