Synthesis of 3D volumetric brain images with diffusion models

Recent advancements in deep learning have significantly impacted medical image analysis. Although mainly used in image classification, neural networks also found its way into synthetic image generation. Early generative networks were hard to train and images they generated lacked diversity, but this has changed with the advent of stable diffusion models. In my research I am using the diffusion models to synthesise 3D volumetric images if the ageing brain. I focus on addressing challenges like inter-subject variation and the scarcity of longitudinal data, and aim to develop models capable of simulating individual brain ageing trajectories without requiring extensive longitudinal datasets. If successful, this approach could enhance the early detection of degenerative diseases and contribute towards fully personalized medicine.

In this talk, I will introduce the diffusion models and explain the network architecture and the mathematics behind it. Then, I will describe my approach to implementing this models and discuss the ongoing and future work.