Automated analysis of mouse rearing using deep learning.

Area of Science:

• Neuroscience and behavioral science
• Artificial intelligence in biological research

Background:

• Rodent rearing behavior is a key indicator for assessing anxiety and exploratory tendencies in animal models.
• Accurate quantification of rearing is crucial for behavioral neuroscience research.

Purpose of the Study:

• To develop and validate a convolutional recurrent neural network (CRNN) model for automated detection of mouse rearing behavior.
• To assess the model's performance against human observation and its ability to detect behavioral changes.

Main Methods:

• Overhead videos of C57BL/6 mice were collected under varying light conditions.
• Manual frame-by-frame labeling of rearing events was performed for supervised learning.
• A CRNN model was trained using the labeled behavioral data.
• Model performance was evaluated using sensitivity metrics on independent test datasets.

Main Results:

• The CRNN model achieved a sensitivity of 89.2%, demonstrating performance comparable to human observers.
• The model successfully detected increased rearing behavior after caffeine administration.
• The model differentiated between diurnal and nocturnal activity patterns in mice.

Conclusions:

• The developed CRNN model offers a reliable and automated method for quantifying mouse rearing behavior.
• This AI-driven approach can enhance the efficiency and objectivity of behavioral assessments in preclinical research.
• The model's ability to detect pharmacological effects and circadian variations highlights its utility in neuroscience studies.