Related Experiment Video
Updated: May 13, 2026
05:54
Quantification of Diabetes-induced Adherent Leukocytes in Retinal Vasculature
Published on: January 24, 2025
Machine learning-based prediction of diabetic retinopathy using clinlabomics: a multi-center study
Lu He1, Mengyu Zhang2, Xuanxuan Wang3
1Ruibao Huibao Pediatric Clinic, Shanghai, 200030, China.
BMC Medical Informatics and Decision Making
|May 12, 2026
Summary
Machine learning models using routine lab data can predict diabetic retinopathy (DR) risk. This cost-effective approach aids early triage, especially where specialized screening is unavailable.
Area of Science:
- Ophthalmology
- Medical Informatics
- Biochemistry
Background:
- Diabetic retinopathy (DR) is a primary cause of vision loss.
- Current screening methods are expensive and require significant resources.
- Machine learning (ML) offers a potential for cost-effective DR risk assessment.
Purpose of the Study:
- To develop and validate ML models for DR risk stratification.
- To utilize routine laboratory data for predicting DR.
- To create an accessible tool for early DR triage.
Main Methods:
- Analysis of 750 patients' data (363 Type 2 Diabetes Mellitus, 387 DR).
- Screening of 50 hematological and biochemical parameters.
- Training six algorithms, with XGBoost selected as the top performer.
- Utilizing SHapley Additive exPlanations (SHAP) for interpretability.
Main Results:
- XGBoost model achieved an Area Under the Curve (AUC) of 0.87.
- A reduced set of four predictors (total cholesterol, blood urea nitrogen, fibrinogen, glucose) maintained AUC of 0.87.
- External validation confirmed model robustness (AUC=0.86) with good sensitivity (0.73) and specificity (0.80).
Conclusions:
- Routine laboratory parameters are effective predictors for DR using ML.
- An interpretable, web-based XGBoost tool can aid in risk scoring and triage.
- This solution is particularly valuable for primary care settings lacking specialized imaging.
