Evaluating Similarity Measures for Brain Image Registration.

Area of Science:

• Medical Image Analysis
• Computational Neuroscience
• Biomedical Imaging

Background:

• Evaluating similarity measures for brain image registration is complex.
• Existing methods are affected by optimization, regularization, and image types.
• A robust evaluation is needed to isolate similarity measure performance.

Purpose of the Study:

• To propose a novel evaluation method for brain image registration similarity measures.
• To introduce a single performance metric, 'robustness', to quantify similarity measure effectiveness.
• To develop a new, highly robust similarity measure for 3D brain image registration.

Main Methods:

• Developed a performance metric named 'robustness' to evaluate similarity measures.
• Designed an evaluation framework that isolates the effect of similarity measures.
• Introduced 'normalized spatial mutual information' as a new similarity measure.

Main Results:

• Higher robustness correlates with better registration of degraded images.
• Robustness effectively predicts success in intermodal brain image registration.
• The proposed normalized spatial mutual information shows significantly higher robustness than existing measures.

Conclusions:

• The 'robustness' metric reliably assesses similarity measure performance in brain image registration.
• The new normalized spatial mutual information measure offers superior performance, especially for degraded and intermodal data.
• This work provides a more effective way to select and develop similarity measures for challenging registration tasks.