Cell nucleus segmentation of skin tumor using image processing

Automation and quantification of diagnosis of tumor cell images have been studied for these three decades in the field of medical imaging technology. Many techniques of image processing were proposed to solve problems such as nucleus segmentation and classification. But these studies have mainly focused on epithelial tumors. Nonepithelial skin tumors such as dermatofibroma (DF) and dermatofibrosarcoma protuberans (DFSP) have not been enough studied. DF is a benign tumorous disease and DFSP is a mid-grade malignant tumor. Recently, it is necessary that criterion of classification between DF and DFSP is quantitatively specified. In this paper a system for segmenting cell nuclei of DF and DFSP is proposed. Nuclei regions are objectively segmented and surrounded using strength edges by the system. Segmentation of arbitrary shaped nuclear regions and weakly stained nuclear regions is made. A dynamic thresholding method combining the Laplacian histogram with Ohtsu´s method is used for segmentation. A segmentation test was done using real tissue cell images of DF and DFSP to evaluate the validity of this system. Shape characteristics such as grade of similarity to a circle were also computed from the segmented regions to assure that some differences between DF and DFSP are expressed in its distribution.