DeepLabStream enables closed-loop behavioral experiments using deep learning-based markerless, real-time posture...

Area of Science:

• Neuroscience
• Animal Behavior
• Computational Biology

Background:

• Animal behavior is characterized by neuronal-driven sequences of postures over time.
• Current pose estimation technologies primarily focus on offline analysis with high spatiotemporal resolution.
• Real-time behavioral detection and reaction are crucial for correlating behavior with neural activity.

Purpose of the Study:

• To introduce DeepLabStream, a versatile closed-loop tool for real-time pose estimation and posture-dependent stimulation.
• To enable the correlation of specific behaviors with underlying neuronal activity in experimental settings.

Main Methods:

• Development of DeepLabStream, a system offering millisecond-range temporal resolution for pose estimation.
• Implementation of a closed-loop system allowing for real-time, posture-dependent stimulations.
• Utilizing DeepLabStream to semi-autonomously conduct a second-order olfactory conditioning task in freely moving mice.

Main Results:

• DeepLabStream provides real-time pose estimation with high temporal resolution.
• The system successfully facilitated posture-dependent stimulations.
• Demonstrated application in a complex behavioral task (olfactory conditioning) with optogenetic labeling of neuronal ensembles.

Conclusions:

• DeepLabStream is a powerful tool for closed-loop neuroscience research, enabling real-time behavioral analysis and intervention.
• The system's versatility supports various experimental designs and input/output devices.
• Facilitates the study of neural mechanisms underlying specific behaviors by linking real-time pose estimation with neural activity.