Task-Based Optimization of Computed Tomography Imaging Systems

TL;DR

This thesis develops a task-based framework using image quality metrics like Hotelling observer SNR to improve CT system design, focusing on analytical and optimization-based reconstruction algorithms.

Contribution

It introduces a formalism for evaluating and guiding CT image reconstruction algorithm design using task-based image quality metrics.

Findings

Task-based metrics can effectively guide CT algorithm development.

Hotelling observer SNR correlates with perceived image quality.

Analysis of noise properties in TV-based iterative reconstruction enhances metric applicability.

Abstract

The goal of this thesis is to provide a framework for the use of task-based metrics of image quality to aid in the design, implementation, and evaluation of CT image reconstruction algorithms and CT systems in general. We support the view that task-based metrics of image quality can be useful in guiding the algorithm design and implementation process in order to yield images of objectively superior quality and higher utility for a given task. Further, we believe that metrics such as the Hotelling observer (HO) SNR can be used as summary scalar metrics of image quality for the evaluation of images produced by novel reconstruction algorithms. In this work, we aim to construct a concise and versatile formalism for image reconstruction algorithm design, implementation, and assessment. The bulk of the work focuses on linear analytical algorithms, specifically the ubiquitous filtered back-projection (FBP) algorithm. However, due to the demonstrated importance of optimization-based algorithms in a wide variety of CT applications, we devote one chapter to the characterization of noise properties in TV-based iterative reconstruction, as the understanding of image statistics in optimization-based reconstruction is the limiting factor in applying HO metrics.