Can Diabetic Eye Testing Be Simplified?

Three-test models matched nine-test battery in classifying disease stages

By
Julie Greenbaum
February 18, 2026
4 min

Conexiant

Overview

Machine learning models can classify stages of diabetic eye disease with high accuracy using minimal visual function tests. This study suggests that a simplified testing approach may be as effective as traditional methods.

Background

Diabetic retinopathy (DR) is a leading cause of blindness, making early detection and management crucial. Current screening protocols often involve multiple tests, which can be resource-intensive. Simplifying the testing process could enhance accessibility and efficiency in diabetic eye care.

Data Highlights

Classification Task	AUC (Top Models)
DM without DR vs No DM	0.94 - 1.00
DR without DME vs DM without DR	0.94 - 1.00
DR with DME vs DR without DME	0.94 - 1.00

Key Findings

Machine learning models achieved AUC values of 0.94 or higher for classifying diabetic eye disease stages.
Top models often included combinations of three visual function tests.
Distance visual acuity was a common feature in the top-performing models.
Exclusion of perimetry tests did not significantly affect model performance.
Longitudinal studies are needed to assess the predictive value of visual function measurements.

Clinical Implications

The findings suggest that healthcare providers may consider using fewer visual function tests for diabetic eye disease classification without compromising accuracy. This could streamline the screening process and improve patient access to care.

Conclusion

The study demonstrates that machine learning can effectively simplify diabetic eye testing, potentially leading to more efficient screening protocols. Further research is warranted to validate these findings in longitudinal studies.