GIFTed Demons: deformable image registration with local structure-preserving regularization using supervoxels for...

Area of Science:

• Medical Imaging
• Computational Anatomy
• Image Processing

Background:

• Deformable image registration is crucial for motion correction in medical imaging, requiring efficient and biologically plausible spatial transformations.
• Standard methods like Demons registration often use Gaussian regularization, limiting their ability to accurately model complex organ motions, such as sliding interfaces.
• Accurate modeling of organ motion is essential for improving diagnostic and therapeutic interventions in dynamic medical imaging.

Purpose of the Study:

• To develop and evaluate a novel regularization method for deformable image registration that can handle complex organ motions, specifically sliding interfaces.
• To improve the accuracy and efficiency of motion correction in medical imaging, particularly for lung and liver applications.
• To introduce a discontinuity-preserving prior for motion estimation using supervoxels and guided filtering.

Main Methods:

• Proposed a novel regularization approach for deformable image registration based on supervoxels.
• Replaced traditional Gaussian smoothing with fast, structure-preserving guided filtering for locally adaptive regularization of displacement fields.
• Applied the framework to estimate sliding motions at lung and liver interfaces using four-dimensional computed tomography (4D CT) and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) datasets.

Main Results:

• Guided filter-based regularization demonstrated improved accuracy in lung and liver motion correction compared to Gaussian smoothing.
• The proposed method effectively handles complex motions, including sliding interfaces, which are challenging for standard techniques.
• Achieved state-of-the-art results on a publicly available CT liver dataset, validating the framework's performance.

Conclusions:

• Supervoxel-based regularization with guided filtering offers an efficient and accurate solution for deformable image registration, particularly for complex organ motions.
• The proposed method enhances the reliability of motion correction in medical imaging, leading to better diagnostic accuracy.
• This approach represents a significant advancement in handling sliding motion phenomena in medical image analysis.