Delta-Diffusion: A New AI Framework for Modeling Brain Amyloid PET Trajectories

Researchers have introduced Delta-Diffusion, a novel AI framework that models longitudinal brain amyloid-PET trajectories using a conditional Poisson Diffusion Bridge. The method, detailed in a paper on arXiv (2606.22216), addresses limitations of existing deep generative models in capturing subtle pathological progression in Alzheimer's disease. Standard diffusion models often fail due to identity drift and bias toward replicating baseline signals. Delta-Diffusion redefines the synthesis task as a conditional distribution transition, anchored to the subject's baseline PET, enabling more accurate modeling of amyloid accumulation over time. This approach could enhance clinical utility by reducing the need for repeated PET scans, which are costly and involve radiation risks. The framework is mathematically grounded in the Poisson diffusion process, offering a progression-aware alternative for longitudinal image synthesis. The paper was announced as a replace-cross update, indicating revisions. The work is relevant to the intersection of AI, medical imaging, and neurodegenerative disease research.