Research On Multi-cell Tracking Based On Interacting Multiple Model Filter Algorithm

The study of cell tracking issues is critical to establish ing cells' growth model and exploring its genetic structure and function.The main content of this article is to trace the plant shoot apical meristem cells in the microscopic image stack sequence.Tracking methods for animal cells and other regular objects is difficult to adapt to track the plant cells.The reasons are as below:(1)the plant cells are densely packed in a specific honeycomb structure;(2)the microscopic images have more noises,which interferes with the identification of cells and the generation of cell trajectories;(3)the sampling time of cells' images is so long that it is difficult to avoid the intermittent during the tracking process.To solve these difficulties,the main work and innovations of this paper are as follows:We propose a machine learning-based cell detection method applicable to different modalities to segment the plant cell image.The method consists of three steps: first,a set of candidate cell-like regions is identified.Then,each candidate region is evaluated using a statistical model of the cell appearance.Finally,dynamic programming picks a set of non-overlapping regions that match the model.The cell model requires few images with simple dot annotation for training and can be learned within a structured SVM framework.For cell tracking part,this paper designs a method that uses local graph matching model and Interacting Multiple Model(IMM)filter to achieve fast cell tracking.The filter interacts with Kalman filter models in three motion forms.Compared with the Kalman filter of a single model,the IMM filter with online parameter adaptation enhances the tracking of varying cell dynamics,and provides the additional capability of motion pattern identification.The local graph matching model proposed previously is used to describe the topological junctions between cells for cell-to-cell matching.The IMM filter is used to predict the movement of the cells,and then the local graph matching approach can search the target cells in the neighborhood of the predicted position in a relatively small area.The matched cell location is then regarded as an observation value to the motion filter for correction so that recursive tracking is achieved.Due to noises,disappearance of cell recording,or segmentation errors,it is easy to make a complete trajectory belonging to one cell divided into multiple discontinuous small tracklets.By using an Interacting Multiple Model filter tracking algorithm combined with local graph matching,a set of cell tracklets with high confidence is obtained.Later,we propose to use the maximum posterior probability method to match the cell trajectories according to their temporal,spatial and visual characteristics,and ultimately the multiple trajectories belonging to the same cell are connected again.Finally,we use multiple image sequence datasets to test the cell segmentation and tracking algorithm.It is verified that the method proposed in this paper can effectively achieve the parallel and stable tracking of plant cells.