Generating OCT B-Scan DME images using optimized Generative Adversarial Networks (GANs).

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic Macular Edema (DME) causes significant vision loss in diabetics.
• Manual interpretation of Optical Coherence Tomography (OCT) B-Scan images for DME is error-prone.
• Reliable diagnostic methods are crucial for effective DME management.

Purpose of the Study:

• To develop an automated model using Generative Adversarial Networks (GANs) to synthesize DME OCT B-Scan images.
• To enhance the robustness of DME detection systems through generated synthetic OCT images.
• To compare the performance of five different GAN architectures for DME image generation.

Main Methods:

• Utilized five GANs: Deep Convolutional GAN, Conditional GAN, CycleGAN, StyleGAN2, and StyleGAN3.
• Generated synthetic OCT B-Scan images from baseline patient OCT images.
• Fine-tuned hyperparameters of the best-performing GAN using Particle Swarm Optimization (PSO).

Main Results:

• Comparative analysis of five GANs for generating realistic DME OCT images.
• Identification of the best-performing GAN for synthesizing high-quality OCT B-Scan images.
• Demonstrated potential for generated images to improve DME detection and severity assessment.

Conclusions:

• The proposed GAN-based model offers a promising approach for generating realistic DME OCT images.
• This method can potentially improve the accuracy and reliability of DME diagnostic systems.
• Insights provided on selecting appropriate GANs for synthetic OCT image generation in ophthalmology.