Research On Tracking Of Plant Cells Across Unregistered Microscopy Image Sequences

Cell analysis is one of the important topics in biomedical image analysis.The analysis and research of cell image sequence is of great significance t o the basic research of life science.The object of this study is the plant Shoot Apical Meristem(SAM)cells,unlike animal cells,plant cells are arranged in a honeycomb structure,and the cell population is highly compact,making the tracking method different from animal cell tracking.In this paper,the main work for plant cell image segmentation and plant cell tracking across unregistered microscopy image sequences are as follows:First,in the cell imaging process,to keep the plant cell alive for a long period of time,it is necessary to limit its exposure to the laser,this results in low contrast of cellular image quality with noise.We use a new segmentation model based on Full Convolutional Neural Network(FCN)called c-Unet to segment plant cell images.This model chooses Dice coefficient as loss function,accordingly,it adopts a series of training strategies to improve the effect of the segmentation,experiments show that the c-Unet model can automatically discard the noise,and accurately segment the cell boundaries.Second,this paper proposes a multi-seed dynamic local graph matching model for cell tracking across unregistered cell images.Firstly,we use a graph model to describe the topological relationship between cells.Then,a translation,rotation,and scaling invariant local graph feature composed of edge length pair ratio,angles between edges,and cell area ratio is extracted to find the multiple seed pairs for initial matching.In order to prevent tracking error accumulation,the existing local graph matching method is improved to be a dynamic matching algorithm.During each cell correspondence growing process,the neighboring cells of the matched cells are updated dynamically as candidate cells,and we match the most reliable cell pair with the least local graph feature distance from candidate cells at one time.Finally,after each group of matches is completed,the cell matching results are the voting output from the matching results produced by the multiple seeds chosen previously.Third,due to the influence of noise,and the disappearance of cells during longterm sampling process,the complete trajectory is cut into many discontinuous small tracklets,so we get a series of reliable collection of cell tracklet fragments after using multi-seeds dynamic local graph matching model for cell tracking.On this basis,we use Markov Chain Monte Carlo data association(MCMCDA)algorithm to analyze and get the complete cell trajectories.Finally,this paper experiments through multiple image sequence datasets with different quality,the experimental results show that our proposed algorithm can efficiently track the plant cells.