Automated analysis of mouse rearing using deep learning

Rodent rearing behavior is frequently assessed as an indicator of anxiety and exploratory tendencies. This study developed a convolutional recurrent neural network (CRNN) model to detect mouse rearing using overhead videos. Behavioral data from C57BL/6 mice under light and dark conditions were manually labeled frame-by-frame and used to train the CRNN model. Model performance was evaluated on separate test videos, achieving a sensitivity of 89.2 %, comparable to human observation. The model reliably detected increased rearing following caffeine administration and distinguished differences between day and night activity patterns.