Date of Award
Doctor of Philosophy (PhD)
Gilat Schmidt, Taly
Selecting the tube current can be challenging when using iterative reconstruction algorithms due to the varying relationship between spatial resolution, contrast, noise, and dose across different algorithms. The objective of this dissertation was to develop and evaluate an Automated Exposure Control (AEC) method for iterative reconstruction algorithms using a task-based image quality metric. To achieve this objective, several CT image quality metrics were evaluated and developed. Two studies were performed to evaluate candidate metrics for use in a task-based AEC method. In Aim 1, a study was performed to identify and experimentally evaluate candidate metrics that have been previously proposed for task-based CT image quality assessment. This study quantitatively evaluated the performance of the Channelized Hotelling Observer (CHO) and the exponential transformation of the free-response operating characteristic curve (EFROC) with respect to sensitivity to changes in dose. The number of images required to estimate the non-parametric EFROC metric was calculated for varying tasks and found to be less than the number of images required for parametric CHO estimation. The EFROC metric was found to be more sensitive to changes in dose than the CHO metric. The Aim 2 study proposed and evaluated fractal dimension as a novel metric for quantifying noise texture. The Aim 2 study demonstrated that fractal dimension was correlated to the previously proposed metric of frequency of the NPS peak and could be estimated in a clinical image from one ROI of size 128 by 128 or four ROIs of size 64 by 64.Based on the Aim 1 and Aim 2 studies, the detectability index, d^', metric was chosen for implementation as part of the proposed task-based AEC method. Detectability index is a task-based image quality metric that combines the contrast-dependent spatial resolution, noise properties and an analytical representation of the task to be detected into a single figure of merit. The detectability index metric was generalized by approximation using a look of table of scaling factors that convert between noise standard deviation and generalized detectability index, d_gen^'. The proposed method leverages existing AEC methods that are based on a prescribed noise level. Generation of the look-up tables requires calibration scans to estimate the task-based modulation transfer function and noise power spectrum. Results demonstrated that the proposed d_gen^'-AEC method provided consistent image quality across different iterative reconstruction approaches, with reduced dose compared to the reference scan.
Available for download on Sunday, October 16, 2022