Pigeons can recognize cancer
A report in the journal PLOS ONE shows that pigeons can learn to recognize tumor tissue in microscopic or mammographic images after several days of training. This highlights the birds’ outstanding visual memory, as well as their ability to categorize similarities.
The brain of a pigeon is as small as the tip of a human finger, yet the tiny organ has impressive capabilities.
Pigeons can count as good as monkeys and even follow abstract counting rules. They can memorize more than 1800 different images, including human faces and letters of the alphabet.
Pathologists and radiologists need months to years of training and experience to be able to understand the tangle of forms and spots in an image and make the correct diagnosis.
To train the pigeons, researchers showed them images with or without tumor tissue, which the birds had to sort by stepping onto one of two buttons. If their response was correct, they received food reward. Later in the training the difficulty was increased: the birds had to recognize tumors also at higher magnification or in black-and-white images. The success rate was initially 50% but rose to nearly 85% after 13-15 days of training.
Another test proved that the birds not only distinguish already seen images, but can perform as good with previously unseen ones. This means, they can generalize what they have learned. When four pigeons were simultaneously shown an image, their cumulative success rate in identification reached 90% – higher than that of some beginner doctors.
In addition to this, the scientists tested pigeons in another diagnostic field: mammography. The birds learned to recognize microcalcifications and suspicious thickenings of the tissue, with a success rate of 70 – 85%. However, it was much harder for them to distinguish between benign and malign thickenings on the mammographic images. They needed weeks instead of days to learn that, and with unknown images they failed completely. “That demonstrates how hard this task is – even human experts are frequently wrong at it”, says Richard Levenson, co-author of the study.
