CT image assessment by low contrast signal detectability evaluation with unknown signal location

Abstract

To devise a new methodology for CT image quality evaluation in order to assess the dose reduction potential of new iterative reconstruction algorithms (IRA). Because of the nonlinear behavior of IRA, the authors propose a task-based methodology consisting of measuring the detectability of small, low contrast signals at random locations. The authors test, via simulations, a phantom design that facilitates human and numerical observer studies in such conditions. The setup allows for the random selection of regions of interest (ROI) around each signal, so that the relative signal location is unknown if the ROIs are shown separately. With such a setup one can perform signal detectability measurements with a variety of image reading arrangements and data analysis methods. In this work, the authors demonstrate the use of the localization relative operating characteristic method. The phantom design also allows for efficient image evaluation utilizing an automatic signal search technique and a recently developed nonparametric data analysis method using the exponential transformation of the free response characteristic curve. The authors present the application of these methods by performing a comparison between the filtered back projection (FBP) algorithm and a polychromatic iterative image reconstruction algorithm. In this generic illustration of the image evaluation framework, the expected improved performance of the IRA over FBP is confirmed. The results demonstrate the ability of these methods to determine signal detectability indices with good accuracy with only a small number, of the order of a few tens, of image samples.