Options
2014
Conference Paper
Titel
Quantitative evaluation of CT Images by means of Shannon Entropy
Abstract
To assess the quality of CT images, a number of statistical measures are available. This work focuses on the Shannon Entropy (SE) as a quantitative measure for the quality of CT images of the kind which they are produced in NDT applications. The aim of this study is to apply SE to simulated CT images which assumes the same geometry, components and photon statistics as in real measurements. By systematically changing the imaging parameters, i.e. camera exposure time, number of projections and X-ray spectrum, we are able to map the parameter space in terms of SE, thus revealing the influence of each parameter on the CT image quality individually, and to determine the optimum. We calculate the SE for a set of simulated CT images of a relatively simple object. Several statistical measures such as signal to noise ratio (SNR), variance (VAR) and variation are calculated from the data in order to compare their sensitivity to changes in the common image acquisition parameters to the SE-sensitivity. Unlike the SNR and VAR, SE appears to be sensitive to all parameters, thus having a very high significance for the evaluation of the CT image. The simulated images are generated by a deterministic simulation tool (Scorpius X-Lab®, Fraunhofer EZRT, Germany). With this tool, almost any acquisition parameter can be modified for evaluation. In addition to SE, well known artifacts of CT imaging are investigated, e.g. beam-hardening artifacts and noise. Their error as well as their sensitivity to the image acquisition parameters is evaluated in terms of SE. We introduce an additional evaluation tool which enables us, e.g., to distinguish between the influences of the X-ray spectrum and of the integration time. Comparing SE to other statistical measures shows: SE has clear advantages in terms of evaluating CT images and SE correlates with CT-artifacts in the simulated images. As far as we can tell from this study SE-calculation is independent of the object size and geometry.