Automated behavioral segmentation and markerless pose tracking of mice during spaceflight

Area of Science:

• Space biology
• Artificial intelligence
• Animal behavior

Background:

• The International Space Station's Rodent Habitat facilitates spaceflight behavioral studies.
• Manual video annotation for rodent behavior is time-consuming and labor-intensive.

Purpose of the Study:

• To evaluate deep learning tools for automated behavioral analysis of rodents aboard the ISS.
• To establish benchmarks for automated monitoring systems in orbital vivaria.

Main Methods:

• Applied pose estimation (SLEAP) and behavioral segmentation (DeepEthogram) to archival footage from the Rodent Research-1 mission.
• Compared deep learning model performance against extensive manual annotations by multiple labelers and behaviorists.

Main Results:

• Pose tracking accuracy closely matched human inter-annotator variability, even with challenging imaging conditions like lens soiling and occlusions.
• Behavioral classification achieved high accuracy (0.86-0.90) across eight categories.
• Kinematic analysis revealed centripetal accelerations during circling behavior approaching 1 g.

Conclusions:

• Deep learning effectively automates rodent pose estimation and behavioral segmentation in the spaceflight environment.
• Findings suggest progressive behavioral adaptations in mice exposed to microgravity.
• This study pioneers the use of AI for rodent behavior analysis in space, setting a precedent for future research.