by ANDREW M. SEAMANn
The pigeons’ training environment included a food pellet dispenser, a touch-sensitive screen which projected the medical image, as well as blue and yellow choice buttons on either side of the image. Pecks to those buttons and to the screen were automatically recorded. PHOTO/Copyright Univ. Iowa/Wassermann Lab
Years of schooling and training are needed to teach pathologists and radiologists to spot cancer on medical images, but a new study finds that pigeons can be about as accurate as these professionals, with the help of a few food pellets.
People don’t have to worry about bird brains diagnosing their cancers any time soon, but the study’s lead researcher says pigeons may have a future standing in for pathologists and radiologists in the kinds of mind-numbing studies of new technologies that involve examining thousands of images.
“If you showed me 10 images, I’d be ok,” said Dr. Richard Levenson, of the University of California Davis Medical Center in Sacramento. “But if you showed me 10,000 images, I would get irritated. Birds don’t have the luxury to be irritated.”
Dr. Levenson thought of experimenting with birds when he heard about work by Dr. Edward Wasserman of the University of Iowa and colleagues that found pigeons’ visual recall is similar to that of humans.
Dr. Levenson teamed up with Dr. Wasserman for the new study, which put pigeons through three sets of tests. Each experiment showed the birds a different type of image: actual breast tissue samples with and without cancerous masses, mammogram images with and without calcifications, and mammograms with benign or cancerous masses.
The birds were taught to spot cancer and potentially cancer-linked calcifications over several days by being rewarded with a pellet of food each time one selected the correct button indicating the image was cancer free or had a malignancy.
To make sure the birds were not simply memorizing which images were cancerous and not cancerous, the researchers also showed the pigeons new images.
Among the images of actual tissue samples, the birds’ accuracy rose from 50% (equivalent to chance) to about 85% 15 days later. They performed just as well when they were shown new images.
Scientific America for more