Group-Aware Registration for Lesion-Level Quantitative Motion Correction in Respiratory-Gated PET/CT Biomedical
Hui Zhou1,2, Longxi He1, Yangsheng Hu1
1Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China.
Sensors (Basel, Switzerland)
|July 28, 2026
Summary
SCAR-Net improves respiratory motion correction in PET/CT scans, enhancing quantitative accuracy for small lesions. This advanced network ensures more reliable imaging data for better medical assessments.
Area of Science:
- Medical Imaging
- Radiology
- Artificial Intelligence in Medicine
Background:
- Respiratory motion significantly impacts PET/CT imaging quality, causing spatial inaccuracies and quantitative instability, particularly for small thoracoabdominal lesions.
- Existing respiratory gating methods struggle with precise motion correction, leading to suboptimal data for lesion characterization and analysis.
Purpose of the Study:
- To introduce SCAR-Net (Similarity-Constrained Adaptive Respiratory Registration Network) for retrospective correction of respiratory motion in gated 18F-FDG PET/CT.
- To address motion correction as a quantitative stability challenge, improving lesion-level accuracy and consistency in phase-resolved PET/CT data.
Main Methods:
- SCAR-Net employs sampled group-aware feature encoding and adaptive group-attentive modulation to model respiratory deformation.
- The network was validated using simulations and a retrospective two-center clinical dataset of 100 patients, evaluating SUV repeatability, spatial correspondence, and image similarity.
- Performance was compared against B-spline FFD, with cross-dataset testing and ablation studies assessing robustness and efficiency.
Main Results:
- SCAR-Net significantly reduced median phase-to-reference variability for small lesions (|ΔSUVmax|: 6.45%, |ΔSUVmean|: 4.73%) and improved median lesion Dice from 0.55 to 0.72.
- For large lesions, SCAR-Net achieved a post-correction Dice of 0.89 while maintaining SUV repeatability.
- Consistent performance was observed across different clinical acquisition settings, demonstrating the network's generalizability.
Conclusions:
- SCAR-Net effectively enhances phase consistency and quantitative stability in respiratory-gated PET/CT, particularly benefiting the assessment of small lesions.
- The proposed method offers a promising approach for improving diagnostic accuracy and reliability in PET/CT imaging affected by respiratory motion.
- Further prospective validation is required to evaluate the impact of SCAR-Net on downstream clinical endpoints.
