A Deep Learning Prognostic Model for Diabetes Patients Using Bilateral Fundus Imaging
Authors' conclusion
We developed a survival-aware, image-based model that can predict mortality risk directly from routine fundus photography, achieving prognostic discrimination which facilitates diabetes management.
Study context
- Population
- Persons with DM who received routine color fundoscopy examinations
- Participants
- 19,029 persons
- Comparison
- Image-trained mortality model risk quartiles vs lowest-risk quartile; comparison with a regulatory-approved commercial ophthalmic system
- Outcome
- Mortality
Methodological detailsdeep learning, Cox proportional hazards modeling
- Deep learning
- Cox proportional hazards modeling
Performance
As reported in the abstract
Mortality by image-trained model risk quartile, testing cohort
Hazard ratio (95% CI)
Second-lowest-risk quartilevs Lowest-risk quartile
1.67(0.80–3.47)
Second-highest-risk quartilevs Lowest-risk quartile
3.02(1.55–5.88)
Highest-risk quartilevs Lowest-risk quartile
7.19(3.91–13.25)
- 5-year survival rate of highest-risk quartile
- lower than 80%
Further findings
Commercial ophthalmic system strata, mortality
- Hazard ratios (HRs) across commercial system strata
- 1.48, 2.02, and 3.11
Model in strata of equivalent sizes to commercial system, mortality
- HR
- 3.64 (3.18-4.16) (95% CI)
- HR
- 4.98 (4.08-6.07) (95% CI)
- HR
- 6.89 (6.01-7.89) (95% CI)
- Stronger separation than the commercial system
5-year survival of top 8% patients identified by the model
- 5-year survival rate
- <70%
Parts of this Brief are withheld: the automatic checks could not confirm them against the abstract.
Original publication
A Deep Learning Prognostic Model for Diabetes Patients Using Bilateral Fundus Imaging. J Diabetes Sci Technol. Published 29 March 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →