CT-Guided Spatially-Varying Regularization for Whole-Body PET Registration
A new method for whole-body PET registration uses CT scans to create voxel-wise regularization maps, addressing anatomical heterogeneity. The approach replaces uniform global regularization with spatially-varying constraints: rigid structures like bones receive stronger regularization, while soft tissues are allowed more flexible deformation. This CT-guided strategy aims to improve deformable registration across different PET tracers for multi-parametric tumor characterization and metastatic disease assessment. The work is presented in arXiv:2604.22905.
Key facts
- Whole-body PET registration is essential for multi-parametric tumor characterization and metastatic disease assessment.
- Deep learning-based deformable registration uses a dense displacement field (DDF) regularizer.
- Anatomical heterogeneity is a key challenge: bones need stronger regularization, soft tissues need weaker constraints.
- The proposed method uses paired CT volume from PET/CT acquisition to construct a voxel-wise regularization map.
- This replaces the conventional single global regularization weight.
- The approach is designed for whole-body cross-tracer deformable PET registration.
- The paper is available on arXiv with ID 2604.22905.
- The method is described as simple yet effective.
Entities
Institutions
- arXiv