A generalized method for automatic segmentation of neighboring cells

Robust and automatic segmentation of the neighboring cells remains a challenging problem due to the diversity of the cell types, frequently occurring artifacts, weak borders between adjacent cells, the arbitrary shape and large number of cells. Currently, the widely used segmentation and quantification tools are still manual or semi-automatic, which is time-consuming and labor intensive. With a breakthrough in the accuracy of threshold selection method proposed by the author, it is feasible now for a completely automatic and generalized method to fulfill the segmentation task. Even if the automatic method might not be perfect for segmenting all the neighboring cells in all the acquired images, the manual work after automatic segmentation will be minor. The proposed generalized method for cell segmentation comprises three parts: 1) Preparatory segmentation; 2) Cell identification; 3) Border delineation. We tested the proposed method on two types of cell images: 1) stained mice skeletal muscle micrograph; 2) human anaplastic astrocytoma micrograph. Experimental results show that the proposed method is effective in segmenting the neighboring cells and is thus promising for the future streamlined cell segmentation tool.