Forget AI. Pigeons Can Spot Some Cancers as Well as Human Experts

The humble bird may help scientists understand how experts learn to see disease.

Long before artificial intelligence began promising to improve medical diagnosis, a small gray bird was already having some pretty good results.

In 2015, common pigeons (Columba livia) learned to do something that almost sounds absurd: they looked at images of breast tissue and learned to tell the difference between benign and malignant samples. With training, their accuracy became comparable to experts. In fact, after roughly two weeks of training, individual pigeons reached about 85% accuracy in distinguishing cancerous from non-cancerous breast tissue.

But it gets even better: when they put their minds together, they reached a whopping 99% accuracy in one diagnostic task.

A Bird Brain Built for Seeing

A pigeon’s brain is tiny, likely no larger than the tip of a human index finger. But small doesn’t mean simple — and it turns out, they’re really good at recognizing patterns. Pigeons have been shown to recognize human faces and emotional expressions, letters of the alphabet, distorted pharmaceutical capsules, and paintings by Monet and Picasso.

In the 2015 study, led by Professor Richard Levenson of the University of California, Davis, and Professor Edward Wasserman of the University of Iowa, pigeons were trained to inspect digitized images of breast biopsy slides. Each bird faced a screen and learned to peck one button for benign tissue and another for malignant tissue. Correct answers were rewarded with food pellets.

For starters, their memory turnet out to be remarkable. Pigeons could recall over 1,800 images, sometimes even images that were remarkably similar. But the important part wasn’t the memorizing, it was the recognition of new images they had never seen before.

The birds were looking for visual features of malignancy (darker, crowded, abnormal-looking cell patterns). In the selected breast biopsy and mammogram tasks, the trained pigeons approached the performance of human experts. But when the researchers combined the decisions of four birds (which funnily enough, they call “flock-sourcing”), it pushed performance to 99% accuracy.

Granted, these were controlled images, and they failed on the more difficult task of judging suspicious mammographic masses. But it’s a stunning result.

×

Get smarter every day...

Stay ahead with ZME Science and subscribe.

Daily Newsletter
The science you need to know, every weekday.

Weekly Newsletter
A week in science, all in one place. Sends every Sunday.
No spam, ever. Unsubscribe anytime. Review our Privacy Policy.

Thank you! One more thing...

Please check your inbox and confirm your subscription.

RelatedPosts

Distinguishing cancer cells using fractal geometry offers faster diagnosis
Innovative pill-on-a-thread offers new hope in the fight against esophageal cancer
Fancy a cup of pigeon milk?
Lung cancer may be detected with a cheek swab

It’s been over a decade since this study came out and we don’t have pigeon teams looking at mammograms. But this all begs the question: why are they so good at it?

They Come Pre-Trained

Well, for starters, the skill makes evolutionary sense. A pigeon survives by scanning messy scenes quickly: gravel, soil, leaves, shadows, food, predators. Its world is full of tiny visual differences that matter. A seed hiding among pebbles is a life-sustaining target, while a small movement overhead may signal danger.

The birds viewed images of breast tissue at different magnifications, much as a pathologist examines slides under a microscope. They learned to classify full-color images, and some were also trained on modified images in which color cues were reduced. They were tested on different levels of image compression, too — a practical concern in modern digital pathology, where slides may be stored, transmitted and viewed on screens.

The pigeons could generalize across these changes. Their performance did not vanish when magnification shifted or images were altered. That suggested the birds had learned something about the structure of the tissue itself, and maybe it’s their eye-in-the-sky scouting that pre-trains them for that.

Nearly a decade after the original pigeon experiment, a 2024 study proposed just that: pigeons arrive in the lab already “pre-trained” by the view from the air. As they fly, pigeons constantly take in landscapes of fields, roads, rooftops, rivers and other repeating textures. Those bird’s-eye scenes, the researchers argued, may have a hidden visual kinship with stained tissue slides. Both are crowded, patchy worlds of color, texture and irregular boundaries.

Birds Eye View

To test the idea, the team trained neural networks on a large aerial-image dataset called BirdsEyeViewNet, then transferred that learning to the same breast histopathology and mammogram tasks used in the pigeon study. The result echoed the birds’ own performance: the aerially pre-trained model did well on histopathology but failed on the mammogram mass task, much as the pigeons had. Still, this just shows that there’s a lot of similarity between the two tasks, it doesn’t prove that’s why pigeons are good at it.

A decade later, the pigeon study is still interesting, not as a joke about birds replacing doctors, but as a reminder of what diagnosis really is. Before it becomes a report, a treatment plan, or a prognosis, it begins as an act of seeing: noticing a pattern, separating signal from noise, recognizing when tissue has gone wrong.

And as medicine leans harder into AI, that lesson matters. The future of diagnosis will not belong to pigeons pecking at screens. But the humble pigeon may still help us build better machines, train sharper doctors, and understand how eyes (human, animal, or artificial) learn to see disease.