Does the application of virtually merged images influence the effectiveness of computer-based training in x-ray screening?

The necessity of computer based training for airport security screening officers to achieve and maintain a high level of x-ray image interpretation competency is well-known. During such training, x-ray images of passenger bags, similar to how they appear at the security checkpoint, are presented to the screening officers on a screen and have to be judged regarding their contents. Certain computer based training systems, such as X-Ray Tutor (XRT), apply a special algorithm that automatically merges images of fictional threat items (FTIs) into x-ray images of passenger bags. The advantage of virtual image merging is that a) a huge variety of different bag images containing different threats can be created, and b) the difficulty of threats (e.g. viewpoint, superposition and bag complexity) can be adapted to individual performance and learning progress. However, merging images virtually can lead to artifacts appearing on the resulting pictures. Therefore, the question arises if the training with merged images actually reflects the reality or if the resulting artifacts on the images actually make it easier to detect threat items during training. If this would be the case, the actual effectiveness of such a CBT would have to be questioned. The aim of this study was to investigate whether threat items that are virtually merged into bags (as they appear during training), are detected more easily than threat items physically embedded in bags and x-rayed as a whole (like at the security checkpoint). A test was conducted with screeners at different international airports. 256 images of passenger bags were presented to screeners, 128 of them contained threat items. Half of these threat images were created through virtual merging, while for the other half of the images threat items had been placed physically into the bags. Half of the used threat items were familiar to the screeners from training (XRT-library), whereas the other half of the items were new. The same study was replicated one year later. In both studies, the screeners achieved high detection performance scores in the test. However, only very small differences in detection performance for the virtually merged threat items and the physically embedded threat items were found. In fact, detection performance was even slightly higher for the physically embedded threat items. In summary, results imply that when well elaborated merging algorithms are used, small artifacts appearing on the test images influence neither the effectiveness of CBT nor the detection of real threat items in real x-ray images negatively.