Assessment of Laboratory Mouse Activity in Video Recordings Using Deep Learning Methods

Area of Science:

• Animal Behavior Analysis
• Machine Learning in Biology
• Laboratory Animal Science

Background:

• Assessing laboratory animal wellbeing is crucial for ethical research.
• Objective and automated methods are needed to analyze animal behavior.
• Current behavioral analysis can be labor-intensive and subjective.

Purpose of the Study:

• To develop and validate a deep learning-based method for automated classification of mouse behaviors.
• To assess the performance of different deep learning models in behavioral analysis.
• To enable objective and efficient monitoring of laboratory animal activity.

Main Methods:

• Filming laboratory mice in observation cages to capture short video clips.
• Labeling video clips into five predefined behavioral categories.
• Applying three distinct convolutional neural network (CNN) methods for video classification.
• Utilizing a two-stream network analyzing frames and optical flow for optimal performance.

Main Results:

• The best performing two-stream CNN model achieved 86.4% accuracy in classifying mouse behaviors.
• The method successfully identified key behaviors, including self-grooming.
• Algorithmic analysis of videos enabled automated assessment of animal behavior.

Conclusions:

• The presented protocol offers a reliable method for automated behavioral assessment in laboratory mice.
• Deep learning significantly enhances the objectivity and efficiency of animal behavior analysis.
• This approach supports improved animal welfare monitoring and research integrity.