The utilization of consistency metrics for error analysis in deformable image registration

Area of Science:

• Medical imaging
• Computational anatomy
• Image processing

Background:

• Deformable image registration is crucial for medical image analysis.
• Assessing the accuracy of deformable registration is challenging.
• Consistency metrics offer potential for error analysis.

Purpose of the Study:

• To investigate the utility of consistency metrics, including inverse consistency, for contour-based deformable registration error analysis.
• To evaluate the correlation between consistency metrics and actual registration error.
• To identify limitations of consistency metrics in predicting registration accuracy.

Main Methods:

• Acquired four images of a phantom with simulated deformations.
• Simulated deformations using deformable image registration algorithms.
• Calculated deformation maps in forward and reverse directions to assess inverse consistency.
• Introduced the generalized inverse consistency error map (Sigma(Chi)) for error quantification.

Main Results:

• The correlation between consistency metrics and registration error varied significantly based on the algorithm and metric used.
• Actual registration error was generally larger than the measured consistency metrics.
• Consistency metrics showed a necessary but not sufficient relationship with registration accuracy.

Conclusions:

• Consistency metrics are valuable but not definitive indicators of deformable registration accuracy.
• Further research is needed to refine error analysis techniques for deformable registration.
• The generalized inverse consistency error map (Sigma(Chi)) provides a novel approach to error quantification.