Encrypted login | home

Program Information

A Novel Volumetric Imaging Method Using a Sparse Subset of CBCT Projections

no image available
Z Zhong

Z Zhong1*, Y Cai1, X Guo1, V Kearney2, L Jiang2, J Wang2, J Yordy2, S Chen2, L Nedzi2, T Solberg2, W Mao2, (1) UT Dallas, Richardson, Texas, (2) UT Southwestern Medical Center, Dallas, Texas

WE-C-WAB-9 Wednesday 10:30AM - 12:30PM Room: Wabash Ballroom

Purpose: : Over the preceding decade, onboard volumetric imaging utilizing cone beam CT (CBCT) has become an indispensible tool for image-guided radiotherapy. A novel method has been developed to deform the planning CT to maximize the spatial congruence between the digitally reconstructed radiographs (DRRs) and CBCT projections. This method maps the planning CT images to the current anatomy using a sparse subset of CBCT projections.

Methods: A finite element modeling methodology was developed to automatically generate adaptive meshes for a given region of interest without the need for segmentation. In this methodology, more control points are generated at organs boundaries, special tissues, contour boundaries, or high intensity gradients. The displacement of boundaries and important features is directly represented by the displacement of control points, instead of interpolation on regular grids. A splatting method was developed to generate DRRs of the deformed CT. These DRRs were compared with corresponding 2D projections from CBCT scans; as a result, the displacement vector fields on the CT were optimized iteratively to obtain a final deformed CT image. A digital phantom was used to evaluate this method.

Results: Typically 60 projections were needed to reconstruct a set of 3D images while several hundreds of projections were required for a conventional CBCT reconstruction. The feature-based meshing method led to better results than either regular orthogonal or regular tetrahedral meshes. The image intensity results of the feature-based meshing method were nearly identical to those of the voxel-based method (normalized cross correlation efficient (NCC) ~ 0.9858 vs 0.9872), while the feature-based meshing method led to much better consistency of displacement vector fields relative to ground truth than voxel-based method (NCC ~ 0.8265 vs 0.6261).

Conclusion: This feature-based meshing method is capable of reconstructing volumetric images with significantly fewer projections while retaining the image quality of planning CT.

Funding Support, Disclosures, and Conflict of Interest: This research is supported by CPRIT Individual Investigator Award RP110329.

Contact Email: