Lung motion estimation by robust point matching and spatiotemporal tracking for 4D CT

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biomedical Engineering

Background:

• Accurate patient-specific lung motion estimation is crucial for 4D CT-based radiotherapy and surgical planning.
• Existing deformable registration methods often struggle with large deformations and temporal inconsistencies.

Purpose of the Study:

• To develop and validate a robust deformable registration approach for precise lung motion modeling in 4D CT.
• To improve the accuracy and temporal stability of lung motion estimation during free breathing.

Main Methods:

• A robust point matching (RPM) algorithm with Scale-Invariant Feature Transform (SIFT) for initial coarse alignment.
• L1 norm regularization for temporal consistency of point trajectories.
• Spatial mean-shift iteration for accurate point tracking across respiratory phases.
• Iterative refinement of RPM and point tracking until convergence.

Main Results:

• The proposed method achieved satisfied accuracy in lung motion registration across multiple public datasets (DIR-lab, POPI-model, CREATIS, COPDgene).
• Demonstrated preservation of topology for image registration with large deformations.
• Enhanced accuracy in establishing correspondences between 4D CT image phases.

Conclusions:

• The developed deformable registration method accurately estimates patient-specific lung motion from 4D CT.
• The approach ensures stable and reliable lung motion modeling, preserving anatomical integrity.
• This technique holds promise for improving precision in thoracic image-guided interventions.