PainFace: A Machine Learning Tool to Measure Spontaneous Pain Behaviors in Rats (Rattus norvegicus)

Background & Aims

Pain assessment in rats is challenging due to the subjective nature, low-throughput, and high resources demands of available pain assessment tools. These tools often rely on naturalistic changes in facial expression associated with spontaneous or induced pain. The rat grimace scale is a validated pain scoring system for adult rats that characterizes four facial features, or ‘action units’, associated with spontaneous pain[1]. Despite its usefulness, limitations have reduced its adaptation in biomedical research. We aimed to address these concerns by expanding the functionality of a validated machine learning algorithm used to assess pain severity called PainFace, originally used with mice[2], for use in rats. We trained rat PainFace to automate the detection of the pain. We hypothesized that rat PainFace would strongly correlate with manual pain scores while allowing automated pain severity scoring.

Methods

Sprague Dawley rats (n= 33 males, 40 females) were randomly assigned by sex to 1 of 6 treatment groups: extended-release buprenorphine (Ethiqa XR; 0.65 mg/kg), buprenorphine (0.01 mg/kg), meloxicam (2 mg/kg), meloxicam with local bupivacaine (2 mg/kg and <2 mg/kg, respectively), saline (5 mL/kg), and unoperated control. After pre-emptive subcutaneous drug administration, we performed a laparotomy (a reliable pain model) by making a 3-cm midline incision penetrating the peritoneal cavity under isoflurane anesthesia. Spontaneous pain behaviors were evaluated using a subjective pain score, and video recordings were collected at baseline, 15-minutes, 24-, 48-, 72-, and 168-hours post-laparotomy. Videos were manually annotated for facial features and scored by a masked observer using established criteria to discriminate pain severity (ground truth).

Results

Preliminary results demonstrate that rat PainFace did not detect facial grimacing in any treatment groups after the 15-minute post-laparotomy time point. However, at 15-minutes post-laparotomy rats in the Ethiqa XR, meloxicam, and saline groups displayed spontaneous pain behaviors detected by PainFace (p=0.021), with significantly higher PainFace scores in Ethiqa XR and saline groups compared to unoperated control rats (p<0.001). Twenty-four hours post- laparotomy, Ethiqa XR rats had lower PainFace scores across treatment groups (p<0.01). These data suggest that facial grimacing is an acute response to pain in rats and preemptive Ethiqa XR may not mitigate post-surgical pain within the first hour of administration.

Conclusions

This study introduces the first automated pain assessment tool to measure pain severity in adult rats undergoing post-surgical pain. Future studies should assess additional features to provide a multimodal temporal approach to optimize pain detection in this population.

References

[1] McCoy ES, Park SK, Patel RP, Ryan DF, Mullen ZJ, Nesbitt JJ, Lopez JE, Taylor-Blake B, Vanden KA, Krantz JL, Hu W, Garris RL, Snyder MG, Lima LV, Sotocinal SG, Austin JS, Kashlan AD, Shah S, Trocinski AK, Pudipeddi SS, Major RM, Bazick HO, Klein MR, Mogil JS, Wu G, Zylka MJ. Development of PainFace software to simplify, standardize, and scale up mouse grimace analyses. Pain 2024.
[2] Sotocinal SG, Sorge RE, Zaloum A, Tuttle AH, Martin LJ, Wieskopf JS, Mapplebeck JC, Wei P, Zhan S, Zhang S, McDougall JJ, King OD, Mogil JS. The Rat Grimace Scale: a partially automated method for quantifying pain in the laboratory rat via facial expressions. Mol Pain 2011;7:55.

Presenting Author

Morika Williams

Poster Authors

Morika Williams

DVM, PhD, DACLAM

UNC Chapel Hill

Lead Author

E-mail