Research On Algorithm Of Cell Detection And Tracking In Microscope Videos

Cell tracking under microscope is an important branch and research area of multi-target detection and tracking.It is also a hot topic which combines computer vision and life science engineering.Accurate tracking of cell population plays an important role in researching directional movement of blood cells,the process of organ repair and the variation of cell cycle influenced by drugs.Different from the multi-target tracking in the natural scene,cells have a wide range of types.Also,the cell phenotype is highly similar to each other and their movements are complex,which make it more difficult to be tracked.Recently,researches on cell tracking have been focused on how to improve the accuracy and quantitatively analyze the changes in cell activity.In this paper,we first analyzed the imaging tools and microscopy image features.After comparing the existing cell detection and tracking algorithms,we proposed a robust and portable of hierarchical tracking framework and tested on cells image sequences with different morphology and imaging tools.The specific works are listed as following:1.We analyzed the cell morphology and microscope image features,and then explored the corresponding image pre-processing algorithms and parameters settings to solve the problem of noise and halo around the cell caused by imaging tools,which can improve the image quality.2.We introduced the classical segmentation algorithms and grouped existing tracking frameworks,including the detection-based association framework,model-based evolution framework and probability-based Bayes framework.The hierarchical idea was applied to optimize segmentation and tracking results by combining the global and local information.3 In the detection module,according to the characters of cell morphology and adhesion,we proposed the method of ellipse fitting to recognize the under-segmentation region and segment the image globally and locally.The image was firstly segmented with unified threshold and then processed with morphological operation and region mark.Each individual region was then fitted with an ellipse,so the under-segmented cell can be selected with ellipse parameters.Lastly,the wavelet method was used to process the rectangle region,and the centroids were extracted to represent cells.4 In the tracking module,we proposed a hierarchical tracking framework which is suitable for cell tracking.The method deals with the cell events in different tracking steps.In the first step,the bipartite graph was used in inter-frame assignment to establish a bunch of tracklets.We also estimated the distance threshold to limit the cell migration,so the tracklets have high reliability.After the bipartite graph matching,the unmatched cells in the fore frame and later frame will be classified into cell division,cell fusion,cell enter and exit events.All these cells were embedded into the potential relevant tracklets with location information.In the second step,the cell events can be automatically identified and corrected based on the biological prior knowledge.At last,the search region of cell motion was calculated with the established tracklets in spatial and temporal.The tracking system is completely automated without initialization and manually correction.5 We obtained the cell lineages by correctly matching the mother cell and daughter cells in cell divisions.The lineages and trajectories were displayed in the 3D space.