Identification of diabetic retinopathy classification using machine learning algorithms on clinical data and optical...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) poses a significant threat to vision.
• Accurate classification of DR severity is crucial for timely intervention.
• Integrating clinical data with OCTA enhances diagnostic capabilities.

Purpose of the Study:

• To develop and validate machine learning (ML) models for multiclass diabetic retinopathy (DR) classification.
• To assess the performance of ML algorithms using combined clinical and OCTA data.
• To identify key predictors for DR classification.

Main Methods:

• A cross-sectional observational study involving 203 diabetic patients for model development and 169 for external validation.
• Utilized random forest, gradient boosting machine (GBM), deep learning, and logistic regression algorithms.
• Evaluated algorithm performance using receiver operating characteristic curves and area under the curve (AUC).

Main Results:

• Random forest and GBM algorithms achieved higher AUC values for DR, referable DR (RDR), and vision-threatening DR (VTDR) classification when using OCTA and clinical data.
• Key predictors included vessel density, retinal thickness, GCC thickness, BMI, waist-to-hip ratio, and glucose-lowering treatment.
• OCTA combined with clinical data demonstrated superior predictive accuracy.

Conclusions:

• ML-based multiclass DR classification using OCTA and clinical data offers reliable assistance for screening, referral, and management.
• The study highlights the potential of integrating multimodal data for improved DR assessment.
• These findings support the clinical utility of ML in managing diabetic retinopathy populations.