Pigeons taught to diagnose breast cancer on X-rays

November 20: Job opportunities for pigeons have been few and far between since electronic communication made their skills as messengers obsolete. But now it seems they could be put to work analysing medical images. So says the team who trained pigeons to distinguish between healthy and cancerous breast tissue.

Richard Levenson at the University of California, Davis, and his colleagues showed pigeons microscope images of breast tissue. Then they rewarded them when they correctly pecked a coloured button that corresponded to either cancerous or healthy tissue. After 15 daily sessions, each lasting an hour, the pigeons got the right answer 85 per cent of the time.

Pooling responses from a panel of four pigeons, or “flock-sourcing” as the researchers call it, increased accuracy to 99 per cent. The pigeons were just as good at spotting small calcium deposits associated with cancer, which appear as white specks on mammograms.

A third task – recognising cancerous breast masses on mammograms – proved too difficult. These features are very subtle and challenging even for humans to detect – radiologists only got 80 per cent correct when they assessed the images in this study.

Pigeons’ visual skills are well studied: they can recognise human faces, letters of the alphabet and even distinguish paintings by Monet and Picasso.

While doctors won’t be turning to pigeons for a cancer diagnosis any time soon, the birds could play a useful role in the development of image analysis technology. Researchers develop software that manipulates medical images so doctors can interpret them more easily, but it takes several hours to work out if the software helps or hinders a diagnosis.

Here’s where the pigeons come in, says Levenson. He says that pigeons’ sensitivity to features in medical images that are important for diagnoses make them ideal for providing feedback on several aspects aof their software development. “They can assist researchers and engineers as they innovate,” he says. Agencies