Ensemble Framework of Deep CNNs for Diabetic Retinopathy Detection

Area of Science:

• Ophthalmology
• Computer Vision
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a progressive eye disease damaging retinal blood vessels, potentially leading to blindness.
• Early detection and classification of DR stages are crucial but challenging for conventional computer vision methods.
• Existing methods struggle with accurately classifying intricate DR features, especially in early stages.

Purpose of the Study:

• To develop and evaluate advanced deep learning models for comprehensive diabetic retinopathy stage detection.
• To improve the accuracy of DR classification, particularly for early and subtle stages.
• To address limitations of current computer vision techniques in analyzing complex retinal pathologies.

Main Methods:

• Two deep convolutional neural network (CNN) models were designed and combined using an ensemble technique.
• Models were trained on a Kaggle dataset, utilizing both balanced and imbalanced data splits.
• Performance was evaluated using balanced and imbalanced test datasets on high-end GPU infrastructure.

Main Results:

• The proposed ensemble CNN models successfully detected all five stages of diabetic retinopathy (0-4).
• The models outperformed existing state-of-the-art methods on the same Kaggle dataset.
• Effective classification was achieved across both balanced and imbalanced datasets, indicating robustness.

Conclusions:

• The developed deep CNN ensemble models offer a significant advancement in automated diabetic retinopathy detection.
• These models provide a more accurate and comprehensive approach to classifying DR stages than conventional methods.
• The findings suggest a promising direction for improving DR screening and management through AI.