The accuracy of a person reading a mammogram is improved when their gaze is subtly shifted toward suspicious areas, and nudged around to ensure that they look at every part of the scan, according to new research.
Such "gaze manipulation" is often used in the making of movies, but could be of real value in helping to catch breast cancers, the study found.
"Using this subtle gaze direction, we can draw someone's eye around an image without distracting their viewing," said study researcher Cindy Grimm, a computer engineer at Washington University in St. Louis.
The same methods, she said, could also be applied to other tasks where someone needs to look over an image — airport security personnel looking at images from scanned luggage, for instance.
Learning from the movies
Artists and movie directors have been using the technique of gaze manipulation for decades, to make sure we see some part of the action we might otherwise miss. People tend to direct their gaze at the parts of an image that are brighter, or that have higher contrast between the darkest and lightest areas, than the rest.
Grimm, who had focused her computer graphics research on how people perceive images, said she realized that the tricks of directing someone's eye to a particular part of a picture may have uses in medicine as well.
To study how gaze manipulation could be used to help those reading mammograms, Grimm and her colleagues used 65 mammogram images with known abnormalities. The team hired an expert radiologist, and tracked his eyes while he read the mammograms. The researchers also recruited 20 mammogram-reading novices, who typically make more mistakes, or don't spot abnormalities.
The novices were divided into four groups: some read the scans as they normally would, some had their gazes subtly manipulated so they followed a gaze path similar to the expert's, some were directed with subtle clues to look only at regions suspected of being cancerous, and some had their gaze paths guided randomly across each scan.
The group that read mammograms as they normally did were accurate in reading 52 percent of the scans, and those who were guided randomly were correct 54 percent of time, whereas the group that was directed, through subtle visual clues, to follow the eye path of the expert was 65 percent accurate. The group whose eyes were drawn to abnormalities was 69 percent accurate.
"To guide their eye paths we picked spots we wanted them to look at, and made it a little bit brighter," Grimm said. "But by the time they actually pick up on that cue and look in that direction, the brightness fades and the image looks normal." Most participants, she said, didn't pick up on anything unusual about the images, or notice that their gaze was being manipulated.
Even experts vary their gaze
While the new study is proof-of-concept that gaze manipulation could improve the accuracy of those reading the scans, there are still questions about what makes one expert more accurate than another.
Anthony Maeder, of the University of Western Sydney in Australia, studies how people interpret mammograms, and said it's not as simple as gaze. "The major challenge is understanding how all the different psycho-visual factors some into play," he said.
Maeder's research has revealed that even "expert" mammogram readers don't follow the same eye gaze pattern when they read the same scan on two different occasions.
"This can be attributed to fatigue, distraction, covert attention and peripheral visual effects, visual memory of similar images, or just random human body system variations," Maeder said.
But while systems relying solely on humans, or solely on computers, both have their drawbacks, subtle gaze manipulation could be a way to link the two, Grimm said.
Computers could pick out suspicious areas of a scan and then subtle gaze manipulation could be used to make sure radiologists look at those spots — as well as the rest of the scan — when they are reading the mammogram.
"This isn't going to be a substitute for training," Grimm said. "But it could be used to keep people on task, or improve their techniques."
Related on MyHealthNewsDaily:
Copyright 2012 MyHealthNewsDaily, a TechMediaNetwork company. All rights reserved.