Practical estimation of detectability maps for assessment of CT scanner performance

摘要

We propose an experimental methodology that facilitates practical assessment of low-contrast detectability maps in real CT images. Specifically, we explain how one may efficiently estimate the area under the ROC curve for a channelized Hotelling observer applied to signal-known-exactly/background-known-exactly lesion detection tasks at multiple points in the imaging field-of-view. Our approach builds on earlier work in which we demonstrated that accurate prior knowledge of channel means may be used to gain a large decrease in the statistical variability of estimated results. Our new procedure is illustrated with detectability maps computed from real CT images of a thorax phantom. These maps provide image quality information that may be regarded as more general than conventional NEQ evaluations.