Professor Richard Levenson, University of California Davis
It turns out that pigeons aren’t as bird-brained as you might think! Research published in the journal PLOS One this week shows that a flock of our feathery friends can tell - almost as well as a pathologist - cancerous or malignant tissue from normal tissue! Richard Levenson made the discovery by showing pictures to the birds of what cancerous and healthy breast tissues look like down a microscope on a slide. But why, you may well ask? Chris Smith found out...
Richard - I had been listening to the radio and heard that a colleague at UC Davis had just published a paper on the ability of pigeons to demonstrate remarkable visual memory and, for no particularly good reason, I wondered how pigeons would do in my field of pathology imaging. So I called up the investigator at Davis and he said “wonderful idea but you really need to talk to the University of Iowa researcher who actually had the pigeons, his name is Ed Wasserman ” and he turned out to be my co-author on this work.
Chris - And so you were seeing whether a pigeon would be able to do what a pathologist does when you look down the microscope and see a particular combination of colours and contrasts that you would say, with a pathologists eye, that’s something that’s cancer. Can a pigeon do that?
Richard - Exactly. And the answer to that is it would be counterintuitive if they could do it well, because pathologists are the product of years and years of training. You know, humans don’t natively have the ability to walk up to a microscope and make these distinctions between benign and malignant for example. So it would be surprising if pigeons could and, that’s the surprising result, the pigeons were very good at this.
Chris - Pigeons don’t tend to mix with microscopes well, do they? So how did you show microscope slides to a bird?
Richard - The pigeons were kept at 80 percent of their comfort level with respect to food, and the only way to get their remaining 20 percent was to perform well on the tasks that they were given. And they were given a task by being put into a box with a touch sensitive screen on one end and a pellet dispenser behind them, and an image would be presented on the screen and, after a bit of habituation, they were given the opportunity to peck a yellow or blue bar depending on whether they thought the image was cancer or benign. Of course, they didn’t understand those terms but, by being rewarded or not, they eventually learnt to discriminate these two classes.
Chris - And what was the sensitivity and specificity, to take two medical terms, in other words how good were the pigeons at picking up cancer?
Richard - The pigeons were shown a series of benign images and malignant images – equal numbers. I think it was 12 and 12 and, over the course of the first 10 days or so they went from chance to being around 80-85 percent accurate. But the danger here is that instead of learning what cancer and benign breast lesions look like, they had simply memorised the images – just by saying, oh I remember that, that was in class A. So what we did then was, after they hit 80-85 percent, we then gave them another 12 and 12 benign and malignant images they had never seen before and, the wonderful and surprising result was, they did just as well on images they had never seen before.
Chris - And how does that compare with a trained pathologist?
Richard - Not all pathology problems are the same. In other words, there are some very easy distinctions to be made, and these were of the easy sort, so pathologists should have no trouble, and all pathologists, after training, should get all of these correct. But each individual pigeon might be around 80-85 percent accurate. If you turn around and say I’m going to show the same image to four pigeons, and then see what the majority says, their accuracy goes up to around 99 percent. So we call that “flock sourcing.”
Chris - Very clever. Now, how do you think that they’re doing this?
Richard - Obviously, pigeons were not driven by evolution to become good diagnosticians. What they’re doing on this test must, in some sense, rhyme with what they have to do in the wild and vision is one of their major ways of interacting with the world. And, an important thing that they have to accomplish in the world is to avoid being targeted by predators, some of whom use extremely sophisticated camouflage and so, the pigeon has to be able to basically see through the clutter and detect things that might kill them and eat them. And so, I think that the tasks that they have to do in the wild are very similar in terms of content with what you have to do when you look at pathology images.
Chris - Now how do you see this discovery being utilised or translated. Are you seriously contemplating boxes full of pigeons in pathology laboratories to do sort of flock sourcing diagnosis?
Richard - I have to say that the vision of flocks of pigeons making diagnosis certainly has a certain strange appeal to it, but the reality is that this world’s complicated interacting forces of training, certification, regulation, financial, and legal parameters would really, I think, prevent this from happening any time soon. Nor is it probably practical at the end because pigeons are very visual, but they have very poor verbal skills and there’s a lot more information about images in patients that have to be communicated than just what thing look like.