The blink of an eye — that's how fast expert radiologists are able to distinguish an abnormal mammogram from a regular one.

Visual attention researchers from Brigham and Women’s Hospital (BWH), along with colleagues from the University of York and MD Anderson Cancer Center, set out to determine not only how quickly radiologists could detect telltale variance in mammogram exams but also what kind of factors informed that knee-jerk interpretation.

They found that radiologists made accurate assessments in 75 percent of the cases and, faced with the most challenging cases, were still correct more than 60 percent of the time.

Previous research by Jeremy Wolfe, Ph.D., senior author of the study and director of the Visual Attention Laboratory at BWH and first author, Karla Evans, PhD, now at University of York has shown that while expert radiologists can make relatively accurate split-second analysis of mammograms, non-expert radiologists can not. In the current study, Wolfe and his team tested whether breast tissue symmetry, breast density, resolution or other factors were helping radiologists successfully detect cancer.

“We showed mammograms or portions of mammograms for .25 to .50 seconds to expert radiologists and asked them to say whether they would recall the woman on a scale from 0 to 100,” Wolfe told HCB News.

“From those data, we can extract ‘signal detection’ measures of how well they do. For instance, if you treat everything above 50 [on the scale] as a yes, recall her decision and everything below as a no, then you can see if observers say yes more often when the case is abnormal,” he said.

They found that breast symmetry and breast density were not contributing factors but finer details of the breast tissue did help radiologists make correct diagnoses more frequently.

"Radiologists can have ‘hunches’ after a first look at a mammogram. We found that these hunches are based on something real in the images,” he said. “We know that the ‘high frequency’ finer detail is important but we don’t know what the observers are using as a cue.”

Even when the images of abnormal breast tissue did not directly capture a cancer lesion, or when the images were taken from the contralateral breast — the breast on the other side of the body — the radiologists were still able to detect breast cancer better than chance would have.

“It's really striking that in the blink of an eye, an expert can pick up on something about that mammogram that indicates abnormality,” Wolfe said. “Not only that, but they can detect something abnormal in the other breast, the breast that does not contain a lesion.”

Learning the signal that experienced radiologists use to detect breast cancer could help researchers refine and improve computer-aided detection (CAD) systems that help in medical screening and could be incorporated into training for physicians to improve breast cancer detection.

The researchers are also interested in exploring whether other medical image experts such as dermatologists or pathologists can use signals to detect problems or abnormalities.

Rad Oncology Homepage