MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Automated Estimation of Deformable Image Registration Uncertainty in the Liver

Y Zhang1*, J Balter2 , R Kashani3 , (1) Univ Michigan, Ann Arbor, MI, (2) Univ Michigan, Ann Arbor, MI, (3) University of Michigan, Ann Arbor, MI

Presentations

(Sunday, 7/14/2019) 4:00 PM - 5:00 PM

Room: 221AB

Purpose: Current techniques for estimation of deformable image registration (DIR) uncertainty rely on limited manual landmarks and contours. The purpose of this study is to automatically create a spatial map of uncertainty within different regions in the liver, with the aim of incorporating this data into plan optimization in the future.

Methods: A previously reported method for estimation of the local degeneracy of B-spline based (DIR) was utilized. We deformably aligned T1-weighted MR images to planning CT scans, using regularized hierarchical B-splines (NiftyReg) for five patients. Normalized Mutual Information was used as an objective function, and bending energy as the regularization term. The optimal deformation generated by the DIR was altered by applying random variations of up to +/-10mm in the local spline coefficients, and re-computing the objective function locally in a region of 10mmx10mmx15mm around each voxel. The largest deformation that didn’t reduce the value of the local objective function was considered as a measure of local uncertainty. The accuracy of resulting uncertainty maps was evaluated by comparing to the target registration error (TRE) for pairs of landmarks. Each landmark pair was identified twice by an expert to establish repeatability.

Results: Good agreement was observed between the estimated error and TRE for points where the calculated TRE was within the 10mm limit allowed for spline coefficient variations (R^2=0.75). Estimated repeatability of landmark selection was 2mm. Out of 18 points used, the agreement between the estimated uncertainty and the TRE was 1mm or better in 50%, and 2mm or better in 78% of the points.

Conclusion: We demonstrated that a voxel level map of distribution of DIR uncertainty can be generated for the liver using this method for any pair of images. The results can be utilized in evaluating the sensitivity of plan optimization to DIR errors for adaptive radiotherapy.

Funding Support, Disclosures, and Conflict of Interest: Supported in part by NIH/PO1 CA059827

Keywords

Registration, Deformation, Mutual Information

Taxonomy

IM/TH- Image registration : Multi-modality registration

Contact Email