Real-Time Closed-Loop Feedback in Behavioral Time Scales Using DeepLabCut.

Area of Science:

• Neuroscience
• Animal Behavior
• Computer Vision

Background:

• Computer vision and deep learning are increasingly used for offline animal behavior tracking and pose estimation.
• Markerless deep learning approaches are gaining traction for behavioral analysis.
• Real-time tracking is crucial for advanced closed-loop experimental systems.

Purpose of the Study:

• To develop and validate a real-time movement estimation approach using DeepLabCut for behavioral tracking.
• To enable precise, markerless tracking of individual mouse whiskers.
• To create a system capable of triggering outputs at behavioral timescales.

Main Methods:

• Trained a deep neural network (DNN) offline using high-speed video data of a mouse whisking.
• Transferred the trained DNN for real-time whisker tip tracking.
• Converted whisker positions into TTL output signals within 10.5 ms timescales.

Main Results:

• Successfully implemented real-time estimation of mouse whisker movement using DeepLabCut.
• Achieved tracking of three whisker tips in an arc with high temporal resolution.
• Demonstrated the ability to trigger outputs based on individual whisker movement or inter-whisker distance.

Conclusions:

• This DeepLabCut-based approach enables real-time, markerless tracking of animal movement.
• The developed system facilitates flexible closed-loop experiments, complementing optogenetics.
• It allows direct manipulation of the relationship between movement and neural activity.