Rectified Flow Matching for High-Resolution CT Reconstruction
August 31, 2026
A high-resolution Rectified Flow Matching model is trained on 256x256 chest images to provide a more efficient alternative to stochastic diffusion models for CT reconstruction. The approach uses a two-stage training strategy with anatomically informed data augmentation to improve structural fidelity and numerical stability.
HOW THIS AFFECTS YOU
●
researcherYou can explore flow matching as a faster, more stable alternative to diffusion for medical inverse problems.
●
healthThis could lead to more computationally efficient and stable CT imaging reconstruction workflows.