Constructing and analyzing circular neighborhood cell-graph model of the histopathological tissue image

The smallest unit of living beings, which has structural and functional properties, is the cell. The investigating and understanding of cell and its behaviors is the basis of all biological sciences. Almost all cell behaviors are regulated by intracellular reactions. These reactions are initiated as a result of communication between cells. The distance between the cells and the locations of the cells influence this communication. In the same way, communication between cells affects the distance between cells and the locations of the cells. Thus, cell location based modeling and analyzing of histopathological tissue images can provide crucial information to the expert during the diagnosis and treatment process. In the literature, there are various cell-graph models such as t-threshold distance, k-nearest neighbors, Voronoi, Delaunay and colored graphs. However, to the best of our knowledge, there is not yet a model investigating the circular neighborhood relation between cells, although it is very important in terms of cell communication. Thus, in this study, a new proximity graph "Circular Neighborhood Cell-Graf" model has been developed, and the features that can be extracted from this model are presented. The cell nuclei are first segmented, and the overlapped nuclei are split to construct the cell-graph models. Then, cell nuclei are considered as vertex and cell-graph models are constructed by establishing connections between cells according to certain criteria. In this study, the "Probabilistic Nuclei Segmentation Algorithm", which does not require pre-processing, post-processing or manual parameter value, has been proposed to segment the nuclei. A novel algorithm, which split the overlapped nuclei from each other by determining the circularity center of the connected components, has also been developed to split overlapped nuclei. The proposed algorithms were tested on healthy and damaged kidney and liver tissue images, and the results were compared with related studies in the literature. The comparison and evaluation results show that the proposed segmentation algorithm is faster and more efficient than the commonly used K-Means; the overlapped nuclei splitting algorithm can successfully split overlapped nuclei with accuracy of 83%; histopatholojical tissue images can be classified with the accuracy of 95.7% by using "Circular Neighborhood Cell-Graph" model.