'If you don't find it often, you often don't find it.'
In findings that highlight the difficulty of making air travel safer, researchers at Brigham and Women's Hospital report that when the target of a visual search -- like a gun or a knife -- occurs only rarely, it is likelier to escape notice.
It was already known that detecting weapons in luggage -- or tumors in breasts, for that matter -- are tremendously difficult visual-search tasks, too complicated to be fully automated and exhausting for fallible human eyes.
But it had not been shown experimentally before that human performance could decline so very miserably when search targets were rare, according to the lead researcher, Jeremy M. Wolfe.
''This is a red flag that says we need to look at any situation where the targets are rare and the searches are difficult, to make sure people are spending enough time to do the search," said Kyle R. Cave, a vision researcher at the University of Massachusetts at Amherst. When the people took more time to look at each image, their performance was more accurate.
Wolfe emphasized that the study, which was funded by the federal Transportation Security Administration and published in the latest issue of Nature, is basic research with uncertain implications in the real world.
But it is suggestive enough that ''we plan to work with the TSA people to ask whether this is a real problem, and if it is a real problem, to see if we can try some things to fix it," Wolfe said.
Since the 9/11 attacks, federal authorities have been funneling money and effort into research on human and machine vision, seeking science-based insights into ways to keep the country's airports safe.
The ''TSA is dedicated to the research and development of improved technologies that will lead directly to increased threat-detection capabilities," Joshua Rubinstein, a TSA engineering research psychologist, wrote in an e-mail. ''In many cases, we must conduct . . . basic research to discover which psychological strengths and weaknesses people have that could affect screener performance."
In the Nature study, Wolfe and colleagues asked volunteers to look at 2,400 images, each containing a set of objects against a busy background, and report whether they saw a tool, like a wrench or a hammer.
In some sequences of images, the tool would be present half the time, and the volunteers missed only 7 percent of them. But when the tools appeared only 10 times in every 1,000 images, their error rate rocketed to 30 percent.
''If you don't find it often, you often don't find it," Wolfe said.
The problem, Wolfe said, is not that people doze off or tune out; it is that when targets are rare, people have a built-in tendency to give up too quickly when hunting for them, and thus are much likelier to miss them.
After being told they had failed to notice a tool, the volunteers would slow down their search a bit, but then quickly speed up again after a few more images with no tools.
Increasing their motivation by using a game-style point system to keep track of their performance brought no improvement. It also did not help to mix the rare tools in with targets that would appear more commonly and get spotted more easily. Even as the volunteers performed well with the more common targets, they still missed a lot of the rare tools.
Already, airport security authorities occasionally project fake images of weapons onto the screens that human scanners monitor, as a check. But the Nature study appears to indicate that such projected images may do little to affect performance, because they're added to the images so infrequently.
The fake images may serve other purposes, said Howard Egeth, a Johns Hopkins University professor and vision researcher: ''For one thing, you can keep an eye on baggage inspectors and make sure they have a suitable catch rate, find out who are the good ones and who are weak. But as far as solving the problem of [rare targets], Jeremy's work suggests it's going to be harder."
One of the central challenges, Egeth and others noted, is finding ways to improve the screeners' performance without making the snaking lines at airports significantly longer.
So what can be done? Ways must be found to force scanners to look longer, Egeth said.
If the experience of mammogram-readers is any guide, training and increased motivation can make a difference, said Dr. Steven Seltzer, chairman of the Radiology Department at Brigham and Women's.
Now that mammograms are done digitally, it has also helped to get ''computerized second opinions," running the data through computers that use algorithms to check for areas of possible interest, he said. Another tactic tried by some is to make sure two doctors check each mammogram in hopes that one will catch something the other missed.
But still, he said, the task is so difficult that mammogram-readers still miss about 10 percent of early cancers; and false positives, in which a spot is biopsied and found to be nothing, are also common.
''An important point is that there is no panacea," Seltzer said. ''This specific perceptual challenge of detecting low-prevalence targets is a tough job, and it's inevitable that human observers will make mistakes."
Carey Goldberg can be contacted at goldberg@globe.com. 