To propose a new metric, strain alignment, for evaluating the mechanical feasibility of predicted displacement fields in soft tissue registration, focusing on deformation direction and magnitude.
Approach:
Metric Definition: The strain alignment metric B3 is adapted from the mean alignment metric used in swarming behavior analysis, focusing on the largest principal strain value to assess deformation direction and magnitude.
Experiments: The applicability of B3 was demonstrated by analyzing inference results from a deep learning method (V2S-Net) and a biomechanical method (BCF-FEM) on synthetic liver mesh data.
Key Findings:
The strain alignment metric B3 provides a more interpretable measure of deformation direction and magnitude compared to the Jacobian determinant and strain norm.
The metric is designed to penalize sign flips between compression and extension, reflecting gradual changes expected in soft tissues.
Interpretation:
Strain alignment provides an alternative for evaluating displacement fields without requiring dense ground truth data.
Limitations:
The metric's effectiveness is contingent on the quality of the underlying mesh and the accuracy of the deformation model used.
Conclusion:
Strain alignment is a tool for assessing the mechanical plausibility of predicted displacement fields in soft tissue applications.
