Reliable and valid measures of threat detection performance in x-ray screening

Abstract

Over the last decades, airport security technology has evolved remarkably. This is especially evident when state-of-the-art detection systems are concerned. However, such systems are only as effective as the personnel who operate them. Reliable and valid measures of screener detection performance are important for risk analysis, screener certification and competency assessment, as well as for measuring quality performance and effectiveness of training systems. In many of these applications the hit rate is used in order to measure detection performance. However, measures based on signal detection theory have gained popularity in recent years, for example in the analysis of data from threat image projection (TIP) or computer based training (CBT) systems. In this study, computer-based tests were used to measure detection performance for improvised explosive devices (IEDs). These tests were conducted before and after training with an individually adaptive CBT system. The following measures were calculated: pHit, d', /spl Delta/m, Az, A' p(c)/sub max/. All measures correlated well, but ROC curve analysis suggests that "nonparametric" measures are more valid to measure detection performance for IEDs. More specifically, we found systematic deviations in the ROC curves that are consistent with two-state low threshold theory of R.D. Luce (1963). These results have to be further studied and the question rises if similar results could be obtained for other X-ray screening data. In any case, it is recommended to use A' in addition to d' in practical applications such as certification, threat image projection and CBT rather than the hit rate alone.