SMART: A Flexible, Interpretable, and Scalable Spatio-temporal Brain Atlas from High-Resolution Imaging Data
Researchers have developed SMART, a novel framework for creating flexible, interpretable, and scalable spatio-temporal brain atlases from high-resolution medical imaging data. Unlike previous black-box models, SMART decouples global disease dynamics from individual anatomical manifestations by learning a continuous disease-time atlas. This approach uses region-specific differential equations to model progression along a shared disease timeline and employs Neural Cellular Automata for personalization. Tested on five longitudinal MRI datasets for Alzheimer's disease, SMART demonstrated state-of-the-art forecasting accuracy and improved temporal consistency. AI
IMPACT Establishes a new paradigm for modeling spatio-temporal changes in medical imaging, potentially improving disease prediction and understanding.