MO-F-BRA-04: Voxel-Based Statistical Analysis of Deformable Image Registration Error via a Finite Element Method

Area of Science:

• Medical Physics
• Radiotherapy
• Image Analysis

Background:

• Clinical implementation of adaptive treatment planning is hindered by the lack of quantitative tools to assess deformable image registration errors (R-ERR).
• Accurate assessment of R-ERR is crucial for improving the precision and efficacy of radiotherapy.
• Existing methods for evaluating R-ERR lack the quantitative precision needed for clinical application.

Purpose of the Study:

• To develop and validate a novel method using finite element modeling (FEM) to estimate registration errors based on their mechanical consequences.
• To establish a quantitative correlation between mechanical changes and registration errors in medical imaging.
• To provide a tool for assessing deformable image registration errors in the context of adaptive radiotherapy.

Main Methods:

• Simulated diaphragm deformation on CT images of lung cancer patients using FEM to generate simulated displacement vector fields (F-DVF).
• Performed B-Spline based registrations (Elastix) to generate registration DVF (R-DVF) and calculated registration error as the difference between F-DVF and R-DVF.
• Computed unbalanced energy (UE) using in-house FEM software and developed a nonlinear regression model to correlate UE with registration errors and DVFs.

Main Results:

• A significant correlation (R̂2=0.73, R=0.854) was found between unbalanced energy (UE) and registration error (R-ERR), including its product with the displacement vector field (DVF).
• The association was independently verified using 40 tracked landmarks, establishing a linear function between mean UE and R-DVF*R-ERR.
• The mean registration error was 0.9 mm, with 85.4% of voxels fitting the model within one standard deviation.

Conclusions:

• An encouraging relationship between unbalanced energy (UE) and registration error (R-ERR) has been identified.
• Unbalanced energy (UE) shows feasibility as a valuable tool for evaluating registration errors in medical imaging.
• This method supports the advancement of 4D and adaptive radiotherapy by providing quantitative error assessment.