A Novel Foundation Model-Based Framework for Multimodal Retinal Age Prediction
Bibliographic record
Abstract
The retinal age gap (RAG; the difference between the retina's biological and chronological age) has recently gained increased attention as a potential image-based, non-invasive, and accessible biomarker for a broad spectrum of ocular and non-ocular diseases. Traditionally, machine learning predictions of biological retinal age utilize convolutional neural network (CNN) architectures and data from color fundus photography (CFP). Despite being previously unexplored, the multimodal fusion of two-dimensional CFP with three-dimensional optical coherence tomography (OCT) data has significant potential to enhance retinal age prediction accuracy and the diagnostic utility of the RAG biomarker. Therefore, this work presents a novel foundation model-based framework for multimodal retinal age prediction. Technology or Method: Feature representations from CFP and OCT images were extracted using RETFound, a powerful foundation model for retinal image analysis. These representations were then combined using an innovative fusion strategy to train a lightweight linear regression head model for predicting retinal age. Training and evaluation of the developed multimodal retinal age prediction model was achieved using retinal images from over 80,000 participants in the UK Biobank. RESULTS: The developed multimodal model sets a new benchmark in retinal age prediction (mean absolute error of 2.75 years), outperforming traditional CNN and single-modality approaches. Additionally, multimodal RAG values demonstrated superior performance in classifying patients with diabetes mellitus type 1, multiple sclerosis, and chronic kidney disease, highlighting the clinical relevance of the proposed multimodal approach for non-ocular disease detection. CONCLUSIONS: This work demonstrates that multimodal fusion of CFP and OCT significantly improves retinal age prediction and subsequent RAG-based analyses. By leveraging foundation models and multimodal retinal imaging, the proposed approach enhances disease classification accuracy and demonstrates the potential of integrating the RAG into clinical workflows as a scalable, non-invasive screening tool. SIGNIFICANCE: The findings underscore the potential of multimodal retinal imaging to transform RAG into a clinically relevant and highly accessible biomarker for disease detection.
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) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 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".