Research On Cell Detection And Tracking Algorithm Based On Image Segmentation

Tracking cells in microscopic images is an important research field in medical image processing.By tracking and monitoring cells,researchers can understand the microcosmic state changes of organisms to predict,diagnose and treat disea ses.Manual observation of cells is a very time consuming and labor-intensive task,while the efficiency of this task can be effectively improved by using computer vision technology for automated cell tracking.Cell tracking algorithm is different from the common object tracking technology,we need to take into account both the tracking of cell trajectory and the detection of cell growth,proliferation and apoptosis.What's more,the intergenerational relationship between cells should be recorded.These are the difficulties that need to be solved in this paper.Other cell tracking algorithm can not be able to get a balance between the real-time of tracking and the accuracy of tracking.In order to improve the accuracy of cell tracking and detection in real-time cell tracking system,we design a detection-based cell tracking algorithm,which is composed of the detector and the tracker.The detector is composed of an image segmentation module and a feature extraction module.The tracker is composed of a data association module,an abnormality matching processing module and a prediction object verification module.In the detector,the image segmentation module performs image binarization and divides each cell into different connected components.The feature extraction module extracts the position,area,shape and color information of each cell by the connected component processing algorithm.By training convolutional neural network to learn the advanced features of each cell,a proliferation and apoptosis classification model is generated.In the tracker,the data association module performs object matching between adjacent frames based on the primary feature of the cell extracted by the feature extraction module.If an abnormal match occurs,the abnormality matching processing module will be started to determine the type of abnormal match according to the information generated by the feature extraction module and the data association module.The abnormal matches are mainly caused by cell proliferation or apoptosis,cell entry or exit events,and false positive detections or false negative detections.The prediction object verification module verifies the tracking result of the data association module and the abnormality matching processing module,and then corrects the tracking data.The modules cooperate with each other and operate interactively to form a robust cell tracking system.In this paper,a new object prediction algorithm is designed in the abnormality matching processing module,which can predict the information of the missed cells to reduce the number of object loss and improve the accuracy of object matching.In the prediction object verification module and the abnormal matching processing module,we combine the spatiotemporal context information of cells with the cell proliferation and apoptosis classification model to improve the accuracy of proliferation and apoptosis detection.Compared with other cell tracking algorithms,the algorithm designed in this paper can not only perform real-time cell tracking,but also obtain more stable tracking trajectory and more accurate cell lineage trees.