Radiologists Can Detect Subtle Breast Cancer In Split Second, Study Finds

A radiologist checks mammograms, an advanced imaging screening that promotes early detection of breast cancer, at The Elizabeth Center for Cancer Detection in Los Angeles, May. 6, 2010. (Damian Dovarganes/AP)
A radiologist checks mammograms, an advanced imaging screening that promotes early detection of breast cancer, at The Elizabeth Center for Cancer Detection in Los Angeles, May. 6, 2010. (Damian Dovarganes/AP)
This article is more than 4 years old.

Years ago, Malcolm Gladwell wrote a mega-bestseller called "Blink," arguing that sometimes we make better choices unconsciously, in the blink of an eye, than we do after careful consideration.

New research suggests that there can be a certain "Blink" element to breast cancer screening — not in medical practice, but in doctors' brains. It found that radiologists can be strikingly good at detecting even subtle abnormalities on a mammogram at first glance, in a split second.

I spoke with Dr. Jeremy Wolfe, senior author on the paper and director of the Visual Attention Lab at Brigham and Women's Hospital. (He's best known as an expert on how we look for things that we want to find: where we left our keys, cancer on scans, bombs in luggage.) Our conversation, edited:

CG: How would you sum up what you found?

JW: The paper really started with the experience we had talking with radiologists — and indeed, any sort of expert visual search experts: It could be the people at the airport, or people who look at spy satellite images. These people often say, 'Sometimes when the image comes up, I just know it's bad.'

The paper's lead author, Karla Evans, who's now at the University of York, wanted to know whether that notion of getting something out of the first glimpse was for real. So we took a collection of images, half of them that had signs of cancer in them and half of them that didn't, and we flashed them for half a second at radiologists. We asked them simply to say whether this woman should be recalled [for further examination].


If they were just guessing, they would be just as likely to say a normal one was abnormal as an abnormal one. But in fact, they could beat chance very reliably. You would never, ever, ever want to do real breast cancer screening in half a second, of course. But they were getting about 75 percent correct.

What we really wanted to find out was: Where is that signal? One possibility was that it's the symmetry between the two breasts. Typically, radiologists look at a pair of breasts, and if the two breasts are asymmetrical, that's a sign that something is amiss. But we repeated the experiment using just a single breast, and they did just fine. So it's not the symmetry.

And that led to what I think is the most interesting piece of this finding. When we flashed just the breast with the cancer in it, they did as well as they did with two breasts. But when we flashed the other breast, the apparently uninvolved breast, they were still beating chance. They could say that this image was apparently not normal, even though the cancer was in the other breast.

That is wild.

It's less wild than one might think. I was talking with colleagues of mine, a molecular researcher. And she said, 'Look, I've been looking at molecular signs of cancer, and we can find those in the breast that has cancer, and lo and behold, we can find those in the other breast, too,' — suggesting that there's something about the breasts of women who develop cancer that is abnormal even before there's a cancer there. So while it's deeply surprising that you can find it in half a second, it's not ridiculous. There is biological evidence that something is going on in that uninvolved breast.

The tantalizing thing is that we don't know what that signal is, though?

No, we're really not sure. One of the experiments we did gets at that a little bit. We did what's called 'spatial frequency filtering' of the breast images. If you lower that filtering, you basically blur the image, so you end up with sort of blobby structures in the breast. And if you go in the other direction, so called high-pass filtering, what you end up with is almost cartoon-like fine detail. People in my trade would have guessed that if there's a signal that you're getting out very quickly, it would be in that low-pass blurred image. There's evidence for that elsewhere. But in this particular case, the information seems to be in that high-pass fine detail.

So we don't know exactly what that signal is, but if you flash the image up for half a second and you only give people the fine detail of the fibrous structures of the breast, they do almost as well as with the whole image. If you blur the image, they do markedly worse. So we know something about where that signal is hiding.

Another really interesting piece of this ability to see something in half a second: The radiologist can beat chance at saying whether the breast is abnormal or not, but if you say, 'OK, we know you're kind of guessing but just put a mark where you think the problem is,' they are completely at chance. The signal seems to be what we would call a global signal — the texture of the breast as a whole — that looks abnormal to an expert radiologist. It's not that sometimes you got lucky and you happened to be staring at the cancer.

So what next?

One interesting next experiment would be to take images from women who developed cancer, and look at their images from four or five years ago, and say, 'OK, if we flash those at a radiologist, can the radiologist beat chance before the cancer ever shows up?'

And overall, the objective is to make breast cancer screening better?

Right. This goes back to the problem of looking for something that's very rare. Radiologists are very good at finding cancers; they probably find 90 percent of cancers that show up in screening. But they recall about 10 percent of women for further exams, and because cancer is so rare, almost every one of those recalls is a false alarm. If there's anything we could do that would cut down that error rate, we would cut down the worry of women and their families, costs, and potentially we would save lives.

How would you relate Gladwell's 'Blink' to these findings?

You could get the impression from 'Blink' that you really want to make the decision about your who to marry or your breast cancer on the basis of the first glimpse, but that would be a huge mistake. What he's pointing out, and what the science points out, is that there is information in that first glimpse. But if you go with that bit of information, you'll be only 75 percent correct. And there are a lot of decisions in your life where you really, really want to be better than that. Taking a second look would be better than going with the initial blink.


Carey Goldberg Twitter Editor, CommonHealth
Carey Goldberg is the editor of WBUR's CommonHealth section.