Region-Adaptive Conditional MeanFlow for CT Image Reconstruction
A new CT image reconstruction method called RA-CMF (Region-Adaptive Conditional MeanFlow) has been developed to address variations in noise, contrast, and texture caused by different imaging protocols and scanner models. The approach introduces a conditional MeanFlow network that predicts image-conditioned flow fields from intermediate image states, trained with a MeanFlow consistency loss and image reconstruction loss. A regional reinforcement learning-driven policy network provides adaptive refinement based on spatial location of enhancements. The method aims to improve CT imaging for lung cancer screening, diagnosis, therapy planning, and prognosis.
Key facts
- RA-CMF stands for Region-Adaptive Conditional MeanFlow.
- The method addresses differences in noise statistics, contrast, and texture in CT images.
- It uses a conditional MeanFlow network to model enhancement trajectories.
- The network is trained with MeanFlow consistency loss and image reconstruction loss.
- A regional reinforcement learning-driven policy network provides adaptive refinement.
- The policy network receives information about MeanFlow rollouts.
- CT imaging is important for lung cancer screening, diagnosis, therapy planning, and prognosis.
- The research is published on arXiv with ID 2605.00901.
Entities
Institutions
- arXiv