Cell Segmentation Using Multiple Instance Learning Based Support Vector Machines

Abstract

In this study, in order to perform cell segmentation on histopathological images, multiple instance learning (MIL) paradigm is applied. In this context, support vector machines adapted to multiple instance learning problems are utilized in the classification modelling. Then, during the test phase, the test images are scanned and classified at pixel level separately and pre-segmentation images are obtained. In the post processing step, the Markov random fields (MRA) method is applied to improve the pre-segmentation results. In the conclusion, the classification performances of multiple instance based support vector machines and the conventional support vector machines are given comparatively.