(Thursday, 7/18/2019) 7:30 AM - 9:30 AM
Room: 225BCD
Purpose: In locally advanced cervical cancer, large inter-fractional organ deformations occur during external beam radiotherapy (EBRT) and brachytherapy (BT). This study investigates a surface projection algorithm to handle the complex deformations of the sigmoid.
Methods: Twenty patients were retrospectively analyzed. The sigmoids were delineated on the pre-BT CT at EBRT boost (without applicator) and BT CT (with applicator). The symmetric thin-plate spline robust-point-matching (sTPS-RPM) was used to estimate the anatomical deformations due to the applicator insertion. The sTPS-RPM was improved using the centerline and the normalized length, as constraints, within a multi-resolution scale of smoothing factors. The classical and constrained sTPS-RPM were compared to ANACONDA, a commercially available hybrid deformable image registration (DIR) method, using the Dice similarity coefficient (DSC), distance to agreement (DTA) and Hausdorff distance (HD). The two sTPS-RPM were compared using surface triangle quality criteria. The mean displacement of the D2cc was reported for 16 of the 20 patients.
Results: Rigid registration was not able to handle the sigmoid deformations resulting in an average DSC of 0.27. The average DSC, DTA and HD were 0.51, 1.6 mm, 28.9 mm for ANACONDA, 0.96, 0.4 mm, 6.1 mm for the sTPS-RPM and 0.98, 0.3 mm, 2.5 mm for the constrained sTPS-RPM. The two sTPS-RPM methods provided an inverse consistency error below 1 mm for all sigmoids. The constrained sTPS-RPM improved the deformed surface quality with a mean absolute difference of the Jacobian and aspect ratio of 0.16 and 0.6 compared to the classical approach. The mean (min-max) D2cc of 12 Gy (1-21) shifted of 26 mm (9-43) between pre-BT and BT anatomies.
Conclusion: Classical DIR methods cannot handle the large and complex deformations of the sigmoid due to applicator insertion. Future investigations will focus on the constrained sTPS-RPM to define boundary condition for multi-organs biomechanical model-based DIR.
Funding Support, Disclosures, and Conflict of Interest: Funding from RaySearch Laboratories
Keywords
Taxonomy
Not Applicable / None Entered.
Contact Email
