The solution is to randomly insert images in the x-rays that are false positives. To keep screeners alert, these false positives need to happen about 5% of the time.
That's one reason that every now and then, a bag gets sent back through the x-ray machine without being opened, despite not having any contraband in it.
The Freakonoics podcast last week talked about boredom. One of the ways to make these types of tasks more interesting, and probably improve the effectiveness is to "gameify" it. Maybe something like this would work: As each tray goes through, the screen displays the image and the screener presses a red button for "bad stuff" and a green button for "everything is ok". Each time the red button is pressed but no "bad stuff" is found in later screening, the "player" looses points. Each time the red button is pressed and "bad stuff" stuff is found in later screening, the screener gets points. Now comes the part that makes it work: since we want to deduct points (presumably a lot of points) on occasions when the screener lets "bad stuff" pass with a press of the green button, but for actual items carried by travellers we do not know who might be carrying the bad stuff so we insert known images of "bad stuff containing" trays (and maybe known images of trays with no "bad stuff"). When the screener encounters these images, they can be scored appropriately, and these known images allow for calibration of the screener's effectiveness. The system can add the "virtual trays" at a rate that keeps the screeners alert, and the penalties for flagging a "clean" tray as dirty can discourage being overly cautious.
Tie the points into some small reward like free lunch, or better yet have them contribute towards some team points totals to get the powers of competition and tribal grouping to focus attention and effectiveness would probably go up as well. Remove people from the screening job who cannot do it well enough.
If implemented, we would probably have to work hard at preventing cheating - making the virtual tray images fit into the regular images in such a way that the screener cannot tell which is which, making sure that other members of the "team" do not somehow help the screener in a way that messes the system up, etc.
To generate "no bad stuff" images for the system you would need to pack some bags with no bad stuff in it, but for "bad stuff" images you could use both bags packed with "bad stuff" in addition to images of actual traveller bags that got flagged by a screener and then were found to have "bad stuff" in them.
All of this presupposes that there is an actual desire to effectively find all the things that are prohibited. I suspect that this is not actually the case. Sure, most people are happier if they feel that knives and guns are not carried by any travellers, but "the powers that be" probably all know that the list of prohibited items is needlessly long. Actually ferreting out all of these things probably would slow things down way too much for no actual improvement in safety. To catch all of the "bad stuff" containing virtual images for example, you probably need to flag every bag that might possibly have a "bad thing" obscured by another thing - it is probably impossible to catch all of the "red" images without also flagging lots of "green" images, which would result in lots of further screening of "clean" trays.