DeepBehavior: A Deep Learning Toolbox for Automated Analysis of Animal and Human Behavior Imaging Data

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Understanding the brain-behavior relationship requires detailed behavioral analysis.
• Current methods for analyzing complex behaviors can be labor-intensive and time-consuming.

Purpose of the Study:

• To present deep learning-based methods for analyzing behavior imaging data in mice and humans.
• To provide detailed instructions and code for rapid implementation by the neuroscience community.
• To expand the available tools for detailed behavior analysis using deep learning.

Main Methods:

• Utilized three different convolutional neural network architectures.
• Applied methods to five different behavior tasks in mice and humans, including 3D kinematic analysis and social interaction tests.
• Employed a transfer learning approach to accelerate network training.
• Provided code for post-processing of deep learning analysis results.

Main Results:

• Demonstrated the effectiveness of deep learning for detailed behavior analysis across various tasks and species.
• Showcased the acceleration of network training using transfer learning.
• Developed comprehensive methods including implementation instructions, diverse applications, and post-processing code.

Conclusions:

• The presented deep learning methods offer a powerful and efficient approach to behavior analysis in neuroscience research.
• These methods can be readily implemented by researchers, accelerating the study of brain-behavior relationships.
• The human motor behavior analysis tools have potential clinical applications for assessing motor function recovery after injury.