Research On Cell Segmentation Algorithm During Freezing Based On Active Contour Model

In biomedicine, cryopreservation is a way to achieve long-term banking of living materials, and it plays an important role in many research fields, such as organ preservation, tissue culture and keeping cells viable. The advancement of cryopreservation depends on the research of optimal cryopreservation protocols. Inappropriate freezing protocols tend to cause cryoinjury of cells during freezing, which is undesirable for cryopreservation. To develop optimal cryopreservation protocols, an intrinsic property of cells called water permeability has to be studied. For this reason, freezing experiments are needed. In these experiments, the projected areas of cells will decrease with the falling of temperature. Through figuring out these very cell areas, the cell volume variation caused by dehydration during freezing can be obtained, thereby determining the water transport parameters of cell membrane with data fitting. As for the calculation of cell areas, it can be easily achieved by cell segmentation. Therefore, it can be concluded that efficient algorithms for cell segmentation are of great significance to the measurement of water permeability and development of cryopreservation.Unfortunately, there exists so far no efficient and accurate segmentation method to handle this kind of image processing task gracefully. During freezing, the existence of extracellular ice and sharp changes of cell shapes altogether make it quite a challenging problem. Traditionally, researchers have to manually segment the cells, a procedure that is unreasonably time consuming and inaccurate. Therefore, in this thesis, we propose a novel approach to reliably extract cells and automatically work out the corresponding areas.The input to our proposed algorithm is temporal image sequences obtained from freezing experiments. Considering that cell positions will drift during freezing, we propose a greedy search strategy to track the approximate locations of cells in motion. Based on such coarse segmentation results, we further utilize a hybrid active contour model to extract the refined boundaries of cells. This model, not only shares both the advantages of edge-and region-based active contour models, but also possesses competitive ability to avoid undesirable segmentation results. As for solving the proposed model, the level set method with narrowband implementation is employed, and a reaction diffusion term is introduced to evolve the level set without reinitialization, which is time consuming and may bring many unexpected results. In addition, in this thesis we also utilize a multi-scale scheme to further accelerate the cell segmentation process. Experimental results demonstrate that, compared to classical image segmentation algorithms, the proposed method is efficient and effective in segmenting cells during freezing.The segmentation method proposed in this thesis not only opens up a new way to settle cell segmentation problems during freezing, but also brings new technical support for cryopreservation research.