Design Of A System For Measuring The Startle Response Of Zebrafish Under Light Stimulation

Gene and drug screening for assessing visual function plays an essential role in thoroughly understanding the mechanism of visual abnormalities,as well as in developing therapeutic methods for visual pathologies.Such screening usually requires to be executed in experimental animal models.The zebrafish model is one of the most commonly employed animal models in biomedical research,especially in screening-targeted behavioral studies.This is because zebrafish possesses multiple merits,including high reproductivity,short lifespan,and satisfactory similarity with humans in vision-related genes and organs.By analyzing the startle response under light stimulation,the visual function of the zebrafish can be efficiently evaluated and,subsequently,the high-throughput screening can be completed.Conventional algorithms for processing the video data collected from the startleresponse experiments of zebrafish suffer from several limitations,such as the low tolerance to either the noise or the postural variation of the animals.To avoid the negative effects induced by these limitations,the convolution neural network-based deep-learning(DL)algorithm has been applied to the behavioral data processing for the first time in this study.In order to enable high-throughput behavioral test of zebrafish,a system has been designed and developed.This system is capable of automatically video recording and measuring the startle response under light stimulation.The hardware of this system includes a high-speed video camera,a disc container(diameter 16 cm)with its capacity of ～20 zebrafish larvae,a UV filter reducing the interference from the white light,and an opaque cover shielding the external light.In addition,a relay has also been employed to activate multiple cameras and,hence,elevate the efficiency by performing the 5 to 10 measurements in parallel.Regarding the software,this study has employed the DL algorithm for automatic tracking and matching.Using the Hungarian matching algorithm and Kalman filter,each individual zebrafish has been successfully identified and traced.The position and motion of zebrafish showing startle response resulted from the applied light stimulation have been specified and profiled,following the analysis of the acquired video recordings with MATLAB.In addition,the outcomes obtained from the matching and tracking algorithms based on different deep-learning models,the YOLO v3 model and Mask R-CNN model,have been compared in this study.