Textural Feature Extraction and Multiple Instance Learning for Cell Segmantation in Histopathological Images

Abstract

Multiple instance learning (MIL) is a learning approach which is based on classification of bags of instances, as opposed to the traditional supervised learning paradigm. Multiple instance learning provides a natural way for modelling some pattern recognition problems which naturally have ambiguity such as object recognition, image and text classification. In addition, multiple instance learning paradigm gives more accurate results for those problems than traditional supervised learning paradigm does. In this study, we have applied multiple instance learning paradigm to cell segmentation problem in histopathological images by employing the intensity values in color space and textural information of pixels as features. Furthermore, to increase segmantation accuracy, we aimed to improve the results of pre-segmentation by implementing Markov random fields (MRF) method in the post processing step. Then, we presented the promising results in tables comparatively.