Automated Behavior Analysis in the Novel Object Recognition Test.

Area of Science:

• Behavioral Neuroscience
• Machine Learning
• Animal Behavior Analysis

Background:

• Manual annotation of animal behaviors is time-consuming, subjective, and costly.
• Automated behavior classification is crucial for advancing behavioral neuroscience research.
• The Novel Object Recognition Test (NORT) is a key paradigm for assessing memory and cognition in rodents.

Purpose of the Study:

• To develop a unified deep learning and heuristic framework for automated behavior classification in NORT.
• To improve the accuracy and scalability of rodent behavior annotation.
• To provide an objective and reproducible method for analyzing NORT videos.

Main Methods:

• Fine-tuned YOLOv11 Pose model to detect rat keypoints (nose, tail-base).
• Developed spatial heuristics from keypoints for behavior identification.
• Trained YOLOv11 classification models on various NORT video datasets (2-object, 5-object, combined).
• Integrated heuristic post-processing for object interaction identification.

Main Results:

• The combined dataset model with heuristic post-processing achieved the highest accuracy and generalization.
• The framework outperformed heuristic-only and subset-specific models.
• Error analysis indicated most errors occurred at behavioral transition points.

Conclusions:

• The proposed framework offers a scalable, reproducible, and objective solution for NORT video annotation.
• This work lays the foundation for temporally-aware behavioral analysis in rodents.
• Publicly available code and models will support future research in rodent behavior recognition.