Impact of point-source injection methodologies on injection studies

Effective interdiction of illicit radioactive material requires decision criteria that are sensitive to the material of interest, while insensitive to material of lesser interest. Injection studies are used to assess the sensitivities of decision criteria to radioactive threat scenarios. Configuring injection studies requires decisions about methodological factors that represent or simplify actual field conditions. Understanding the relative impact of different source injection methods on alarm rates and other outcome measures is vital for interpreting results of injection studies.The most complete and realistic way to inject point-source counts into a vehicle count profile is with the randomly positioned source profile approach. In this paper, we report a study of the impacts of methodological simplifications to this approach. We assessed relative sensitivity of two field-deployed detection algorithms to two experimental factors in point-source injection studies. The first factor featured “centered” versus “random” point-source position along the vehicle count profile. The second factor featured either adding the entire injection profile (“profile injection”) or adding only the central “spike” portion of the injection profile nearest to the point-source position (“spike injection”). A collection of 75 059 archived vehicle count profiles from an operating border crossing were injected with a simulated low-energy gamma-ray point-source intended to emulate surrogates of highly enriched uranium or plutonium. Outcomes were assessed by means of injection curves (alarm rate vs. injection source size) and a derived measure, minimum detectable count rates (MDC) required to attain a detection probability of 95%.Results show that injection methodology combined with algorithm impacts study outcomes in different ways. For gross count results, centered spike injections yield higher detection sensitivity than centered profile injections, and random spike injections yield even higher detection sensitivity than centered spike injections. Conversely, for energy window ratio results, centered spike injections yield less sensitivity than centered profile injections, and random spike injections yield even less sensitivity than centered spike injections. These effects may be partially attributable to background suppression in the vehicle profile of detection statistics. Although the observed effects (up to 10%) are not large, the injection method impacts alarm rates from the two detection algorithms in opposite directions. These results inform methodological decisions and assessments of results in injection studies.