Classification of diabetic maculopathy based on optical coherence tomography images using a Vision Transformer model.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Diabetic maculopathy (DM) is a leading cause of vision loss.
• Accurate staging of DM is crucial for timely intervention and prognosis.
• Optical coherence tomography (OCT) is a key imaging modality for DM assessment.

Purpose of the Study:

• To develop and evaluate a Vision Transformer (ViT) deep learning model.
• To detect and classify different stages of diabetic maculopathy using OCT images.
• To assess the model's performance in differentiating early diabetic macular edema, advanced DME, severe DME, and atrophic maculopathy.

Main Methods:

• A dataset of 3319 retrospective OCT images from DM patients was curated.
• Images were randomly split into training (70%) and validation (30%) sets.
• A deep learning model based on the Vision Transformer architecture was trained for DM stage detection.

Main Results:

• The ViT model achieved an overall accuracy of 82.00% and an F1 score of 83.11%.
• Area under the curve (AUC) for overall detection was 0.96.
• Specific AUCs for early DME, advanced DME, severe DME, and atrophic maculopathy were 0.96, 0.95, 0.87, and 0.98, respectively.

Conclusions:

• The developed Vision Transformer model demonstrates high accuracy in grading diabetic maculopathy from OCT images.
• This AI tool can assist in preliminary screening, identifying high-risk patients for further evaluation.
• The study highlights the potential of AI in supporting clinical decision-making for diabetic maculopathy management.