Markerless Mouse Tracking for Social Experiments.

Area of Science:

• Ethology
• Biomedical Engineering
• Computer Vision

Background:

• Accurate automated behavior quantification is crucial for studying animal social interactions.
• Existing tracking methods often struggle with identity switches and loss of anatomical data, especially with visually similar subjects.
• Increased human effort in training or post-processing is typically required to overcome these limitations.

Purpose of the Study:

• To develop a markerless, video-based tool for simultaneous, accurate tracking of two interacting mice with identical appearances.
• To quantify social behaviors like sniffing, touching, and locomotion without additional human intervention.
• To enhance tracking accuracy in controlled experimental settings.

Main Methods:

• A hybrid approach combining traditional computer vision techniques with deep learning.
• Training the tool on a limited dataset of manually annotated images.
• Outputting body masks and key anatomical coordinates (snout, tail-base) for each mouse.
• Testing under various conditions, including bedding and surgical implants.

Main Results:

• Near elimination of identity switches compared to existing methods.
• Approximately 15% improvement in tracking accuracy over pure deep-learning pose estimation.
• Successful application in quantifying social behaviors in mice with altered olfaction.
• The tool functions effectively even with common experimental setups like bedding and implants.

Conclusions:

• The proposed markerless tracking tool significantly improves the accuracy and reliability of automated behavioral quantification in socially interacting mice.
• This method reduces the need for extensive human post-processing, making it a valuable tool for rodent social behavior research.
• The approach offers potential for broader applications in studying group behaviors in various animal models.