Mathematical observer for 3D radiographic scanner design

In the engineering processes of medical 3D radiographic scanner, detailed design elaborates each aspect of the device with complete descriptions, which may include operating parameters as well as the scanner specifications. For assessing the developments mathematically and objectively, task-based approaches are advocated for the optimization of image quality through the scanner design and imaging parameters. In this paper, we present a framework for assessing tomosynthesis scanner performance in lung lesion signal detectability through imaging parameters such as projection frames, lesion depths and anatomical background by using Hotelling observer (HO). Moreover, a channelized Hotelling observer (CHO) with efficient choices of non-linear iterative partial least squares (NIPALS) channels is implemented to reduce the large data dimensionality for computations. Area under the receiver operating characteristic curve (AUC) is computed as an objective figure-of-merit for the performance measurement that described the entire imaging chain. The presented framework is validated with our laboratory x-ray scanner and an anthropomorphic chest phantom. In our result, the performance is, evaluated as functions of AUC to the imaging parameters. Performance map shows 5 mm lung lesion detectability is worsened with cardiac projection-overlapping. Scanner performance is highly related to anatomical structures at the image background. The developed framework has presented its value on assess tomosynthesis scanner performance through anthropomorphic phantom imaging. It can be an objective method for determining scanner detailed design and imaging parameters. (C) 2017 Elsevier Inc. All rights reserved.