A viscous fluid model for large-scale motion estimation in image-guided radiotherapy.

Area of Science:

• Medical Physics
• Radiotherapy
• Image Processing

Background:

• Image-guided radiotherapy (IGRT) accuracy is compromised by anatomical and physiological motion.
• Existing deformable image registration (DIR) methods struggle with large displacements in the thorax, abdomen, and pelvis.

Purpose of the Study:

• To propose a physics-derived DIR algorithm for accurate motion estimation in highly deforming anatomical regions during IGRT.
• To develop a computationally efficient DIR solution suitable for online adaptive radiotherapy workflows.

Main Methods:

• Anatomical motion modeled as viscous fluid dynamics, solving simplified Navier-Stokes equations.
• Dissimilarity between images acts as the driving force for deformation estimation.
• FFT-based numerical solver implemented on GPUs for high computational performance; Jacobian determinant used for deformation analysis.

Main Results:

• Achieved 1-2 mm accuracy for thoracic CT and 0.8-0.9 Dice similarity coefficient for pelvic MR images.
• Demonstrated smooth Jacobian determinant distribution, indicating plausible deformations.
• Average computational latency ranged from 20 seconds to 3.5 minutes.

Conclusions:

• The viscous fluid dynamics-based DIR algorithm effectively estimates large deformations in the thorax and pelvis for CT and MR images.
• The algorithm's accuracy and computational performance align with clinical acceptability for online adaptive IGRT motion management.