Task based characterization of spectral CT performance via the Hotelling Observer

Many different approaches to spectral CT, also called dual-energy CT, are being investigated for clinical use, but there is somewhat of a lack of quantitative metrics for evaluating performance with regard to clinical tasks. In this paper we apply a Hotelling Observer model with signal variability to the task of distinguishing two different contrast materials in a known background. The calculation of the associated Hotelling SNR gives a quantitative description of how well a particular spectral CT system performs. Unlike the detectability studies that are carried out in conventional CT, the inclusion of signal variability is key to applying the Hotelling Observer to dual-energy data. This model could be applied to many tasks that are being studied in clinical trials, such as creating virtual non-contrast images, bone removal, or kidney stone identification. Additionally, the proposed model can be adapted to examine the information loss associated with various dual-energy reconstruction algorithms.