JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

DTG: Dual transformers-based generative adversarial networks for retinal 2D/3D OCT image classification

  • 0Department of Ophthalmology, Université de Montréal, Montreal, Québec, Canada; Centre Universitaire d'Ophtalmologie (CUO), Hôpital Maisonneuve-Rosemont, CIUSSS de l'Est-de-l'Île-de-Montréal, Montréal, Québec, Canada.
Medical Image Analysis +

|

January 10, 2026

|

January 10, 2026

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces Dual Transformers-based Generative Adversarial Networks (DTG) for improved retinal disease classification from Optical Coherence Tomography (OCT) scans. DTG enhances diagnostic accuracy by leveraging advanced AI and data augmentation techniques.

Area Of Science

  • Ophthalmology
  • Medical Imaging
  • Artificial Intelligence

Background

  • Automated retinal disorder identification is a key ophthalmology application.
  • Transformer models show promise in image classification but require more data for medical applications.
  • Current methods for retinal data classification need performance improvements.

Purpose Of The Study

  • To propose a novel deep learning architecture, Dual Transformers-based Generative Adversarial Networks (DTG), for enhanced retinal disease classification.
  • To address the data-hunger issue of Transformer models in medical imaging.
  • To improve the accuracy and reliability of automated diagnosis using Optical Coherence Tomography (OCT) data.

Main Methods

  • Utilized Vision Transformer and Multiscale Vision Transformer for encoding 2D and 3D OCT images.
  • Employed a Generative Adversarial Networks (GAN) architecture for high-quality semantic data representation.
  • Implemented a patient instance-based data augmentation technique to increase training data.
  • Applied a weighted classifier for the final retinal disease classification task.

Main Results

  • The proposed DTG architecture achieved superior performance across multiple metrics including accuracy, precision, recall, f1-score, and AUC.
  • DTG outperformed popular Convolutional Neural Networks and Transformer models in classifying 2D and 3D OCT images.
  • The approach demonstrated significant improvements over existing methods for retinal data classification.

Conclusions

  • The DTG architecture offers a powerful solution for automated retinal disease classification from OCT data.
  • The combination of Transformers, GANs, and data augmentation effectively addresses data limitations in medical AI.
  • This work advances the field of AI in ophthalmology, promising more accurate and efficient disease diagnosis.

Keywords: