Inter-fraction deformable image registration using unsupervised deep learning for CBCT-guided abdominal radiotherapy

• Authors: Huiqiao Xie, Emory University; Yang Lei, Emory University; Yabo Fu, Emory University; Tonghe Wang, Emory University; Justin Roper, Emory University; Jeffrey Bradley, Emory University; Pretesh Patel, Emory University; Tian Liu, Emory University; Xiaofeng Yang, Emory University
• Journal Title: Physics in medicine and biology
• Volume: Volume 68, Number 9
• Publisher: IOP Publishing | 2023-04-13
• Type of Work: Article | Final Publisher PDF
• Publisher DOI: http://doi.org/10.1088/1361-6560/acc721
• Subjects: Health Sciences, Radiology; Engineering, Biomedical
• Keywords: cbct; deep learning; deformable image registration; radiotherapy; Humans; Deep Learning; Image Processing, Computer-Assisted; Spiral Cone-Beam Computed Tomography; Neoplasms; Radiotherapy, Image-Guided; Cone-Beam Computed Tomography; Radiotherapy Planning, Computer-Assisted

Abstract

Objective. CBCTs in image-guided radiotherapy provide crucial anatomy information for patient setup and plan evaluation. Longitudinal CBCT image registration could quantify the inter-fractional anatomic changes, e.g. tumor shrinkage, and daily OAR variation throughout the course of treatment. The purpose of this study is to propose an unsupervised deep learning-based CBCT-CBCT deformable image registration which enables quantitative anatomic variation analysis.Approach.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) to predict the coarse- and fine-scale motions, respectively. The network was trained by minimizing the image similarity loss and the deformable vector field (DVF) regularization loss without the supervision of ground truth DVFs. During the inference stage, patches of local DVF were predicted by the trained LocalGAN and fused to form a whole-image DVF. The local whole-image DVF was subsequently combined with the GlobalGAN generated DVF to obtain the final DVF. The proposed method was evaluated using 100 fractional CBCTs from 20 abdominal cancer patients in the experiments and 105 fractional CBCTs from a cohort of 21 different abdominal cancer patients in a holdout test.Main Results. Qualitatively, the registration results show good alignment between the deformed CBCT images and the target CBCT image. Quantitatively, the average target registration error calculated on the fiducial markers and manually identified landmarks was 1.91 ± 1.18 mm. The average mean absolute error, normalized cross correlation between the deformed CBCT and target CBCT were 33.42 ± 7.48 HU, 0.94 ± 0.04, respectively.Significance. In summary, an unsupervised deep learning-based CBCT-CBCT registration method is proposed and its feasibility and performance in fractionated image-guided radiotherapy is investigated. This promising registration method could provide fast and accurate longitudinal CBCT alignment to facilitate inter-fractional anatomic changes analysis and prediction.