A Quantitative Analysis of Lung Registration Error in a Clinically Used Deformable Image Registration Implementation Using Manually Identifiable Features in Fast Helical Free-Breathing CT Scans
L Naumann*, B Stiehl, M Lauria, K Singhrao, A Santhanam, D Low, University of California, Los Angeles, Los Angeles, CA
Presentations
PO-GePV-M-144 (Sunday, 7/25/2021) [Eastern Time (GMT-4)]
Purpose: Deformable image registration (DIR) of fast helical free-breathing CT (FBFHCT) scans produces deformation vector fields (DVFs) which describe the motion within the lung at the voxel level. The accuracy of these results are difficult to quantify and remain unverified. In this study, we aim to measure the error of a clinically used lung DIR implementation using identifiable landmark feature tracking.
Methods: We developed a feature tracking algorithm of manually identified and selected structures within the deformably registered reference and target lung images. A set of 25 patient datasets (source-target image pair and associated DVFs) were retrospectively employed for this study. DVFs were generated using dense displacement sample (deeds) DIR. 50 manually identifiable structures were selected on each reference scan and the corresponding structures were then identified on the associated target scan. The algorithm accepts the input of a precise point (single voxel) on the identified feature selected by a user in the reference image. After displaying the corresponding deformed voxel location in accordance with the DVF, the algorithm accepts a true voxel location, which was selected based on the corresponding feature on the target image. The Euclidean distance and percent error between the deformed voxel location and the true voxel location was then calculated and reported by the algorithm for each set of feature points.
Results: The average distance/registration error measured across all features and datasets was 0.83mm (min: 0.28mm, max: 1.65mm, std: 0.37mm). The average percent error was observed to be 13.18% (min: 4.74%, max: 30.84%, std: 7.76%).
Conclusion: This work employed a feature tracking tool to quantify the image registration accuracy of a clinically used DIR lung registration algorithm for FHFBCT scans.
ePosters
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email