The Americans taught the computer to feel the level of pain in the muzzle expression of mice.

Scientists have created a program to recognize external signs of pain in mice. Preprint article about it published In the bioRxiv.org repository.

Animals do not know how to talk, and during laboratory experiments scientists are forced to determine their state by external signs. This also applies to the level of pain they experience.

Mark Zilk of the University of North Carolina and his colleagues have developed a computer system that automatically recognizes the facial expressions of mice. It is based on pattern recognition using neural networks and allows you to measure pain level on a scale of zero to eight.

Three groups of mice were used to train the neural network: one had its gut removed a few centimeters, the other had its entire gut removed, and the third was anesthetized without surgery. After that, each mouse was recorded on high-definition video for half an hour. After the workout, PainFace managed to score on the grimace scale. As the training progressed, the accuracy of their predictions improved, and at the peak they were 0.75 and higher, meaning that they coincided with a person’s predictions in 75 out of 100 cases.

Before this system can be recommended for use in real animal experiments, other biologists will need to test whether the Zylk group’s results are indeed reproducible with other mice.