Model-Based Sparse-to-Dense Image Registration for Realtime Respiratory Motion Estimation in Image-Guided...

Area of Science:

• Medical Imaging
• Radiation Therapy
• Image-Guided Interventions

Background:

• Accurate intra-interventional respiratory motion estimation is crucial for radiation therapy and focused ultrasound.
• Current methods using external indicators lack precision.
• Real-time motion tracking using MRI or ultrasound is needed for improved dose delivery.

Purpose of the Study:

• To develop a new approach for real-time respiratory motion tracking applicable to MR-Linac and 3D-Ultrasound.
• To improve the accuracy of motion estimation beyond indirect breathing indicators.

Main Methods:

• Combines GPU-accelerated feature point tracking with a patient-specific motion model.
• Employs a unified optimization framework for regularization and sparse-to-dense interpolation.
• Utilizes contrast-invariant feature descriptors for robust tracking.

Main Results:

• Achieved high accuracy in motion predictions (≈1 mm landmark error for MRI, ≈2 mm for US).
• Demonstrated substantial improvements over traditional template tracking methods.
• Successfully modeled complex physiological respiratory motion, including lung-rib cage sliding.

Conclusions:

• The model-based sparse-to-dense image registration enables accurate, real-time respiratory motion tracking.
• This technique enhances the quality of dose delivery in image-guided interventions.
• The approach is suitable for challenging real-time imaging modalities.