Temporally Consistent and Controllable Video Generation of 2D Cine CMR via Latent Space Motion Modeling
Researchers have developed a novel text-to-video framework for generating realistic and controllable 2D cine cardiac magnetic resonance (CMR) sequences. This method addresses the scarcity of public CMR datasets by decoupling spatial structure from temporal motion. A diffusion model creates an initial frame based on clinical text prompts, while a latent flow model generates the cardiac motion, ensuring temporal coherence and fidelity to the prompts. The system achieved a FID score of 31.68 for image realism and a CLIP score of 31.04 for text-image alignment, demonstrating its potential for on-demand medical data synthesis. AI
IMPACT Enables on-demand generation of high-fidelity medical imaging data, potentially accelerating research and clinical applications.