Many abnormalities missed in clinical practice may, in fact, be detected by radiologists without their knowledge. Researchers from the University of Massachusetts Medical School measured physiological arousal while radiologists read studies, demonstrating the potential for a second-look warning system.
These results demonstrate the potential for a second look warning system, where a radiologist would be prompted to re-check a study they declared normal if their arousal increases, according to an RSNA 2016 presentation given by Gregory Di Girolamo, PhD, associate professor of psychology at UMass.
Even as imaging technology has improved, error rates for radiology remain around 30 percent—similar rates to what was experienced 50 years ago when the specialty was relying on just x-ray. This is a testament to how plain difficult it is to efficiently read studies and identify abnormalities. While clinical decision support and computer-aided detection tools strive to make radiologists more accurate, Girolamo believes there is another untapped resource.
“A radiologist will be exposed to an average of 50,000 reads a year, meaning the brain is exposed to quite a lot of abnormalities—some of which we catch wind of, and some of which we don’t,” he said. “There’s going to be learning accrued in both the conscious and unconscious brain.”
“We’re looking for markers that show detection of abnormalities outside the radiologist’s conscious awareness,” Girolamo added.
One of the first signs of physiological arousal in humans is pupil dilation, so Girolamo used high speed video eye tracking to track radiologists’ gaze while they read axial CT exams, getting a precise measure of where they’re looking, when they’re looking, and how big their pupils are.
To identify all conscious thought, Girolamo had the radiologist subjects left click on lung nodules and right click on areas where they considered identifying a lung nodule.
“The radiologists were not shy about this, on average they right clicked about 60 times per image—clicking on everything they remotely think might be an abnormality,” said Girolamo.
The researchers used studies with difficult-to-identify lung lesions, resulting in a miss rate of 60 percent.
Naturally, radiologists showed more arousal as measured by pupil diameter when they discovered an abnormality and identified it by left clicking. However, they also showed greater physiological arousal when there was an abnormality present and they weren’t consciously aware of it at all.
“The most interesting part of the data is when the nodule is present and they don’t even look at it—their eye doesn’t land on it during the entire screening—their physiological arousal goes up. Even when they’re not looking at it, these unconscious processes are firing.”
This phenomenon could be leveraged to assist radiologists, according to Girolamo.
“We could say, ‘Your arousal has increased when you’re looking at this image, are you sure nothing’s there?’” said Girolamo. “The most straightforward way of doing this would be attaching an Apple Watch to these radiologists and notifying them when their heart rate changes, but the eye tracking we’re using is remote and unobtrusive—it could be a camera at the bottom of the screen, requiring nothing exceptional of the radiologist.”