Abstract 12631: Phenome-Wide Prediction of Incident Circulatory Diseases Using Machine Learning-Derived Features From Retinal Imaging
Bibliographic record
Abstract
Purpose: The retina is a window for assessment of cardiovascular health. Here, we performed machine learning-based assessment of retinal vascular density (RVD) from fundus photos, and combined this with retinal layer thicknesses from optical coherence tomography (OCT) imaging, to predict incident circulatory conditions. Methods: We utilized UK Biobank participant retinal fundus images (N=54,813), and OCT images (N=44,823). Deep-learning of fundus photos was used to remove poor-quality images and segment the retinal microvasculature to calculate RVD. Using the Topcon Segmentation algorithm, we quantified retinal layer thicknesses. We performed Cox survival analyses, separately associating RVD and retinal layer thicknesses with 130 circulatory conditions from Phecode ICD-9 and ICD-10 composite phenotypes (median 10 year follow-up), adjusting for age, sex, smoking, ethnicity, height, weight, and spherical equivalent. False Discovery Rate (FDR<0.05) correction was performed to identify significant associations. Results: Participants’ mean age was 56 (SD 8 years), 55% were female, and 56% were non-smokers. Associations were observed for each SD of photoreceptor thinning with incident abdominal aortic aneurysms (HR 1.47, P=6e-6), peripheral vascular disease (HR 1.32, P=4e-5), nonhypertensive heart failure (HR 1.20, P=2.8e-6), myocardial infarction (HR 1.17, P=8e-7), cerebrovascular disease (HR 1.15, P=1.9e-5) and hypertension (HTN) (HR 1.09, P=1e-12). Each SD of lower RVD was associated with incident HTN (HR 1.15, P=3.8e-39) and hypertensive heart disease (HR 2.07, P=6e-4). Further incident heart failure analysis additionally adjusting for RVD, prevalent HTN, and type 2 diabetes identified a persistent association for photoreceptor thinning. Conclusions: Our results indicate that photoreceptor thinning and RVD from retinal imaging can serve as a biomarker for future cardiovascular conditions, independent of other common risk factors.
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 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 | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, 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".