Using Novel Fundus Image Preprocessing to Improve the Classification of Retinopathy of Prematurity (ROP) Using Deep Learning
Bibliographic record
Abstract
Retinopathy of Prematurity (ROP) is a condition which can affect babies born prematurely. It is a potentially blinding eye disorder as a result of damage to the eye's retina. Tortuosity, presence and intensity of demarcation line are key indicators of ROP. Screening of ROP is a laborious and manual process which requires a trained physician performing a dilated ophthalmology examination. Automated diagnostic methods using Artificial Intelligence (AI) can assist ophthalmologists increase diagnosis accuracy using ROP specific features in the patient's digital retina image. For clinical use, pediatric fundus images captured using digital camera such as RetCam [1] are challenged with image quality that reduces visibility of retinal features and therefore must be pre-processed. This paper presents two improved image pre-processing methods that blend traditional and restoration methods to enhance retinal features, making the images more effective for clinical use and AI for diagnosing ROP. These new methods demonstrated improved CNN accuracy compared to traditional image pre-processing on our ROP RetCam dataset. Using ResNet50 and InceptionResNetV2 classifiers, the best results for ROP classification were as follows: Plus Disease: Accuracy 0.98, sensitivity 0.98, specificity$\mathbf{1. 0}$, precision$\mathbf{1. 0 0}$, F1-score$\mathbf{0. 9 8}$. Stages: Accuracy 0.92, sensitivity 0.72, specificity 0.96, precision 0.79, F1-score 0.76. Zones: Accuracy 0.90, sensitivity 0.85, specificity 0.93, precision 0.85, F1 -score 0.85. These results match or surpass those of comparable studies using limited data. This is the first known application of restoration-based image pre-processing for ROP RetCam images, showing enhanced effectiveness in ROP classification.
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 | 0.001 |
| 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 | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
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".