AI framework for multidisease detection via retinal imaging
Study context
- Population
- Colour fundus photographs from community-based and hospital-based cohorts; six external cohorts from resource-limited and high-resource settings
- Participants
- 107,730 colour fundus photographs
- Comparison
- Reti-Pioneer retinal imaging framework for multidisease detection
- Outcome
- AUC on internal test data
Methodological detailsmultitask deep learning framework, community-based and hospital-based cohorts, external validation in six cohorts, silent trial, clinical pilot
- Multitask deep learning framework
- Community-based and hospital-based cohorts
- External validation in six cohorts
- Silent trial
- Clinical pilot
Performance
As reported in the abstract
- AUC for T2D
- 0.833 (95% CI 0.810-0.856)
- AUC for gout
- 0.832 (0.799-0.866)
- AUC for osteoporosis
- 0.787 (0.742-0.833)
- AUC for hypertension
- 0.740 (0.726-0.755)
- AUC for hyperlipidemia
- 0.736 (0.721-0.751)
- AUC for thyroid disease
- 0.699 (0.667-0.730)
Further findings
External validation
- The framework generalized well to six external cohorts from both resource-limited and high-resource settings
Screening time per case in primary care silent trial
- Screening time per case; notably faster than standard laboratory workflows
- 30.6 ± 6.0 s
Clinical pilot for T2D
- AUC
- 0.776 (0.710-0.842)
- NPV
- 0.966 (0.946-0.983)
- Surpassed the Finnish Diabetes Risk Score
Original publication
AI framework for multidisease detection via retinal imaging. Nat Med. Published 28 April 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →