Multimodal image registration using IECC as the similarity measure

Area of Science:

• Medical Imaging
• Image Registration
• Radiophysics

Background:

• Image registration aligns positron emission tomography (PET) and magnetic resonance imaging (MRI) scans.
• Mutual Information (MI) is a common but occasionally unreliable registration criterion.
• Limitations of MI include small overlap regions, low-activity PET defects, and differing spatial resolutions.

Purpose of the Study:

• To introduce the pixel-based individual entropy correlation coefficient (IECC) as a novel registration criterion.
• To evaluate IECC's accuracy and robustness compared to existing methods like MI, normalized mutual information (NMI), and entropy correlation coefficient (ECC).
• To assess IECC performance in head/brain image registration using simulated and clinical data.

Main Methods:

• Rigid-body registration was applied to brain PET and MRI data.
• IECC was compared against MI, NMI, and ECC using both simulated and clinical datasets.
• PET data included both normal-activity and perfusion-defect models (30% and 50% defects), with reconstructions from filtered backprojection and ordered subset-expectation maximization.

Main Results:

• IECC achieved lower mean errors and standard deviations in translation and rotation compared to MI, NMI, and ECC across all tested conditions.
• In clinical PET, IECC showed mean errors of 1.17 ± 0.85 mm (translation) and 1.04 ± 1.28 degrees (rotation).
• IECC significantly outperformed other criteria (p < 0.01) in reducing misregistration, particularly evident in models with perfusion defects.

Conclusions:

• The individual entropy correlation coefficient (IECC) offers a more accurate and robust criterion for PET-MRI image registration.
• IECC demonstrates statistically significant improvements in accuracy over traditional methods like MI, NMI, and ECC.
• IECC's enhanced performance is particularly valuable in challenging registration scenarios, such as those involving image defects.