Direction-aware deformable registration for respiratory motion correction in PET/CT imaging.
1Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, Yunnan, China; College of Engineering, Yuxi Normal University, Yunnan Key Laboratory of Smart City in Cyberspace Security, Yuxi 653100, Yunnan, China.
Radiography (London, England : 1995)
|May 8, 2026
Summary
A new Sparse Direction-aware Registration Network (SDR-Net) improves respiratory motion correction in PET/CT scans. This method enhances lesion evaluation accuracy, especially for smaller lesions, by stabilizing quantitative measurements.
Area of Science:
- Medical Imaging
- Radiology
- Computational Biology
Background:
- Respiratory motion significantly degrades image quality and quantitative accuracy in gated thoracic PET/CT.
- Accurate lesion localization and uptake measurement are crucial for effective diagnosis and treatment monitoring.
Purpose of the Study:
- To develop a deformable registration method for retrospective motion correction in gated PET/CT.
- The method aims to accurately capture directional respiratory motion while ensuring deformation stability.
Main Methods:
- Developed SDR-Net (Sparse Direction-aware Registration Network) for retrospective phase-to-reference motion correction.
- Employed rotation-aware features, sparse directional modeling, and bidirectional training with inverse consistency.
- Evaluated performance on simulated and multi-center clinical 18F-FDG PET/CT datasets.
Main Results:
- SDR-Net demonstrated superior phase-to-reference alignment and reduced deformation folding compared to other methods.
- Clinical evaluation showed lower residual mismatch and more consistent lesion-based PET measures.
- Benefits were most pronounced for small and medium-sized lesions, improving quantitative stability.
Conclusions:
- Direction-aware and deformation-stable registration effectively corrects respiratory motion in gated thoracic PET/CT.
- This approach enhances the consistency and reliability of lesion-based quantitative assessments.
- SDR-Net can be implemented as a retrospective post-processing step without altering acquisition protocols.
