Tracking Alive Cells In Cell Microscopy Sequence Images

Observing and analyzing live cells is one of the most important methods to study themicrocosmic world, and it is the frontier fields of clinics and bio-engineering. Cell segmentationand detection is very important for disease diagnosis and prevention, and cell movement trackingis very important to biological research and drug development.The article has first analyzed some representative or familiar target detecting method andtracking method, and according to the actual situation, used some proper segmentation andtracking algorithms. We implemented the entire process from segmentation to tracking and theintermediate processing. The main work is as follows:First, in target detection, according to the actual situation the subjects adopted someappropriate segmentation algorithms. Active contour model based on the hybrid model-basedalgorithm is used to segment MCAK, the video provided by LSDCAS is segmented by algorithmprovided by the Flux Tensor. Then we implement a series of subsequent operations to thesegmentation results. To the adhesion of cells, this paper uses watershed algorithm method basedon the regional characteristics of current frame for separating part of the adhesive, and unitedover-segmentation brought by watershed to reduce the over-segmentation brought by watershedseparating.Second, in the video processing field of live cells, for existing of segmentation errors suchas over-segmentation or under-segmentation, this paper first uses the theory of multiplehypothesis tracking algorithm for error correction of cell segmentation results, and uses thetopological structure and a certain threshold scheme to decide over-segmentation andunder-segmentation hypothesis. And false cell segmentation innovatively is modified, so that thecalculated cell parameters become more accurate, the topology used during tracking andcorrection becomes simple, thereby reducing calculation quantity, finally results in more correcttracking results. Experiments show that the effect of our method is obvious. Finally, For one-to-one matching algorithm such as Mean-Shift algorithm can hardly solvethe over-segmentation (i.e., fragments), cell division, under-segmentation (i.e., cell adhesion)in which the number or character of objects largely change, we use multiple hypothesis trackingalgorithm. Multiple hypothesis tracking algorithm is adaptive and has strong expansibility, caneffectively deals with the above problems.