Bilkent University
Department of Computer Engineering


A Clustering Approach for Histopathological Image Segmentation using High Level Semantic Features


A.Çağrı Şimşek
MSc. Student
Computer Engineering Department
Bilkent University

Cancer is one of the most important health problems that threat the human life. The survival of the patient heavily depends on the early diagnosis and correct grading, for which histopathological examination is routinely used. However this examination is subject to considerable amount of observer variability as it mainly relies on visual interpretation of pathologists. There are studies to automate the diagnosis and grading process for reducing the subjectivity in histopathological examination. The images used in these studies are assumed to be homogeneous. However, this may not always be true. Thus, before classification of the grade of cancer, the images should be segmented into homogeneous regions. Homogeneous regions of histopathological images have features which exhibit structural and spatial characteristics and they are more texture like features. In this work we use graph run-length matrices to extract these domain specific features and segment images. We approach the image segmentation problem as a clustering problem, since they are both unsupervised and try to group similar objects together. We investigate different clustering techniques like agglomerative clustering, graph partitioning based clustering and cluster ensembles.


DATE: 20 December, 2010, Monday @ 16:10