Use of Spatial Information via Markov and Conditional Random Fields in Histopathological Images

Abstract

This study aims to increase the segmentation accuracy by using spatial information in biomedical histopathological images. The first step in the study is to provide pre-segmentation of H & E stained images using supervised learning methods, which are k-nearest neighbors algorithm, support vector machine and random forest. In order to build necessary classifier models, several training sets are created from intracellular and extra-cellular image patches extracted from histopathological images. As a two-class classification approach, supervised learning based segmentation are applied to test images in the evaluations. Spatial information should be used to improve the segmentation accuracy of output image obtained in the classification step. In the second step of the study, Markov and conditional random fields methods are utilized to exploit spatial information in histopathological images as a post processing approach. Comparative results prove that the use of spatial information via Markov and conditional random fields can be used to improve the segmentation accuracy of histopathological images.