But what happens when the radiologists are pigeons? It sounds weird, but it’s the basis of a study published Wednesday in PLOS ONE.
Armed with a touchscreen-outfitted training chamber and 16 pigeons, scientists set out to see if birds could identify malignant and benign tissue in medical images with their beady little eyes. They trained pigeons to peck at a monitor after being presented with images from mammograms. When pigeons correctly pecked a touchscreen button that corresponded with the category of the image being shown, they were given food.
Footage of pecking, mammography-reading pigeons may seem hilarious, but it’s serious business for Edward A. Wasserman, an experimental psychologist who specializes in comparing human cognition with that of other animals. Reading medical images, he said, “requires a kind of perceptual sophistication beyond mere words.” That’s where the pigeon steps in. “In some sense, the pigeon and the person are starting at the same place,” said Wasserman. “They’re equally naive.”
That naiveté quickly faded as pigeons looked at benign and cancerous slides. In the first experiment, eight pigeons were presented with histology images from mammograms — four with normal images and four with images that had been color-corrected. Pigeons had a 50/50 chance of being correct — they were presented with half malignant slides, half benign ones. But within days, it became clear that they were really, really good at their task. After just nine days, the cohort was getting more than 80 percent of its classifications correct.
By day 15, the success rate had risen to 85 percent.
Wasserman began to worry that perhaps the pigeons had simply memorized the slides, so after 15 days he began to present them with rotated versions as well. Five days later, the pigeons moved on to “novel stimulus testing.” They were shown the original trial slides and additional, brand-new slides. They also viewed slides at different levels of magnification.
And here’s where the power of pigeons as medical interpreters comes into play: The birds were really good at applying the knowledge from the familiar slides to examples of benign and cancerous tissue they’d never seen before. Their recognition of new slides was only a few percentage points below their success rate for familiar ones. The whole flock was even more effective as a group.
In another experiment, pigeons developed roughly equal abilities to spot microcalcifications — small changes in breast tissue that can sometimes be the only sign of cancer. They weren’t as good at classifying suspicious masses in a third study, but neither are experienced radiologists at the same or a similar task.
So should human radiologists watch their back for oncoming pigeons? Not exactly. Since similar studies on the effectiveness of humans at reading medical imagery don't exist, it’s not possible to compare the pigeons’ skills to those of humans. Robots are much more likely to replace human radiologists than pigeons — but the experiment could eventually help inform new technologies for image interpretation.
It could also make future research cheaper, said Wasserman, who notes that pigeons “might be able to be an effective human stand-in” in additional experiments related to medical imagery. It’s not the first time pigeons have been suggested as human stand-ins for life-or-death situations — B.F. Skinner, the pioneer of behavioral psychology, trained the birds to guide rockets during World War II.
Maybe the true moral of the mammogram-reading pigeons has more to do with respect than their real-world radiological prowess. “The pigeon has kind of a bum rap,” said Wasserman, who insists that the birds are interesting and challenging enough to merit serious study. After all, he said, “humans are not the only intelligent animals walking and swimming and flying on earth.” Or, for that matter, looking at mammograms.