Pigeons Against Breast Cancer

November 19, 2015
Greg Kennelty

Despite the small size of the pigeon brain, our Columba livia counterparts have performed just as well as humans in distinguishing between normal tissue and breast cancer tissue in a study performed by researchers at the University of Iowa and the University of California, Davis.

Despite the small size of the pigeon brain, ourColumba liviacounterparts have performed just as well as humans in distinguishing between normal tissue and breast cancer tissue in a study performed by researchers at the University of Iowa and the University of California, Davis.

According to a press release1from the University of Iowa, the pigeons were initially taught which digitized slides and mammograms were malignant and which were benign. When shown a different set of slides, the pigeons were able to recall characteristics that distinguished between slides that were benign and those that were not.

Researchers taught the birds to distinguish between cancerous and noncancerous images by rewarding the birds with food when a correct choice was made, otherwise known as "operant conditioning." The slides used were from a set of routine cases at the UC Davis Medical Center.1

"These results go a long way toward establishing a profound link between humans and our animal kin," said Edward Wasserman, PhD, professor of psychological and brain sciences, University of Iowa, in the press release. "Even distant relatives — like people and pigeons — are adept at perceiving and categorizing the complex visual patterns that are presented in pathology and radiology images, surely a task for which nature has not specifically prepared us."

The study,2published in the journalPLOS One, stated that its purpose was to understand how the process of visually identifying breast cancer unfurls and what image characteristics and properties are crucial for accurate diagnostics. The abstract stated that pigeons were chosen due to their visual system similarities with humans.

According to the study, the pigeons were split into three experiments. The first experiment involved a cohort of 8 pigeons, while the second and third involved 4 pigeons each. In the first experiment, 4 pigeons were trained with normal images and the other 4 pigeons were trained with hue- and brightness-balanced monochrome images. In the second and third experiments, all of the training images were of the same type for each of the 4 birds.

The first experiment involved a cohort of 8 pigeons that were subjected to 244 breast tissue sample images of varying malignancies. These stained slides were split into three groups of 48 that were digitally scanned using an Aperio whole-slide scanner at a maximum resolution setting of 20x and then pared down to provide representative fields at nominal 4x and 10x resolutions. Each image was originally 388 x 388 and resized to 308 x 308.

The second experiment featured a cohort of 4 pigeons studying 40 regions of interest that were taken from mammograms, 20 of which contained subtle clusters of microcalcifications and 20 without clusters. The third experiment also featured 4 pigeons but studied mammogram images without microcalcifications. According to the study, 40 images were used in total, with 20 samples containing malignant masses and 20 samples containing benign masses.

“The pigeons learned to discriminate benign from cancerous slides as fast in this research as in any other study we’ve conducted on pigeons in our laboratory,” said Wasserman, coauthor of the study, in the press release. “In fact, when we showed a cohort of four birds a set of uncompressed images, an approach known as 'flock-sourcing,' the group’s accuracy level reached an amazing 99% correct, higher than that achieved by any of the four individual birds.”

Success rates improved over the course of the study.2In the first experiment, the pigeons had a 50% rate of success at the outset, and this rate rose to 85% over the course of the15-day training, making the results statistically significant (P= .001). Over the next 5 days of the first experiment, the pigeons were trained with all rotations of the original image set. On day 1, the pigeons' accuracy to the original 0° averaged an 86% rate of success, and accuracy to five newly rotated slides averaged 77% (P= .161). Accuracy over the next 5 days showed 88% accuracy in the 0° slides and 83% accuracy in the rotated slides.

In the second experiment, accuracy rose from 50% to over 85% in the pigeons' 14-day training. Rotation and flipping of the images affected accuracy nonsignificantly, according to the study, and with additional training, success rates rose to 86% in the nonrotated images and 82% in the rotated images.

In the third experiment, the pigeons' accuracy scores began as 71% in nonrotated images and 50% in rotated images, with these scores reaching 74% accuracy in nonrotated images and 44% in rotated images over the course of 14 days.

“Research over the past 50 years has shown that pigeons can distinguish identities and emotional expressions on human faces, letters of the alphabet, misshapen pharmaceutical capsules, and even paintings by Monet versus Picasso,” said Wasserman. “Their visual memory is equally impressive, with a proven recall of more than 1800 images.”

With this information, physicians could potentially put pigeons to use as aids to researchers in exploring image quality and the impact of color, contrast, brightness, and image compression artifacts on diagnostic performances, according to the press release.

Despite their success, the study stated that while the pigeons were helpful in identifying breast cancer, they were unsuccessful in identifying mammographic densities beyond rote memorization, as shown in the third experiment.

“The data suggest that the birds were just memorizing the masses in the training set, and never learned how to key in on stellate margins and other features of the lesions that can correlate with malignancy,” said Richard Levenson, MD, FCAP, professor and vice chair for strategic technologies in the Department of Pathology and Laboratory Medicine, UC Davis Health System, and coauthor of the study. “But, as this task reflects the difficulty even humans have, it indicates how pigeons may be faithful mimics of the strengths and weaknesses of humans in viewing medical images.”

Looking toward the future of utilizing pigeons in breast cancer, Levenson said he had a few things in mind.

“The birds might also be able to assess the quality of new imaging techniques or methods of processing and displaying images—without forcing skilled humans to spend hours or days doing detailed A:B comparisons to figure out whether certain innovations are in fact better or worse than current methods,” said Levenson in an interview withTargeted Oncology. “We are currently thinking about exactly which problems to go after, in radiology and pathology, and don’t have specific projects lined up yet.”

References

  1. One very brainy bird.http://www.uiowa.edu/2015. Available at: http://itsnt774.iowa.uiowa.edu/distrib/lewis stuff/pigeons-medical imaging-copy.pdf. Accessed 18 November 2015.
  2. Levenson R, Krupinski E, Navarro V, Wasserman E. Pigeons (Columba livia) as trainable observers of pathology and radiology breast cancer images.PLOS One. 2015. doi:10.1371/journal.pone.0141357.