Researchers at The University of British Columbia examined the use of denoising diffusion probabilistic models (DDPMs) to generate synthetic retinal images for classifying retinal diseases to overcome data scarcity and enhance deep learning-based medical diagnostics. They found that augmenting training data sets with these synthetic images improved the performance of a deep convolutional neural network (CNN) ensemble in classifying retinal disease. Generated image quality was better than conditional GAN-generated fluorescein angiography images, but worse than OCT image generation using the ProGAN methodology. The researchers hypothesized this image quality, as well as ophthalmologists' accuracy in distinguishing real images from generated ones, may have been a result of the number and nature of ophthalmic conditions included in the image generation.

Source: British Journal of Ophthalmology