Learning Patient-Specific Disease Dynamics with Latent Flow Matching for Longitudinal Imaging Generation
Researchers have developed a new framework called \u0003-LFM to model patient-specific disease progression using latent flow matching. This approach treats disease dynamics as a continuous velocity field, capturing intrinsic progression for better interpretability. The framework addresses challenges in latent space alignment by enforcing patient trajectories to correlate with clinical severity indicators, leading to a more semantically meaningful latent space. Empirical results on three longitudinal MRI benchmarks demonstrate \u0003-LFM's strong performance and offer novel visualization capabilities for disease dynamics. AI
IMPACT Offers a novel framework for interpreting and visualizing disease dynamics, potentially improving clinical diagnosis and personalized treatment.