This video demonstrates the application of DeepLabCut™, a robust, open-source tool for markerless pose estimation, in tracking mouse locomotion in an open field test. The video presents an accelerated top-down view of a box divided into nine sub-boxes, each containing a freely exploring mouse. 

DeepLabCut is based on transfer learning with deep neural networks, achieving high accuracy with minimal training data (50-200 frames). This versatile framework allows precise tracking of body parts, including the nose and tail base, across species and behaviors, supporting both 2D and 3D pose estimation, including multi-animal tracking. 

Experimental Highlights: 

Behavioral Paradigm: 

– Mice were observed across multiple days to evaluate changes in exploratory behavior. 

– Normal mice exhibit reduced exploration over time, indicating memory of the environment. 

– Mice with long-term memory impairments show consistent exploratory patterns across sessions. 

DeepLabCut Implementation: 

– Body Part Tracking: The network accurately detects the nose and tail base of each mouse, enabling detailed locomotion analysis. 

– Network Training: Trained on 50 labeled frames over ~300,000 iterations. 

– Accuracy: Achieved a train error of 1.13px and a test error of 4.82px, comparable to commercial solutions like Ethovision. 

– Visualization: Detection trails of the tail base are retained for 50 frames to visualize movement dynamics. 

Detection Limitations: 

– Some mice in hidden corners of the boxes may not be detected due to limited camera visibility and display settings. 

DeepLabCut Advantages: 

– Open source and fast, developed collaboratively by the Mathis Group and Mathis Lab at EPFL. 

– Capable of tracking multiple animals and computing 3D pose estimates. 

This study underscores the power of DeepLabCut for behavioral analysis, providing high-precision data to investigate exploratory behavior, locomotion, and memory in animal models.