Simultaneous registration of multiple images: similarity metrics and efficient optimization

Area of Science:

• Medical image analysis
• Computer vision
• Computational imaging

Background:

• Simultaneous image registration is crucial for aligning multiple images.
• Existing frameworks like accumulated pair-wise estimates (APE) require efficient optimization.
• Extending registration to multimodal datasets presents significant challenges.

Purpose of the Study:

• To mathematically derive and optimize the accumulated pair-wise estimates (APE) framework for image registration.
• To develop efficient gradient-based optimization strategies for APE, including Gauss-Newton and efficient second-order minimization (ESM).
• To extend APE and ESM for successful multimodal image registration.

Main Methods:

• Mathematical deduction of APE from a maximum-likelihood framework.
• Extension of the congealing framework with neighborhood information.
• Development and application of Gauss-Newton and efficient second-order minimization (ESM) optimization strategies.
• Incorporation of structural image representations for multimodal registration.

Main Results:

• Efficient gradient-based optimization strategies (Gauss-Newton, ESM) were derived for APE.
• The efficient second-order minimization (ESM) demonstrated very good performance.
• The framework was successfully extended to perform multimodal registration using ESM.
• Evaluation on publicly available datasets with ground-truth alignment confirmed performance.

Conclusions:

• The derived optimization strategies significantly improve the efficiency of simultaneous image registration.
• The extension to multimodal registration using structural representations is a key advancement.
• The enhanced APE framework with ESM offers a robust and accurate solution for both monomodal and multimodal image alignment.