Class balanced diabetic retinopathy image synthesis using a latent diffusion framework
This study establishes the Diabetic Retinopathy Latent Diffusion Synthesizer (DR-LDS) as a highly resource-efficient solution to the medical data bottleneck by leveraging a fine-tuned Variational Autoencoder for domain-adapted latent space compression, alongside an optimized U-Net.