Research On Neutrophil Recognition And Tracking Algorithm In Zebrafish Based On Instantaneous Orthogonal Imaging System

The identification and tracking of biological cells is an important way to explore the laws of biological individual movement.Zebrafish cells have become one of the important research carriers because of good optical transparency and being easily bred.Before tracking the biological cells,it is necessary to select the appropriate imaging method.The angular multiplexed optical projection tomography(OPT)system has the advantages of small phototoxicity and wide imaging field of view compared with the traditional optical imaging system,which has preferentially deployed for observing the motion law of living organisms.Then,due to wide field imaging,cells seriously aggregate in the image,and thus it is necessary to find a suitable segmentation algorithm to identify cells.In the cell tracking stage,the angular multiplexed optical projection tomography system synthesizes the three-dimensional state of the cell from the projection data in two directions.The traditional two-dimensional target tracking algorithm and the three-dimensional tracking algorithm are not well suitable for the OPT system,and therefore,there is an urgent need to develop the cell tracking algorithm for the angular multiplexed OPT system.In this paper,the traditional watershed algorithm and improved algorithm are used to segment the zebraish cell image.The watershed algorithm based on distance transformation can not demonstrate the minimum value of cell region in the cell aggregation area easily causing under-segmentation.And the watershed algorithm of gradient transformation is greatly sensitive to image gradient edges,which will cause excessive segmentation.Although the watershed algorithm based on marker control can describe the cell aggregation,threshold segmentation algorithm as the basis of the segmentation is needed.Therefore,in terms of the comparison of the characteristics of watershed algorithm and improved algorithm,this paper proposes a zebrafish neutrophil segmentation algorithm based on Mask-RCNN,and changes the Mask-RCNN detection network for multi-category detection to a single-category detection network for zebrafish neutrophils.The single-category detection network constructs the zebrafish neutrophil data set through adjusting the model hyperparameters,and iteratively trains the model,so that a higher detection accuracy at 89.29%of the final model was obtained in the YZ direction series picture and 86.20%in the XZ direction series picture,which outperforms the traditional watershed algorithm.In the aspect of cell tracking,based on the two-dimensional tracking algorithm DeepSort algorithm based on the two-dimensional tracking algorithm DeepSort algorithm combined with the target motion characteristics of the OPT imaging system,a three-dimensional tracking algorithm based on DeepSort for zebrafish neutrophils was proposed.Aiming at the trajectory id switching problem in the algorithm,combined with the motion characteristics of zebrarish cells in the OPT imaging system,an optimization algorithm combining the cumulative distance error is proposed.When the zebrafish neutrophils are linked frame by frame,the OPT rotation information pair is combined.The cell is predicted by the position information of the next frame.The prediction method conforms to the actual movement rule of the cell,and effectively ensures the spatial and temporal continuity of the reconstructed zebrafish neutrophil trajectory,which is helpful for obtaining a complete and reliable zebra.Three-dimensional trajectories of fish neutrophils in the body.Finally,seventy-nine quantifiable zebrafish embryo neutrophil trajectories were reconstructed.