Y Sang*, D Ruan, Department of Radiation Oncology and Department of Bioengineering, University of California, Los Angeles, CA

• 
  Y Sang

• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  

TU-EF-TRACK 4-9 (Tuesday, 7/27/2021) 3:30 PM - 5:30 PM [Eastern Time (GMT-4)]

Purpose: Cone-beam CTs (CBCTs) are readily available for on-table acquisition in companion with treatment and have lower imaging dose compared to fan-beam CTs (FBCTs). Deformable registration between phases in a 4D-CBCT can potentially provide on-table 4D volumetric information of target location and motion. However, CBCT registration is very challenging due to strong artifacts and the lack of structural details. In this study, we aim to develop an enabling approach for thoracic 4D CBCT registration.

Methods: We propose a novel approach to incorporate a learned implicit regularizer as a feasibility prior of respiratory motion, into the training of an unsupervised image registration network to improve registration accuracy and robustness to noise and artifacts. Specifically, a feasibility descriptor from a set of deformation vector fields (DVFs) is first generated from FBCTs. Subsequently, this FBCT-derived feasibility descriptor was used as a spatially variant regularizer on DVF Jacobian during the unsupervised training for 4D-CBCT registration. In doing so, the higher-quality, higher-confidence information from FBCT is transferred into the much challenging problem of CBCT registration, without explicit FB-CB synthesis. The method was evaluated using manually identified landmarks on five real CBCTs and automatically detected landmarks on nine simulated CBCTs.

Results: The method presented good robustness to noise and artifacts and generated physically more feasible DVFs. The target registration errors on the real and simulated data were (1.63±0.98) and (2.16±1.91) mm, respectively, significantly better than the classic bending energy regularization in both the conventional method in SimpleElastix and the unsupervised network. The average registration time was 0.04 s.

Conclusion: The proposed method imposes a strong and robust motion feasibility prior and enhances registration networks’ resistance against undesired image qualities. It demonstrates the strength of a learned and implicit feasibility prior, in comparison to the pre-defined regularization energies.

Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by a research grant from Varian Medical Systems, Inc.

Handouts

Keywords

Registration, Cone-beam CT

Taxonomy

IM/TH- Image Registration Techniques: Machine Learning

Contact Email