Interpolation artifacts in multimodality image registration based on maximization of mutual information.

Area of Science:

• Medical Imaging
• Image Processing
• Computer Vision

Background:

• Mutual Information (MI) is a key metric for multimodality image registration.
• Interpolation significantly affects MI-based registration accuracy.
• Understanding and mitigating interpolation artifacts is crucial for reliable results.

Purpose of the Study:

• Characterize interpolation artifacts on MI for image registration.
• Investigate strategies to overcome these artifacts and enhance registration robustness.

Main Methods:

• Evaluated eight interpolators: nearest neighbor, linear, cubic Catmull-Rom, Hamming-windowed sinc, partial volume, and NN variants (JIT, BLUR).
• Assessed impact of interpolation on MI across various translational and rotational misregistrations in 2D.
• Tested strategies including jittered sampling (JIT), histogram blurring (BLUR), and image resampling.

Main Results:

• Interpolation introduced spurious MI fluctuations, especially with coinciding periodicities and near alignment.
• Wider support interpolators (e.g., cubic Catmull-Rom) did not reduce artifacts.
• Partial volume interpolation caused arch or inverted-arch artifacts.
• JIT, BLUR, controlled intensity bins, and rotated resampling improved robustness.

Conclusions:

• Interpolation artifacts pose a significant challenge for MI-based image registration.
• Jittered sampling and histogram blurring are effective strategies to mitigate these artifacts.
• Findings are applicable to related metrics like normalized MI and joint entropy.