Program Information
Optimizing B-Spline Deformable Image Registration Between Planning CT and Cone-Beam CT in Head and Neck Cancer Patients
M Nakano1,3*, R Kierkels2 , S Hanaoka3 , E Korevaar2 , T Imae3 , A Haga3 and N M Sijtsema2 , (1) Cancer Institute Hospital, Japanese Foundation for Cancer Research, Koto-ku, Tokyo, Japan, (2) University of Groningen, University Medical Center Groningen, Groningen, The Netherlands, (3) University of Tokyo Hospital, Bunkyo-ku, Tokyo, Japan
Presentations
SU-I-GPD-J-43 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: Deformable image registration (DIR) between planning CT (pCT) and cone-beam CT (CBCT) is increasingly-demanded technique for adaptive radiotherapy to accommodate interfractional organ deformation. The purpose of this study was to optimize free-form B-spline registration parameters for CT-to-CBCT DIR in head and neck region.
Methods: The image series of pCT and CBCT of five head and neck patients treated by 35-fractionated IMRT were investigated. DIR was performed using Elastix v4.7. The control point grid spacing (GS) and the weight of the bending energy penalty (BEP) were optimized for the pCT-CBCT registrations. Optimization was implemented using five patients’ pCT and CBCT of day-32 as moving and fixed image, respectively. Then the optimized parameter set was verified in the performance test using pCT and CBCT of day-2, -12, -22 and -32. DIR results were evaluated by errors between landmarks identified by scale invariant feature transform, and normalized mutual information (NMI) values using 256 histogram bins.
Results: The NMI values increased with smaller GS values. Overall, grid spacing of 5.00 mm provided the best results for all five patients. BEP = 0.05 gave the highest result in one patient and second highest by a narrow margin in three patients. The results of landmark errors in the performance test with GS = 5.00 mm and BEP = 0.05 demonstrated that the median error was around 1.00 mm. The median error was 2.8 mm using rigid registrations only. Meanwhile, unrealistic deformations were observed in areas near the field-of-view boundaries around the shoulders. This implies that the image quality of CBCT affects the deformation quality.
Conclusion: The B-Spline deformable registration algorithm was optimized for pCT-CBCT registrations and resulted in a median DIR accuracy of less than an image voxel. The errors were classified by landmarks derived from the scale invariant feature transform algorithm.
Funding Support, Disclosures, and Conflict of Interest: The present research was supported by JSPS Core-to-Core Program No. 23003 and JSPS KAKENHI No. 16H07429
Contact Email: