Machine Learning Guided Video Analysis Identifies Sound-Evoked Pain Behaviors from Facial Grimace and Body Cues in Mice

Humans can experience auditory pain in response to sound, either from extremely loud noise or in cases of pain hyperacusis, where typically tolerable sounds become painful. However, the mechanisms underlying auditory pain remain poorly understood. Developing behavioral methods to measure sound-evoked pain in animal models is critical for elucidating these mechanisms. Here, a deep learning-based approach was developed to measure auditory pain in freely moving mice by analyzing facial grimace and body position from video recordings during sound exposure. Facial grimace, a validated marker of spontaneous, ongoing pain in mice, was quantified using a deep neural network trained to extract established facial features. Postural changes, additional indicators of pain, were analyzed from the same camera angle. To validate the model, a known painful state, migraine induced by injection of the neuropeptide calcitonin gene-related peptide (CGRP) was used. With this approach the ability to quantify a pain response distinct from baseline behavior was demonstrated, resulting in a defined pain threshold. Sound exposure at high intensities elicited significant changes in facial grimace and body posture, in comparison, surpassing the pain threshold established during migraine validation. These behavioral changes were absent in TMIE-knockout mice, which lack functional cochlear transduction. This automated, high-throughput framework enables objective and sensitive analysis of sound-evoked pain and provides a foundation for future studies investigating the peripheral and central mechanisms of auditory pain.