According to a study recently published in PLOS ONE, reading medical images is for the birds—but it’s not quite how it sounds.
Richard M. Levenson, MD, of the UC Davis Medical Center department of pathology and laboratory medicine, and colleagues set out to see if pigeons could discriminate between benign and malignant pathology and radiology images, all without any verbal cues or instructions.
This may sound far-fetched, but the authors explained that pigeons share numerous visual system properties with humans. In addition, the animals have been impressing scientists for decades with their ability to learn and retain information.
“Research over the past 50 years has revealed that pigeons can be prodigious discriminators of complex visual stimuli, and are able to detect or discriminate: foreground from background; misshapen pharmaceutical capsules; letters of the alphabet; basic object categories such as cats, flowers, cars, and chairs; identities and emotional expressions of human faces; and even paintings by Monet vs. Picasso, among many other impressive feats,” Levenson et al. wrote. “Pigeons’ visual memory is also outstanding, as they can recall more than 1,800 images.”
Sixteen pigeons—Columba livia, to be exact—were used in the experiments. Each pigeon was trained in a 36 cm x 36 cm x 41 cm chamber, pecking a touchscreen connected to a food delivery system with their beak. The authors noted that all housing and training was approved by the University of Iowa’s Institutional Animal Care and Use Committee.
The study’s results showed that pigeons were quickly able to tell the difference between benign and malignant breast tissue, reaching 85 percent accuracy within 15 days of training. Unsure if the birds were truly learning to classify images or just memorizing specific images, the researchers switched things up and mixed in brand new images. The pigeons still excelled.
“Overall, the results show that pigeons were capable of accurately discriminating between benign and malignant breast tissue samples at different levels of magnification, and that they could reliably transfer that performance to novel tissue samples at the same level of magnification,” the authors wrote. “While rotating previously learned images, or presenting similar images, but at higher magnifications than the birds had been trained with, did monetarily affect accuracy, the birds were capable of overcoming these immediate negative effects with the help of additional differential reinforcement-based training.”
The pigeons also displayed the ability to differentiate between images with clusters of microcalcifications and those without. When it came to discriminating mammograms containing masses, however, the bird struggled. In fact, it took nearly 12 weeks of training for some of the pigeons to achieve 60 to 80 percent accuracy, and one never got the hang of it.
Pigeons aren't the first animals to have their diagnostic ability put to the test. The authors listed examples of others being used as “front-line diagnosticians” in the medical community—researchers have tried to leverage a dog's excellent sense of smell to sniff out cancer, for example—but concluded that pigeons weren’t quite ready for such a responsibility.
“We are not (yet) proposing such a role for pigeons,” the authors wrote. “Instead, we have shown that pigeons can be effective, tractable, relevant, informative, statistically interpretable, and cost-effective medical image observers. Their performance on visual discrimination tasks might prove useful in guiding basic vision research into feature saliency and could play a key role in the development of computer-assisted medical image recognition tools.”