Locally Adaptive Total p-Variation Regularization for Non-Rigid Image Registration With Sliding Motion

Area of Science:

• Medical Imaging
• Image Registration
• Computational Anatomy

Background:

• Respiratory motion estimation is complex due to intricate thoracic movements, including lung surface sliding and organ motion.
• Accurate lung motion recovery is crucial for various clinical applications, such as radiotherapy planning and disease diagnosis.

Purpose of the Study:

• To introduce a novel regularization term, locally adaptive total p-variation (LaTpV), for accurate lung motion recovery within a parametric registration framework.
• To evaluate the performance of LaTpV in handling both smooth and sliding motions in thoracic images.

Main Methods:

• Proposed LaTpV, derived from a modified L_p-norm constraint (1 < p < 2) with a Dirac-shaped prior distribution for adaptive voxel weighting.
• Integrated LaTpV into a parametric registration framework and analytically deduced the gradient of the cost function.
• Validated the method on synthetic, pulmonary CT, and thoracic CT-PET datasets.

Main Results:

• LaTpV adaptively balances smoothness and discontinuity in the displacement field, effectively modeling expected sliding interfaces.
• Quantitative and qualitative results demonstrated that LaTpV significantly outperforms existing regularizers like bending energy and parametric total variation.
• The method showed superior performance in sliding motion correction across various imaging modalities.

Conclusions:

• The proposed LaTpV-based registration scheme offers improved accuracy for lung motion estimation, particularly for sliding motion.
• This approach holds significant potential for clinical applications, including the diagnosis of pleural mesothelioma and refinement of radiotherapy plans.