Development of a list-mode ideal observer to perform classification tasks when imaging nuclear inspection objects under signal-known-exactly conditions

Abstract

We developed a signal-known-exactly version of the ideal observer that processes data in list-mode format to perform binary classification, a useful task for arms-control treaty applications. This observer offers the best possible performance and future observer models developed in our work will be compared to this model. The two examined sources were plutonium inspection objects developed by Idaho National Lab. We modeled a fast-neutron coded-aperture imager, developed by Oak Ridge National Lab and Sandia National Labs to acquire simulation data. Monte Carlo simulations using the GEANT4 toolkit tracked photons and neutrons from these objects to the imager. The observer model was evaluated using the area under the ROC curve for multiple background strengths.