Automated Optic Disc Tilt Classification in Fundus Photography: Segmentation and Elliptical Ratio Across External Clinical Validation (Preprint)
Bibliographic record
Abstract
Abstract Background Optic disc tilt is a morphological change in myopic eyes that complicates clinical interpretation and artificial intelligence (AI)–based analysis of fundus images. Accurate detection of optic disc tilt is necessary to avoid misinterpretation of disc morphology and enhance diagnostic reliability across different disease types. Objective This study developed and externally validated an end-to-end AI-based pipeline for optic disc segmentation and quantitative tilt classification in color fundus photographs (CFPs), offering an objective alternative to manual segmentation and subjective clinical assessments. Methods We trained a nnU-Net–based optic disc segmentation model on the Standardized Multi-Channel Dataset for Glaucoma (SMDG; n=3103 CFPs) and externally validated it on the Samsung Medical Center (SMC) dataset (n=2448 CFPs from n=1370 patients). Model generalizability was assessed using both a fixed 80:20 random split and 5-fold cross-validation. Tilt was classified using the ratio of the long-axis diameter to the short-axis diameter, with a ratio of ≥1.3 indicating tilt. Segmentation performance was evaluated using the Dice similarity coefficient, intersection over union, and pixel accuracy on the SMDG dataset and using the clinical acceptance rate determined by 2 independent ophthalmologists on the external SMC dataset. Results Using the SMDG dataset, nnU-Net achieved consistently high performance, with mean Dice similarity coefficients of 0.956 (SD 0.042) across 5-fold cross-validation and 0.961 (SD 0.055) for the best-performing single-fold model across 8 datasets. On the SMC dataset, 2 independent expert reviews yielded mean clinical acceptance rates of 98.61% and 98.86% across disease types, with acceptance rates ranging from 81.63% and 93.88% for edema to 99.59% and 99.17% for pallor, respectively. Tilt was detected in 7.5% (186/2448) of images, with rates of 9.7% (118/1215) for normal images, 3.9% (35/894) for glaucoma, 7.8% (19/241) for pallor, and 14.2% (14/98) for edema. Segmentation errors were observed in 1.39% (34/2448) and 1.14% (28/2448) of cases by the 2 reviewers, mainly due to edema-related swelling, peripapillary atrophy, and vessel confusion. Conclusions Our pipeline provides objective and reproducible detection of optic disc tilt in CFPs, with strong generalizability to clinical images. By replacing manual segmentation and subjective assessments, the pipeline supports tilt-aware AI diagnostics and scalable screening for myopia-related conditions, with future refinements needed to address edema-related challenges.
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 | 0.005 | 0.012 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.002 |
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".