Rodent Social Behavior Recognition Using a Global Context-Aware Vision Transformer Network

Area of Science:

• Neuroscience and computational biology
• Utilizes advanced computer vision for animal behavior analysis.

Background:

• Manual animal behavior coding is time-consuming and error-prone.
• Machine learning offers automated solutions for analyzing animal behavior.
• Existing methods require improvement for accurate rodent social interaction identification.

Purpose of the Study:

• To develop and evaluate an automated system for recognizing rat social behaviors.
• To leverage state-of-the-art computer vision, specifically the Global Context Vision Transformer (GC-ViT), for this task.

Main Methods:

• Proposed a novel approach: Vision Transformer for Rat Social Interactions (ViT-RSI).
• Adapted the GC-ViT architecture for identifying social interactions in rodents.
• Utilized the publicly available Rat Social Interaction (RatSI) dataset for experiments.

Main Results:

• ViT-RSI achieved high accuracy in identifying five distinct rat social behaviors.
• The model outperformed prior literature results for four behaviors: Approaching (F1=0.81), Following (F1=0.81), Moving away (F1=0.86), and Solitary (F1=0.94).

Conclusions:

• The ViT-RSI approach demonstrates significant potential for accurate and automated animal behavior analysis.
• This method offers a more efficient and reliable alternative to manual coding in research settings.
• ViT-RSI advances the field of machine learning applications in behavioral neuroscience.