D-GET: Group-Enhanced Transformer for Diabetic Retinopathy Severity Classification in Fundus Fluorescein Angiography

Area of Science:

• Ophthalmology and Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Diabetic Retinopathy (DR) diagnosis requires timely detection to prevent vision loss.
• Manual DR diagnosis using Fundus Fluorescein Angiography (FFA) is labor-intensive, costly, and prone to inaccuracies.
• Existing DR classification models struggle with subtle lesion variations and detecting small lesions.

Purpose of the Study:

• To develop a deep learning model for accurate DR classification from FFA images.
• To enhance the detection of small-scale DR lesions, which are often missed by current methods.
• To improve the efficiency and accuracy of DR diagnosis.

Main Methods:

• Proposed a novel deep learning model named D-GET (Diabetic Retinopathy - Group-Enhanced Transformer).
• Incorporated a Full-Scale Transformer Block with a Group-Focal module for multi-scale feature capture and contextual integration.
• Integrated a Channel Adaptive Attention Module (CAAM) for improved feature detection and localization.

Main Results:

• The D-GET model demonstrated superior performance compared to existing methods on a custom dataset.
• The model significantly improved the detection of small-scale lesions crucial for early DR diagnosis.
• D-GET enhances the classification of DR lesion severity in FFA images.

Conclusions:

• The D-GET model offers a significant advancement in automated DR classification using FFA.
• Improved detection of small lesions by D-GET aids in earlier and more accurate DR diagnosis.
• This deep learning approach provides a foundation for broader applications in ophthalmic and medical imaging.