Zheng Zhang MA, BS, Presenter: Nothing to Disclose

Junguo Bian PhD, Abstract Co-Author: Nothing to Disclose

Xiao Han MSc, Abstract Co-Author: Nothing to Disclose

Daxin Shi PhD, Abstract Co-Author: Employee, Toshiba Corporation

Alexander Zamyatin PhD, Abstract Co-Author: Employee, Toshiba Corporation

Emil Y. Sidky PhD, Abstract Co-Author: Nothing to Disclose

Xiaochuan Pan PhD, Abstract Co-Author: Research Grant, Koninklijke Philips Electronics NV Research Grant, Toshiba Corporation Consultant, UtopiaCompression Corporation

In diagnostic computed tomography (CT) imaging, it is often of interest to obtain detailed information within a region of interest (ROI), while rough knowledge outside the ROI may be sufficient. This novel imaging approach leads to an image reconstruction problem that requires voxels of different sizes within and outside the ROI. In this work, we develop and investigate an optimization-based algorithm to reconstruct images with variable spatial resolution, that is, images with non-uniform voxel sizes.

We used a Toshiba 320-slice diagnostic CT scanner to collect data from a patient and a swine using a circular geometry. Both data sets were acquired at 1200 views over 2π. We performed image reconstruction by using a modified adaptive-steepest-descent-projection-onto-convex-sets (ASD-POCS) algorithm, which is specifically adapted to accommodate image arrays with variable resolution. Using this modified algorithm, we performed reconstruction on a variable-resolution array, which consists of voxels of size 0.06 cm within a selected ROI, and voxels of size 0.12 cm outside the ROI. We then carried out additional reconstructions by further increasing the voxel size outside the ROI to 0.24 and 0.48 cm. As references, we also applied ASD-POCS algorithm to reconstructing images on uniform-resolution arrays, consisting of voxels of sizes ranging from 0.06 cm to 0.48 cm.

By visual comparison, we observed that in the variable-resolution images, as the ratio of voxel size outside the ROI to that within the ROI increased from 1 to 8, although the region outside the ROI becomes progressively coarser, the image quality within the ROI remains virtually identical to that of the reference image reconstructed with uniform voxels of size 0.06 cm. 

The results demonstrate that by employing an optimization-based algorithm tailored to variable-resolution images, we are able to reconstruct images within ROI of quality comparable to that obtained with uniform, high-resolution arrays.

Variable-resolution optimization-based reconstruction can reduce computation time and memory consumption. It may also have potential impact on ROI images reconstructed from data containing truncation.

Zhang, Z, Bian, J, Han, X, Shi, D, Zamyatin, A, Sidky, E, Pan, X, Optimization-based Image Reconstruction with Variable Resolution in Diagnostic CT.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13014700.html Accessed November 7, 2025