Sara Brüningk, University of Bern, Center for AI in radiation oncology, Switzerland.

10:30h.

This presentation introduces a mechanistic learning framework for modeling the spatio-temporal growth of (pediatric) brain tumors following radiotherapy. By combining data-driven generative AI with mathematical modeling, we capture both anatomical progression and dynamic growth patterns over time, even in settings with sparse longitudinal data. Generative models, such as denoising diffusion models, produce realistic, probabilistic maps of tumor expansion, offering insights into variability and potential growth trajectories. These methods integrate prior biological knowledge with machine learning flexibility, enabling more accurate, personalized predictions of tumor behavior. Our results highlight the promise of mechanistic learning in advancing tumor growth modeling beyond conventional, isotropic assumptions.