Enhancing spatiotemporal disease progression models via latent diffusion and prior knowledge

Daniele Ravi

Abstract: Computational models of disease progression are crucial for understanding neurodegenerative conditions and predicting patient outcomes. This work presents an enhanced framework that combines spatiotemporal disease progression modeling with latent diffusion models and biologically-informed prior knowledge. We leverage diffusion-based generative models to capture complex disease dynamics while incorporating domain knowledge about disease mechanisms to improve model accuracy and interpretability. Our approach demonstrates improved performance in longitudinal neuroimaging analysis and disease trajectory prediction.