Automated identification of fleck lesions in Stargardt disease using deep learning enhances lesion detection sensitivity and enables morphometric analysis of flecks
Bibliographic record
Abstract
Purpose To classify fleck lesions and assess artificial intelligence (AI) in identifying flecks in Stargardt disease (STGD). Methods A retrospective study of 170 eyes from 85 consecutive patients with confirmed STGD. Fundus autofluorescence images were extracted, and flecks were manually outlined. A deep learning model was trained, and a hold-out testing subset was used to compare with manually identified flecks and for graders to assess. Flecks were clustered using K-means clustering. Results Of the 85 subjects, 45 were female, and the median age was 37 years (IQR 25–59). A subset of subjects (n=41) had clearly identifiable fleck lesions, and an AI was successfully trained to identify these lesions (average Dice score of 0.53, n=18). The AI segmentation had smaller (0.018 compared with 0.034 mm2, p<0.001) but more numerous flecks (75.5 per retina compared with 40.0, p<0.001), but the total size of flecks was not different. The AI model had higher sensitivity to detect flecks but resulted in more false positives. There were two clusters of flecks based on morphology: broadly, one cluster of small round flecks and another of large amorphous flecks. The per cent frequency of small round flecks negatively correlated with subject age (r=−0.31, p<0.005). Conclusions AI-based detection of flecks shows greater sensitivity than human graders but with a higher false-positive rate. With further optimisation to address current shortcomings, this approach could be used to prescreen subjects for clinical research. The feasibility and utility of quantifying fleck morphology in conjunction with AI-based segmentation as a biomarker of progression require further study.
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) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 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 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| 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".