Voxel-By-Voxel Validation of Deformable Image Registration Between External Beam and Intracavitary Brachytherapy Images Using Artificial CT Images Created From Patient Populations
Z Yu1*, J Yang1, L Dong2, Y Zhang2, L Court1, F Mourtada3, S Tucker1, A Klopp1, P Eifel1, R Kudchadker1, (1) UT MD Anderson Cancer Center, Houston, TX, (2) Scripps Proton Therapy Center, San Diego, CA, (3) Christiana Care Hospital, Newark, DESU-C-WAB-2 Sunday 1:00PM - 1:55PM Room: Wabash Ballroom
Purpose: This study aims to develop a novel method for volumetric validation of deformable image registration. The developed method was used to determine the accuracy in bladder and rectum registration between external beam radiation therapy (EBRT) and intracavitary brachytherapy (ICBT) CT images.
Methods: Twenty patients treated for intact cervical cancer were used for this study. All patients had EBRT CT (CT1) and ICBT CT (CT2). A single patient from the population was chosen as the reference patient. CT1 from the remaining 19 patients were then deformed to the reference CT1 (CT1_ref). The mean deformation and principle components of deformation fields were determined and used to simulate a known inter-patient deformation. For each patient, CT1 was also deformed to CT2 to quantify deformation within each patient. The mean deformation and principal components of those fields were used to create a known intra-patient deformation field. The inter-patient deformation field was applied to the CT1_ref and inter- and intra-patient deformation field was applied to CT2_ref to create artificial CT1 and CT2, respectively. Our DIR algorithm was tested by deforming artificial CT1 to CT2. Errors in DIR were determined by taking the difference between calculated and the known fields.
Results: We have successfully created artificial CT images that were realistic, represented our patient population, and had realistic deformations between EBRT and ICBT. The mean errors in registration for the bladder were 0.98 mm, 1.09 mm, -0.17 mm, and 3.47 mm in the RL, AP, SI directions, and total magnitude, respectively. For the rectum, the mean errors in registration were 0.02 mm, -0.86 mm, -0.37 mm, and 2.74 mm in the RL, AP, SI directions, and total magnitude, respectively.
Conclusion: We have developed a method to validate any DIR algorithm for any particular patient population and organs of interest.