Adaptive local boundary conditions to improve deformable image registration.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Image Registration

Background:

• Accurate assessment and correction of movement are vital in image-guided therapy.
• Deformable image registration (DIR) aligns moving images with fixed ones.
• Boundary conditions are critical for DIR accuracy, especially near image borders.

Purpose of the Study:

• To customize boundary conditions for diverse medical DIR tasks.
• To address the limitations of global boundary conditions in DIR.
• To develop an automated framework for adaptive boundary conditions.

Main Methods:

• Analyzed homogeneous Dirichlet and Neumann boundary conditions.
• Proposed a generic, locally adaptive, Robin-type boundary condition.
• Automated the framework using hyperparameter optimization via energy minimization.

Main Results:

• Tested on CT thorax and CT-to-MRI abdominal registration tasks.
• Achieved up to 12% relative improvement in target registration error for thorax CT.
• Demonstrated near-optimal results for the abdominal CT-to-MRI task.

Conclusions:

• Tailoring boundary conditions is crucial for optimizing image registration.
• Introduced a novel voxel-by-voxel adaptive boundary condition method.
• The framework provides optimized DIR results without prior image assumptions.