EchoLVFM
Research · arXiv 2025One-step echocardiogram video synthesis via Latent Flow Matching. Given a target ejection fraction (EF), the model generates a realistic cardiac ultrasound video in a single forward pass — no iterative diffusion steps required.
Video Generation
Generating synthetic echocardiograms conditioned on different ejection fraction values.
Video Reconstruction
Reconstructing real echocardiograms in latent space, with matched ejection fraction values.
Latent Flow Matching
Rather than learning a denoising process over many steps, LFM learns a straight-line flow in latent space between noise and data — enabling one-step generation.
Clinical Conditioning
The model is conditioned on ejection fraction (EF), a key cardiac health metric, allowing targeted synthesis of pathological and healthy cardiac motion.
Application
Synthetic data augmentation for rare cardiac conditions, and a controllable simulator for training downstream echocardiogram analysis models.