Diffusion models in medical imaging
Invited speaker, Spring school on Deep Learning in Medical Imaging (DLMI), Lyon, France
Diffusion-based generative models have recently emerged as a powerful framework for modeling complex data distributions, especially in high-dimensional domains like medical imaging. In this lecture, I introduced the core concepts behind Denoising Diffusion Probabilistic Models (DDPM), a prominent class of diffusion models grounded in stochastic processes. I then illustrated their practical relevance through applications to data augmentation, semantic segmentation, and anomaly detection, highlighting how such models can generate anatomically plausible and diverse samples.