Denoising diffusion probabilistic model for generating histopathology images
Mahfujul Islam Rumman, Naoaki Ono, Kenoki Ohuchida, Md. Altaf-Ul-Amin, Ahmad Kamal Nasution, Shigehiko Kanaya · 2025
Generative models are machine learning models that focus on learning the pattern or distribution of original data to generate new and similar-looking data. In the world of artificial intelligence, generative models play a crucial role in tasks that need the creation of new data. The ability to create new data makes generative models a powerful tool in computer vision. On that note, we utilized diffusion models to synthetically create new histopathology images. In recent times, diffusion-based generative models have shown promise in generating high-quality images in image-generation tasks. In this work, we used a denoising diffusion probabilistic model with three different variance schedulers to generate images that resemble our original dataset. We evaluated the quality of the generated images using Fréchet Inception Distance, which is a metric used to measure the similarity between the original and generated images. Using our model, histopathology images can be artificially created for plenty of other tasks in the field of computer vision.