Voxel-based statistical analysis of uncertainties associated with deformable image registration

Area of Science:

• Medical Imaging
• Computational Anatomy
• Radiotherapy Physics

Background:

• Deformable image registration (DIR) algorithms generate displacement vector fields (DVFs) with inherent uncertainties.
• Accurate estimation of these uncertainties is crucial for reliable medical applications, particularly in adaptive radiotherapy.

Purpose of the Study:

• To develop and validate an optimal metric for estimating uncertainties in DIR-generated DVFs.
• To compare the performance of the proposed metric against existing error metrics.

Main Methods:

• Developed six computational phantoms from patient CT images using finite element method (FEM).
• Generated standard DVFs using FEM and evaluated registration DVFs using a mechanics-based metric, unbalanced energy (UE).
• Validated results using landmark approach (POPI-model) and compared UE with inverse consistency (IC) and image intensity difference (ID) metrics.

Main Results:

• UE showed a higher Pearson correlation coefficient (r=0.50) with DIR error compared to IC (r=0.29) and ID (r=0.37).
• UE demonstrated strong correlations with the product of DIR displacements and errors (r=0.62 to 0.73).
• The UE metric outperformed IC and ID in estimating DIR uncertainties.

Conclusions:

• Unbalanced energy (UE) is a robust and accurate metric for quantifying deformable image registration uncertainties.
• UE can serve as a valuable tool for enhancing adaptive treatment strategies and probability-based treatment planning.