Related Experiment Video
Updated: Oct 2, 2025
Author Spotlight: Unraveling the Pathogenesis of Age-Related Macular Degeneration and Discovering Potential Therapies
Published on: July 28, 2023
A deep learning model for screening type 2 diabetes from retinal photographs.
Jae-Seung Yun1, Jaesik Kim2, Sang-Hyuk Jung3
1Department of Biostatistics, Epidemiology and Informatics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA; Division of Endocrinology and Metabolism, Department of Internal Medicine, St. Vincent's Hospital, College of Medicine, The Catholic University of Korea, Seoul, Republic of Korea.
A new deep learning algorithm using retinal images can help screen for type 2 diabetes. This non-invasive tool improves risk prediction when combined with traditional risk factors.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Type 2 diabetes poses a significant global health challenge.
- Early screening and risk stratification are crucial for effective management.
- Retinal imaging offers a non-invasive window into systemic health.
Purpose of the Study:
- To develop and evaluate a non-invasive deep learning algorithm for type 2 diabetes screening.
- To assess the algorithm's performance using retinal images from UK Biobank participants.
- To compare the algorithm's predictive power against traditional risk factors.
Main Methods:
- A deep learning model was trained on retinal images from over 50,000 participants.
- The model's ability to predict traditional risk factors (TRFs) and diabetes risk was evaluated.
- Performance was compared between an image-only model, TRFs, and a combined model.
Main Results:
- The deep learning algorithm achieved high AUCs for predicting age (0.931), sex (0.933), and HbA1c status (0.734).
- The image-only model had an AUC of 0.731 for type 2 diabetes prediction, while TRFs achieved 0.810.
- Combining the algorithm with TRFs improved the AUC to 0.844 and showed a 50.8% net reclassification improvement.
Conclusions:
- The developed deep learning algorithm is a valuable tool for non-invasive type 2 diabetes risk stratification.
- Integrating retinal image analysis with TRFs enhances predictive accuracy.
- This approach holds promise for identifying high-risk individuals in the general population.
