An adaptive motion regularization technique to support sliding motion in deformable image registration.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biomedical Engineering

Background:

• Conventional isotropic smoothing in deformable image registration (DIR) struggles with complex tissue deformations like sliding motion.
• This limitation hinders accurate modeling of organ boundaries in thoracic and abdominal imaging.

Purpose of the Study:

• To develop and evaluate an adaptive direction-dependent regularization technique for deformation vector fields (DVFs).
• To accurately model and estimate sliding tissue motion in medical images, improving DIR accuracy.

Main Methods:

• An iterative DVF computation using intensity differences, regularized with an adaptive direction-dependent filter (Gaussian normal, bilateral tangential).
• Automatic delineation of sliding surfaces (e.g., chest/abdominal walls) and adaptive parameter adjustment based on distance maps.
• Validation on 14 4D-CT datasets (lung, upper abdomen, digital phantom) comparing Target Registration Error (TRE) with existing methods.

Main Results:

• Preserved sliding motion near chest and abdominal walls.
• Reduced average TRE by 35.1% on lung datasets compared to five other DIR methods.
• Achieved lower Sum of Squared Differences (SSD) and significant TRE reduction near liver and spleen on a digital phantom.

Conclusions:

• The developed adaptive direction-dependent DVF regularization method effectively models sliding tissue motion.
• This approach enhances overall motion estimation accuracy, particularly in regions with significant organ sliding, such as the chest and abdominal walls.