Abstract

In order to transition the GPU-accelerated CT reconstruction algorithm to a more clinical environment, a graphical user interface is implemented. Some optimization methods on the implementation are presented. We describe the alternating minimization (AM) algorithm as the updating algorithm, and the branchless distance-driven method for the system forward operator. We introduce a version of the Feldkamp-Davis-Kress algorithm to generate the initial image for our alternating minimization algorithm and compare it to a choice of a constant initial image. For the sake of better rate of convergence, we introduce the ordered-subsets method, find the optimal number of ordered subsets, and discuss the possibility of using a hybrid ordered-subsets method. Based on the run-time analysis, we implement a GPU-accelerated combination and accumulation process using a Hillis-Steele scan and shared memory. We then analyze some code-related problems, which indicate that our implementation of the AM algorithm may reach the limit of single precision after approximately 3,500 iterations. The Hotelling observer, as an estimation of the human observer, is introduced to assess the image quality of reconstructed images. The estimation of human observer performance may enable us to optimize the algorithm parameters with respect to clinical use.

Committee Chair

Joseph A. O’Sullivan

Committee Members

Joseph A. O’Sullivan David G. Politte Ulugbek Kamilov

Comments

Permanent URL: https://doi.org/10.7936/K7JQ10G3

Degree

Master of Science (MS)

Author's Department

Electrical & Systems Engineering

Author's School

McKelvey School of Engineering

Document Type

Thesis

Date of Award

Spring 5-18-2018

Language

English (en)

Download

Share

COinS