A Markov random field approach for topology-preserving registration: application to object-based tomographic image...

Area of Science:

• Medical Imaging
• Computational Anatomy
• Image Processing

Background:

• Accurate interpolation of medical imaging data is crucial for diagnosis and analysis.
• Existing registration methods may struggle with preserving topological structures during multiresolution transformations.

Purpose of the Study:

• To develop and validate a topology-preserving multiresolution elastic registration method for object-based interpolation of tomographic slices.
• To enhance the accuracy and robustness of medical image interpolation.

Main Methods:

• A discrete Markov random field of deformations combined with a block-matching procedure.
• Incorporation of intensity- and gradient-based features for data fidelity.
• Application of hard constraints to maintain local field configurations and ensure topology preservation.
• Unsupervised estimation of field parameters to balance fidelity and smoothness terms.
• Forward and backward registration with combination based on local information content.

Main Results:

• The proposed method successfully interpolates tomographic slices while preserving topological features.
• Quantitative comparisons show improved performance over existing methods in magnetic resonance and computed tomography brain and torso acquisitions.
• Component analysis on cardiac magnetic resonance images demonstrates the statistical significance of each algorithmic component for performance.

Conclusions:

• The topology-preserving multiresolution elastic registration method offers a significant advancement in medical image interpolation.
• The method's components are essential for achieving high-performance interpolation, particularly in complex cardiac imaging.
• This approach holds promise for improving the quality and reliability of medical image analysis.