Deep learning-derived retinal biomarker associated with diabetes-related amputation in type 2 diabetes

By
Junseok Park
Jung Soo Yoon
Sahil Thakur
Dongjin Nam
Sunjin Hwang
July 1, 2026
0 min

Frontiers In Endocrinology

Overview

This study investigates the association between a deep learning-derived retinal biomarker and the risk of diabetic foot-related amputation in patients with type 2 diabetes.

Background

Diabetic foot-related amputation is a significant complication in type 2 diabetes, with high morbidity and mortality rates. Identifying biomarkers that can predict amputation risk is important. The study explores the potential of a retinal biomarker to assess macrovascular disease burden associated with diabetic foot complications.

Data Highlights

Measure	Value
AUC (full model)	0.791
Sensitivity (rule-out cutoff)	87.5%
Negative Predictive Value	92.6%
Specificity (rule-in cutoff)	90.4%
Positive Predictive Value	56.3%

Key Findings

The study included 392 individuals with type 2 diabetes, of whom 79 had DF-related amputation.
The deep learning model achieved an AUC of 0.791 in predicting DF-related amputation.
Adding the retinal biomarker to a clinical model improved predictive performance (cNRI 0.629).
Sensitivity and specificity at the rule-out and rule-in cutoffs were 87.5% and 90.4%, respectively.
The association between the retinal biomarker and amputation risk remained consistent across sensitivity analyses.

Clinical Implications

The findings indicate that a deep learning-derived retinal biomarker can enhance the prediction of amputation risk in patients with diabetic foot complications.

Conclusion

The study demonstrates that a retinal biomarker derived from deep learning is significantly associated with the risk of diabetic foot-related amputation.