
A new artificial intelligence (AI) estimates how much pain a mouse is in by analysing its facial grimaces. Researchers hope the tool will give scientists a better understanding of how effective a pain-relief drug is and increase the number of pain-related experiments they can analyse at one time.
Many pain treatments are first tested in mice by assessing a rodent鈥檚 discomfort after drug administration. But these assessments can be subjective and differ between researchers, says at the University of North Carolina.
鈥淭he new person you train up, say they鈥檙e a grad student, may not score the mouse鈥檚 pain in precisely the same way as you,鈥 he says.
Advertisement
In an effort to overcome this issue, Zylka and his colleagues devised a tool to do the job, focusing on so-called black-coated mice. Also known as C57BL/6, these animals are commonly used in pain research.
They recorded more than 270 videos of black-coated mice in various settings, including after a laparotomy, which is an abdominal cut. The researchers specifically looked at the rodents鈥 upper face, whiskers, nose and ears.
Next, they took hundreds of frames from each video, scoring them based on a commonly used pain grimace scale for mice. Altogether, more than 70,000 frames were scored.
The researchers then fed this data into a machine-learning model.
The resulting AI, coined PainFace, allows researchers to upload a video of their black-coated mouse experiment. The AI then judges how much pain a given rodent is in on a scale of 0 to 8.
When put to the test, the AI was no worse at judging a mouse鈥檚 pain than when the rodent鈥檚 discomfort was assessed by a human. Moreover, the AI was at least 100 times faster, says Zylka.
He hopes the tool will one day change the way new pain-relief drugs are tested.
鈥淚n a lab, you may take a 30-minute video of a mouse and then pain-score about 10 frames,鈥 says Zylka. 鈥淲hereas now, you can take that same video and score thousands of frames and it basically takes no time.鈥
The tool will also make it easier to scale up the number of animals and drug doses you can study at one time, he says.
Although only tested in black-coated mice to date, Zylka says there is no reason why PainFace wouldn鈥檛 work with other laboratory animals. The team is working on adding white mice images to the model.
PainFace鈥檚 main limitation is mice must be videoed in specially designed cages to capture their grimaces in the clearest detail, says Zylka.
鈥淭his new software can improve our way of analysing grimace score [in mice],鈥 says at the University of Reading, in the UK. 鈥淢oreover, the use of PainFace can reduce human error behind manual scoring and produce data which can be easily compared across different pain labs鈥.
Reference: