DeepEthoProfile-Rapid Behavior Recognition in Long-Term Recorded Home-Cage Mice

Area of Science:

• Neuroscience
• Computational Biology
• Animal Behavior Analysis

Background:

• Understanding animal behavior is vital for neuroscience research, particularly in studying brain function and dysfunction.
• Accurate and efficient behavioral analysis in laboratory settings is a significant challenge.

Purpose of the Study:

• To develop an open-source software, DeepEthoProfile, for efficient and accurate classification of mouse behavior.
• To introduce EthoProfiler, a novel mobile system for simultaneous video recording of multiple mice.

Main Methods:

• Developed DeepEthoProfile, a deep convolutional neural network-based software requiring no spatial cues.
• Introduced EthoProfiler, a mobile cage rack system for recording up to 10 mice.
• Utilized 36 hours of manually annotated video data for training and validation.

Main Results:

• DeepEthoProfile achieved over 83% classification accuracy, comparable to human-level performance.
• The software demonstrated performance on par with state-of-the-art solutions on existing datasets.
• DeepEthoProfile processes data efficiently, annotating nearly 2,000 frames per second.

Conclusions:

• DeepEthoProfile offers a reliable and efficient solution for automated mouse behavior analysis in home environments.
• The software is robust to variations in lighting and cage conditions, suitable for real laboratory settings.
• DeepEthoProfile and the accompanying dataset can advance research in animal behavior and neuroscience.