Related Experiment Videos
Detection of referable diabetic retinopathy using machine learning on routine clinical data
Young Joon Jeon1, Jae Shin Song1, Shubham Borghare2
1Department of Ophthalmology, Seoul National University College of Medicine, Seoul National University Bundang Hospital, Seongnam, Republic of Korea.
Frontiers in Medicine
|June 4, 2026
Summary
A machine learning model accurately predicts referable diabetic retinopathy (RDR) using clinical data, avoiding the need for eye imaging. This tool aids early RDR detection, especially in underserved areas, preventing vision loss.
Area of Science:
- Ophthalmology
- Medical Informatics
- Machine Learning
Background:
- Early detection of referable diabetic retinopathy (RDR) is vital for preventing vision loss.
- Machine learning (ML) models can predict RDR using clinical and laboratory data, bypassing the need for ophthalmic imaging.
Purpose of the Study:
- To develop and validate an ML model for predicting RDR without fundus imaging.
- To assess the model's performance and identify key predictors of RDR.
Main Methods:
- A cohort of 562 adults with diabetes was analyzed retrospectively and prospectively.
- Four ML models were trained on clinical and laboratory variables, with performance evaluated using AUROC.
- Predictor importance was determined using Shapley Additive Explanations (SHAP).
Main Results:
- The random forest model achieved an AUROC of 0.932 in the validation set.
- Key predictors included age, diabetes duration, fasting glucose, BMI, and blood pressure.
- The model demonstrated high sensitivity (85.8%) and specificity (91.2%).
Conclusions:
- A random forest model effectively identifies RDR using routine clinical data, eliminating the need for fundus imaging.
- This ML tool can facilitate early RDR detection in resource-limited settings.
- The model supports timely referrals and integration into clinical decision support systems.
