Radiologists looking for lung cancer nodules maintain high accuracy rates even if they've been given false expectations, but the resulting change in search methods could cause unexpected errors, a new study has shown.
The work, by an international research team, appears in the latest edition of Radiology.
Scientists have long suspected that a clinician's ability to spot abnormalities is affected by what they expect to find. It also seems the less they expect to see abnormalities, the less likely they are to spot one when it does appear.
In this study the researchers tested whether accuracy rates improved or declined if radiologists were told to expect a different number of cancer nodules in chest x-rays than were actually present.
Lead author Warren Reed of the Sydney University Health Sciences Faculty, said the study was done with 22 highly experienced radiologists. They were divided into three groups and given different expectations of how many nodular lesions they would find in a set of 30 x-ray images. Half of the images showed healthy lungs and the other half had been digitally altered to contain the nodular lesions which normally indicate lung cancer.
Reed explains that digitally altered images were used, rather than actual images of lesioned lungs, because it eliminated any uncertainty as to how many lesions were present.
Each altered image contained up to three pulmonary lesions and on the first read the radiologists were told to expect that number of abnormalities on 15 of the 30 images.
The three groups of radiologists were then shown the same set of images a second time and were told to expect nine, or 22 abnormal images; or they were not told how many to expect.
Reed says there was no significant change in the number of abnormalities which observers noticed after being given a prior expectation.
Looking too long reduces acccuracy
"What tended to change was the way they looked at the images," Reed says.
The more lesions a radiologist expected to find, the longer they spent on the image. Eye movement also increased when observers had been given high expectations. The number of fixations, that is longer periods looking at one spot, also increased.
"It was quite a revelation. You'd assume that a radiologist would sit down and look at things in the same way, time and time again," says Reed.
Reed says radiologists normally spend about 12 seconds looking at an x-ray.
"Experts tend to look at images for less time and have fewer fixations," he says.
"That's almost counterintuitive. A lot of people think they'll sit down and look carefully for a long time. Actually the reverse works. The longer you spend looking at an image the more chance that confusion and uncertainty will set in."
Reed says that could lead to fatigue when reading large numbers of images.
He says the study raises several other questions he hopes to address.
"Anecdotally, from radiologists," Reed says, "a low number of positives can lead to lapses in concentration and attention."
Reed says that might lead to lower accuracy in mammography interpretations, for example, where a radiologist might expect to find only one positive in 200.
He says that if observers do not find an abnormality fairly regularly, they often fail to notice the abnormality when it does appear.
"You look at most breast screening services. Lesions can occasionally be missed. We wish to fully understand how this happens and suggest solutions," Reed says.