Room: AAPM ePoster Library
Purpose: Combining dose from external beam and brachytherapy components of combined modality therapy requires accurate image registration. The purpose of the current study was to compare rigid image registration, point-based registration, and deformable image registration (DIR) in a prostate seed implant (PSI) phantom after translation, rotation, and deformation.
Methods: A PSI phantom was implanted with 3 fiducial markers and non-radioactive 103Pd seeds in two stages, first a partial implant and then a full implant. Computed tomography was acquired for several translations, rotations, and deformations of the phantom in partial and full seed implantation. The shifted phantom CTs were registered via rigid image registration, point-based registration using fiducials, and B-spline DIR to the “reference” phantom CT which only contained fiducials to simulate external beam CT simulation. The Dice Similarity Coefficient (DSC) index of registered prostate and urethra contours were compared for each registration technique as per AAPM TG-132.
Results: The phantom was partially implanted with 26 seeds on the right side, then fully implanted with a total of 58 seeds. All three registrations yielded DSC indices of 0.96-0.97 between contours when the phantom was either rigidly translated or rotated. When the phantom was physically deformed, DIR had the greatest DSC index for both the prostate and urethra at 0.95 and 0.86, respectively. The rigid with fiducial image registration yielded slightly lower indices than DIR at 0.95 and 0.81 for prostate and urethra respectively. Rigid image registration gave the lowest DSC index for both organs at 0.92 and 0.62. On average, the full seed implant yielded slightly lower DSC values than the partial seed implant phantom by 0.7-1.7%.
Conclusion: Rigid registration using fiducials and DIR yielded similar registration accuracy. Given the uncertainties and complexities associated with DIR, point-based rigid registration with fiducial markers is a feasible alternative.