Artificial intelligence–based automated scoring of the mouse grimace scale: a systematic review
Source: PubMed Central Open Access, NCBI / U.S. National Library of Medicine
Supplemental Digital Content is Available in the Text. Deep learning models automate Mouse Grimace Scale scoring with high accuracy, yet critical gaps in external validation, strain diversity, and reporting limit generalizability. toc Abstract The Mouse Grimace Scale (MGS) is a widely adopted facial expression–based tool for pain assessment. However, manual scoring is labor-intensive and subject to both intra- and inter-observer variability. Artificial intelligence–based approaches have recently been developed to automate MGS analysis, yet their performance and methodological quality have not been systematically evaluated. This systematic review was undertaken to evaluate studies applying artificial intelligence–based methods for automated MGS scoring. We searched PubMed, Web of Science, Scopus, and IEEE Xplore up to January 14, 2026. Descriptive statistics were used to summarize quantitative outcomes. Eight studies published between 2018 and 2025 met the inclusion criteria. All used deep learning–based approaches to perform grimace score prediction or facial action unit analysis. Training data set sizes varied across the studies, ranging from less than 100 images to more than 70,000 images. Reported performance was generally high, with classification accuracies up to 0.97 and strong correlations between automated and human scores (up to 0.87). However, most evidence is based on internal validation only. Artificial intelligence–based approaches to MGS scoring have the potenti
Abstract
Supplemental Digital Content is Available in the Text. Deep learning models automate Mouse Grimace Scale scoring with high accuracy, yet critical gaps in external validation, strain diversity, and reporting limit generalizability. toc Abstract The Mouse Grimace Scale (MGS) is a widely adopted facial expression–based tool for pain assessment. However, manual scoring is labor-intensive and subject to both intra- and inter-observer variability. Artificial intelligence–based approaches have recently been developed to automate MGS analysis, yet their performance and methodological quality have not been systematically evaluated. This systematic review was undertaken to evaluate studies applying artificial intelligence–based methods for automated MGS scoring. We searched PubMed, Web of Science, Scopus, and IEEE Xplore up to January 14, 2026. Descriptive statistics were used to summarize quantitative outcomes. Eight studies published between 2018 and 2025 met the inclusion criteria. All used deep learning–based approaches to perform grimace score prediction or facial action unit analysis. Training data set sizes varied across the studies, ranging from less than 100 images to more than 70,000 images. Reported performance was generally high, with classification accuracies up to 0.97 and strong correlations between automated and human scores (up to 0.87). However, most evidence is based on internal validation only. Artificial intelligence–based approaches to MGS scoring have the potential to make preclinical pain assessment more efficient and consistent. However, current evidence is limited by a lack of external validation, partial implementation of the full MGS, and variability in reporting performance metrics. Addressing these limitations will be critical to support the robust and reliable integration of automated grimace analysis into future pain research.