Noninvasive Coronary Artery Disease Detection Using Retinal Images: A Multimodal Study
Authors' conclusion
Our findings suggest that retinal images and vasculature may provide useful information for CAD detection.
Study context
- Population
- Patients who underwent successful coronary angiography at 4 hospitals, November 2022–September 2024
- Participants
- 383 patients
- Exposure
- Deep learning models using retinal images, clinical indicators or both for coronary artery disease (CAD) detection
- Outcome
- CAD detection (retinal images only)
Methodological detailsmulticentre, 4 hospitals
- Multicentre
- 4 hospitals
Main finding
As reported in the abstract
| Group | AUC | Sensitivity | Specificity |
|---|---|---|---|
| Visual algorithm | 0.80 (95% CI: 0.75-0.85) | 90.5% | 59.6% |
Visual algorithm
AUC0.80 (95% CI: 0.75-0.85)
Sensitivity90.5%
Specificity59.6%
Observational data: associations, not effects.
Further findings
Visual algorithm vs CAD consortium clinical score (test group)
- Higher accuracy with visual algorithm
- 76.2%
- Higher sensitivity with visual algorithm
- 87.4%
Intermediate-risk population (clinical score 15%-85%) (subgroup)
- Visual algorithm outperformed clinical indicators-only model: AUC 0.79 vs 0.75, higher accuracy and sensitivity
CAD detection, multimodal model (retinal images + clinical indicators)
- Best multimodal model AUC
- 0.91 (95% CI: 0.88-0.94)
- Best multimodal model accuracy
- 87.0%
- Best multimodal model sensitivity
- 92.1%
Original publication
Noninvasive Coronary Artery Disease Detection Using Retinal Images: A Multimodal Study. JACC Adv. Published 19 November 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →