Assessing ResNeXt and RegNet Models for Diabetic Retinopathy Classification: A Comprehensive Comparative Study.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy is a major cause of vision loss globally.
• Automated classification systems are vital for early detection and management.
• Retinal fundus images are key for diagnosing diabetic retinopathy.

Purpose of the Study:

• To compare deep learning models for diabetic retinopathy classification.
• To evaluate ResNeXt and RegNet families using retinal fundus images.
• To assess model performance in binary and multi-class settings.

Main Methods:

• Trained and tested ResNeXt and RegNet models.
• Utilized a 70-20-10 data split for training, validation, and testing.
• Assessed performance using precision, sensitivity, specificity, F1-score, and AUC.
• Employed SHapley Additive exPlanations for model interpretability.

Main Results:

• Both ResNeXt and RegNet achieved high performance in binary classification.
• ResNeXt excelled in detecting early diabetic retinopathy stages.
• RegNet demonstrated balanced performance across all stages, especially advanced cases.

Conclusions:

• ResNeXt models effectively identify early diabetic retinopathy signs.
• RegNet models provide more consistent classification across multiple severity stages.
• Combining quantitative metrics and interpretability enhances decision support systems for diabetic retinopathy screening.