Room: Karl Dean Ballroom C
Purpose: Analysis of radiation therapy response for liver cancer is often hampered by the difficulty to register longitudinal images when large tumor responses occur. As a new alignment method adapted to extreme liver deformation cases, we propose a deformable image registration (DIR) method based on a biomechanical model driven by boundary conditions (BC) on the segmented liver vasculature.
Methods: Pre- and post-RT contrast enhanced-CT from 65 patients treated with radiotherapy for cholangiocarcinoma and presenting large tumor changes after treatment were retrospectively selected. Contours of the liver were delineated and the surfaces were used in Morfeus, a biomechanical model-based DIR method, to solve the main deformations and deform the post-RT CT onto the pre-RT CT. The vasculature was then segmented on the pre-RT and deformed CT using a vessel enhancement filter followed by adaptive thresholding. A surface projection algorithm robust to partial inconsistencies was applied to align the segmentations. The displacements estimated on the vasculature centerline of the pre-RT image were used as boundary conditions in a second run of Morfeus in order to solve residual misalignments. For each patient, five to eight corresponding landmarks between the pre-RT and post-RT images were manually identified as close as possible to the tumor area and used to measure the target registration error (TRE).
Results: To date, five patients with large tumor response were analyzed. According to the proposed vasculature matching algorithm, the group mean of the displacements to be applied to the vasculature centerline was 7.6Â±3.8 mm after rigid registration. The group mean TRE was 8.2Â±3.5, 8.8Â±4.8 and 3.5Â±1.4 mm after rigid registration, standard Morfeus and Morfeus with additional BC, respectively.
Conclusion: The proposed method to align longitudinal CT images of the liver presenting large anatomical and intensity changes should allow more efficient studies on the relationship between delivered dose and treatment outcome.
Funding Support, Disclosures, and Conflict of Interest: Guillaume Cazoulat and Kristy Brock have research funding from RaySearch. Kristy Brock has a royalty agreement with RaySearch
Image Processing, Image Analysis