Prediction of the risk of diabetic foot from corneal nerve images using deep learning algorithms
Authors' conclusion
Our DLAs integrating corneal nerve images with HbA1c or serum creatinine have good performance in predicting and stratifying diabetic foot risk, providing a new screening approach.
Study context
- Population
- Participants with corneal nerve images (23,550 images from 942 eyes)942 eyes
- Participants
- 471 participants
- Comparison
- Deep learning algorithm (DLA) using corneal nerve images alone vs hybrid DLAs adding clinical risk factors or quantitative corneal nerve parameters
- Outcome
- AUC for predicting diabetic foot risk categories
Methodological detailsdeep learning algorithm development
- Deep learning algorithm development
Performance
As reported in the abstract
Prediction of diabetic foot and identification of high-risk patients
| Test or model | AUC, prediction of diabetic foot | AUC, identification of high-risk patients | AUC, prediction of diabetic foot (+HbA1c) | AUC, identification of high-risk diabetic foot (+serum creatinine) |
|---|---|---|---|---|
| DLA (images alone) | 0.69 | 0.76 | — not stated in the abstract | — not stated in the abstract |
| Hybrid DLAs | — not stated in the abstract | — not stated in the abstract | 0.94 | 0.93 |
AUC, prediction of diabetic foot
DLA (images alone)0.69
Hybrid DLAs— not stated in the abstract
AUC, identification of high-risk patients
DLA (images alone)0.76
Hybrid DLAs— not stated in the abstract
AUC, prediction of diabetic foot (+HbA1c)
DLA (images alone)— not stated in the abstract
Hybrid DLAs0.94
AUC, identification of high-risk diabetic foot (+serum creatinine)
DLA (images alone)— not stated in the abstract
Hybrid DLAs0.93
— not stated in the abstract
Original publication
Prediction of the risk of diabetic foot from corneal nerve images using deep learning algorithms. Diabetes Res Clin Pract. Published 6 November 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →