Search papers, labs, and topics across Lattice.
The SMART framework introduces a novel approach for constructing a spatio-temporal brain atlas from high-resolution longitudinal imaging data, overcoming limitations of existing black-box models in flexibility, interpretability, and scalability. By utilizing anatomically inspired priors and region-specific differential equations, SMART effectively models global disease dynamics while personalizing trajectories to individual anatomies through a multi-scale Neural Cellular Automata. Evaluated on over 1,300 subjects across five longitudinal MRI datasets, SMART achieves state-of-the-art accuracy in forecasting disease progression in Alzheimer's, demonstrating its potential for impactful clinical applications.
SMART redefines brain atlas construction by achieving state-of-the-art forecasting accuracy while maintaining interpretability and scalability in high-dimensional medical imaging.
We introduce SMART, a framework for learning a flexible, interpretable, and scalable spatio-temporal brain atlas from longitudinal high-resolution 3D medical images. Existing approaches to spatio-temporal atlas construction rely on black-box generative models that lack flexibility, limit interpretability, and struggle to scale to high-dimensional data. SMART addresses these challenges by learning a continuous disease-time atlas that decouples global group-wise disease dynamics from their patient-specific anatomical manifestation. Guided by anatomically inspired priors, SMART models interpretable global trajectories of regional progression along a shared disease timeline through region-specific differential equations. Global trajectories are further personalized to individual anatomies via dense diffeomorphic displacements parameterized by a flexible and scalable multi-scale Neural Cellular Automata. Evaluated on five longitudinal MRI datasets in Alzheimer's disease (ADNI-1/GO/2, OASIS-3, AIBL; > 1,300 subjects), SMART produces anatomically meaningful predictions of disease progression and achieves state-of-the-art forecasting accuracy and improved temporal consistency over adversarial and diffusion baselines. Our approach establishes a new paradigm for flexible, interpretable, and scalable modeling of spatio-temporal change in high-dimensional medical image time-series.
