Establishment of a deep learning-based open field refined behavioral analysis system and its evaluation and application in two murine models

AimTo develop a fine-grained open-field behavioral analysis system based on the deep learning algorithms DeepLabCut(DLC)and DeepEthogram(DEG), so to enhance the detection sensitivity of emotion-related behaviors in mice of acute restraint stress (acute restraint stress, ARS)and adolescent intermittent ethanol exposure(adolescent intermittent ethanol exposure, AIE).MethodsUnder the single-camera open-field test (open field test, OFT) setup, a system was established by integrating the DLC and DEG algorithms to analyze fine-grained horizontal and vertical behaviors in mice of ARS or AIE. The influence of vertical behaviors on horizontal metrics was further assessed.ResultsCompared to the control group, the ARS group exhibited significant reductions in traditional travel speed in central area. Moreover, fine-grained metrics obtained by this system was significant decreasing including horizontal travel speed in central area, supported rearing duration, number of supported rearing and supported rearing latency in the ARS group. Notably, traditional travel duration was significantly greater than horizontal travel duration. Additionally, significant correlations existed between supported rearing duration and traditional travel duration, as well as between supported rearing duration and traditional resting duration. Compared with controls, the AIE group showed a significant increase in traditional travel and resting duration indices. Moreover, metrics of fine-grained horizontal travel and resting, apart from horizontal travel duration, significantly increased. Specifically, a significant increase occurred in supported rearing duration, number of supported rearing, supported rearing distribution index, supported rearing distribution density and unsupported rearing distribution index. Traditional travel and resting metrics significantly exceeded corresponding horizontal travel and resting metrics.ConclusionsThe deep learning-based behavioral analysis system developed in this study addresses limitations of traditional OFT by enabling automated quantification of vertical rearing behaviors. By elucidating the influence of these two vertical behaviors on horizontal metrics, to propose corresponding solutions, this system provides a more sensitive and comprehensive analytical tool for neuropsychiatric disorder research and drug evaluation.