Hierarchical multimodal image registration based on adaptive local mutual information.

Area of Science:

• Medical image analysis
• Computer vision
• Computational anatomy

Background:

• Multimodal image registration is crucial for integrating information from different imaging modalities.
• Existing methods, such as mutual information, face limitations in accuracy and robustness for complex registration tasks.
• Adaptive local measures can potentially improve registration performance by considering image characteristics.

Purpose of the Study:

• To develop and evaluate a new adaptive local measure for multimodal image registration.
• To improve the robustness and accuracy of image registration through a hierarchical framework.
• To compare the proposed method against established metrics like mutual information.

Main Methods:

• A novel adaptive local measure based on gradient orientation similarity was developed.
• The metric was embedded within a hierarchical registration framework.
• Computationally efficient gradient orientation estimation using patch-wise rigidity was proposed.

Main Results:

• The proposed method demonstrated improved registration robustness and accuracy by adapting the similarity metric and pixel selection.
• It outperformed mutual information (MI) and its local approximations in multimodal brain image registration (CT/MRI).
• Significant improvements in meanTRE (mTRE) were observed in registering pre-operative brain MRI to intra-operative US images.

Conclusions:

• The adaptive local measure based on gradient orientation similarity offers a superior alternative for multimodal image registration.
• The hierarchical framework enhances registration performance by adapting to image scale and modalities.
• The method shows particular promise for challenging clinical applications, such as neurosurgery planning and guidance.