YOLOv11-Based quantification and temporal analysis of repetitive behaviors in deer mice

Abstract

Detailed temporal dynamics of deer mouse (Peromyscus maniculatus bairdii) behavior remain poorly characterized. This study presents an integrated automated system combining YOLOv11 deep learning for direct behavior classification, post-processing for bout reconstruction, and a comprehensive temporal analysis suite tailored for deer mice. YOLOv11 performs frame-by-frame classification of key whole-body behaviors (e.g., Exploration, Grooming, Rearing types) using bounding boxes, bypassing initial kinematic feature engineering. This methodology facilitates objective, high-throughput quantification of behavior frequency, duration, and complex temporal organization, including transition patterns and sequential structure revealed through analyses like transition probabilities, behavior sequence mining, and lead-follower behavior relationships. In addition to describing a new methodology, this study provides initial baseline temporal data for deer mice, a powerful suite of analyses for future Peromyscus studies evaluating natural variation and experimental manipulations, or for use in other movement disorder models. In conclusion, the described YOLOv11-based system provides an efficient, reliable, and accessible methodology for both detailing behavioral activity and enhancing investigations of temporal dynamics in deer mice and other animal models.