Deformable Image Registration Using Spatial Transformation-Based Network for CBCT-Guided Radiotherapy
H Xie1, Y Lei1, Y Fu1, T Wang1,3, J Roper1,3, X Tang2,3, J Bradley1,3, P Patel1,3, T Liu1,3, X Yang1,3*, (1) Department of Radiation Oncology, (2) Department of Radiology and Imaging Sciences and (3) Winship Cancer Institute, Emory University, Atlanta, GA
Presentations
MO-IePD-TRACK 3-3 (Monday, 7/26/2021) 3:00 PM - 3:30 PM [Eastern Time (GMT-4)]
Purpose: Longitudinal CBCTs from image guided radiation therapy reveal volumetric changes in patient anatomy that may warrant plan changes or be predictive of outcomes. However, quantitative assessments are challenged by poor CBCT image quality, especially in the abdomen, and by the need for fast and accurate deformable registration to make real-time clinical decisions. To address these challenges, we propose a deep learning-based CBCT-CBCT deformable registration method.
Methods: The proposed deformable registration workflow consists of training and inference stages that share the same feed-forward path through a spatial transformation-based network (STN). The STN consists of a global generative adversarial network (GlobalGAN) and a local GAN (LocalGAN), which capture the overall motion and the local motion respectively. The network was trained by minimizing the losses combining image similarity with the deformable vector fields (DVFs) regularization without the supervision of ground truth DVFs. In the inference stage, the global DVF and patches of local DVF were predicted by the trained GlobalGAN and LocalGAN. The local DVF patches were subsequently fused to generate a local whole-image DVF, which is summed with the global DVF to obtain the final DVF. The proposed method was investigated using 100 abdominal CBCTs from 20 radiotherapy patients. Each patient was treated in five fractions over 1 to 2 weeks.
Results: The resulting images show that the deformably registered CBCT images are visually similar to the target CBCT image. Additionally, quantitative assessment shows that the target registration error (TRE) across all patients and fractions is 2.91±1.16 mm after registration with the proposed method. An accurate CBCT-CBCT deformable registration took less than 3 seconds.
Conclusion: We developed a deep learning-based CBCT-CBCT registration method and investigated its feasibility and performance in fractionated radiotherapy. This registration tool could provide fast and straightforward image registration for CBCT-guided radiotherapy.
Funding Support, Disclosures, and Conflict of Interest: This research is supported in part by the National Cancer Institute of the National Institutes of Health under Award Number R01CA215718, and Pilot Grant by the Winship Cancer Institute of Emory University.
ePosters
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email