Development and validation of PainFace, a software platform that simplifies and standardizes mouse grimace analyses

Abstract

Facial grimaces are now commonly used to quantify spontaneous pain in mice and other mammalian species, but scoring remains subjective and relies on humans with very different levels of proficiency. Here, we developed a Mouse Grimace Scale (MGS) for black-coated (C57BL/6) mice consisting of four facial action units (orbitals, nose, ears, whiskers). We used this scale to generate ground truth data from over 70,000 images of black mice in different settings. With this large data set, we developed a deep neural network and cloud-based software platform called PainFace (http://painface.net) that accurately scores facial grimaces of black mice on a 0-8 scale. PainFace generates over two orders of magnitude more MGS data than humans can realistically achieve, and at superhuman speed. By analyzing the frequency distribution of grimace scores, we found that mice spent >7x more time in a high grimace state following laparotomy surgery relative to sham surgery controls. The analgesic carprofen reduced the amount of time animals spent in this high grimace state after surgery. Specific facial action unit score combinations were overrepresented following laparotomy surgery, suggesting that characteristic facial expressions are associated with a high grimace state. We performed validation experiments in two labs located in different countries to demonstrate reproducibility of the PainFace platform. To further enhance rigor and reproducibility, we will invite pain researchers to beta test PainFace and then incorporate their feedback into the software and manuscript prior to peer review. While this study is focused on mice, PainFace was designed to simplify, standardize, and scale up grimace analyses with many other mammalian species, including humans.