Design And Implementation Of Behavior Analysis System For Open Field Tests In Mouse

Animal behavior is the body language of animals to express their psychology and physiology,as well as the embodiment of their own comprehensive function.Mouse are typical model animals,their different behaviors in the open field are widely used to measure the experimental results in biology,neuroscience,pharmacology,genetics and other fields.The existing methods for mouse behavior analysis are time consuming and poor in generalization ability.The design and implementation of a mouse open field test behavior analysis system is of great significance to reduce the workload of related researchers.In this paper,the recognition of behavioral patterns in open field test of mouse is studied,designed and implemented a mouse behavior analysis system.The mouse key point detection model based on convolutional neural network and the mouse behavior recognition model based on ConvLSTM network are designed,to calculate behavioral indicators and apply them to the acute stress experiment.The main work and results of this paper are as follows:(1)The framework of mouse behavior analysis system was designed.Analyzed the specific requirements of behavior analysis in the mouse open field test,selected the model from the experiment box,camera and PC,and designed the basic framework of the mouse behavior analysis software system.The functions of graphical user interface,image acquisition,data processing and result display are realized by using toolkits such as Py QT5,Tensor Flow,Keras,Open CV,Num Py and Matplotlib,and the algorithm flow of the system is determined.(2)A mouse key point detection model based on convolutional neural network was designed to realize key point detection of five typical behavior patterns under multiple influencing factors.The deep fusion of dilated convolution and transposed convolution were used to design the mouse key point detection model,and the coordinates of key points were accurately predicted by combining the heatmap and coordinate offsets.The key point detection data set of mouse was established,and the experimental results under different dilation rates were compared to determine the best model parameters.The average PCK of the present model for detecting key points of nose,left ear,right ear and tail root reached 91%,higher than the performance of CPM,Stacked Hourglass and Deeplabcut models.The trajectory of key points was further processed by Kalman filter,and the motion parameters such as velocity,angular velocity,distance and trajectory were calculated.(3)A mouse behavior recognition model based on ConvLSTM network was designed to recognize the five basic behaviors of mouse: go straight,motionless,groom,rear and body rotation.A mouse behavior recognition model combining key point detection,ConvLSTM network and full convolutional network was designed.The mouse behavior recognition data set was established,the behavior recognition results under different time windows were compared,and the best time window parameters were determined.The average recognition accuracy of the proposed model for the five basic behaviors of go straight,motionless,groom,rear and body rotation reaches 94.27%,which is higher than the performance of LSTM,BI-LSTM and 3DCNN models.The error classification results were corrected by mode filtering,and the behavior parameters such as the number of behavior occurrence and the duration of behavior were measured.(4)The mouse behavior analysis system was implemented and applied to the acute stress experiment.The software and hardware parameters of the mouse behavior analysis system were selected to realize the behavior analysis function of the system.A mouse acute stress model was constructed by using the footshock method,and the behavioral differences of mouse before and after acute stress were measured based on our system.The experimental results showed that after acute stress,the autonomous activities of the mice were weakened,and they tended to be more peripheral activities,and the degree of anxiety and tension increased.The experimental results were consistent with the results of existing literature.Compared with traditional behavior analysis methods,our system provides comprehensive and detailed behavioral indicators.