Wenjin Dong
Research Mentor(s): Bing Ye
Mentor Department: Life Sciences Institute, University of Michigan
Authors: Wenjin Dong, Bing Ye
Session: Session 1 (9:00am – 9:50am)
Presentation Type: Poster 95
Abstract
AI has become a significant tool for animal behavioral analysis since it is time-saving and accurate compared with manual analysis. Our lab harvests AI’s power in behavioral analysis by creating LabGym, a behavioral analysis software tool based on deep learning. We are exploring facial expression analysis to determine pain in mice. We hypothesize that LabGym, a tool in our lab for animals behavior analysis, could be also used for analyzing facial expressions. When mice experience pain, they usually droop their ears and constrict their eyes. When mice experience pain, they usually droop their ears and constrict their eyes. These subtle facial expression changes are likely to be recognized by LabGym after training several models for detectors and categorizers. As a result, LabGym can generate an annotated video to show whether a mouse experiences pain. Moreover, the mouse’s facial expressions are possible to be quantified by LabGym. Through statistical analyses such as UMAP, it is possible to identify which facial regions are strongly associated with the pain recognized by LabGym. This allows us to observe the areas LabGym focuses on the most when identifying the facial expressions of mice. Finally, Grad-CAM can be applied to reveal the areas that LabGym emphasizes when recognizing mouse facial expressions and compare them with the areas that humans focus on. This comparison helps us understand the logic behind AI’s analysis of mouse facial recognition and provides insights for improving the AI algorithm. The research will provide an approach for scientists to analyze pain in mice, which is important for studying the mechanisms underlying pain and for developing treatment for alleviating pain.
