Statistical-Spatial Approach for Cell Classification in Histopathological Imagery

Abstract

In this paper, the effects of spatial relationships in the classification of labeled cells in histopathological images have been evaluated. Firstly, the features of the square windowing cells in different sizes related to Haralick, Tamura and color spaces have been extracted and have been merged. After that, the training and test data sets have been created by using 10 fold cross-validation. Then, data sets have been classified by k-nearest neighbors, random forest and support vector machine algorithms. The accuracies of different window sizes on these classifiers have been observed. As a conclusion, a method for optimal window size has been recommended and comparative results have been presented in the graphics and in the tables.