Histopatolojik görüntü bölütlemesi için çok seviyeli kümeleme bileşimi

In cancer diagnosis and grading, histopathological examination of tissues bypathologists is accepted as the gold standard. However, this procedure has observervariability and leads to subjectivity in diagnosis. In order to overcome suchproblems, computational methods which use quantitative measures are proposed.These methods extract mathematical features from tissue images assuming theyare composed of homogeneous regions and classify images. This assumption isnot always true and segmentation of images before classification is necessary.There are methods to segment images but most of them are proposed for genericimages and work on the pixel-level. Recently few algorithms incorporated medicalbackground knowledge into segmentation. Their high level feature definitionsare very promising. However, in the segmentation step, they use region growingapproaches which are not very stable and may lead to local optima.In this thesis, we present an efficient and stable method for the segmentationof histopathological images which produces high quality results. We use existinghigh level feature definitions to segment tissue images. Our segmentation methodsignificantly improves the segmentation accuracy and stability, compared to existingmethods which use the same feature definition. We tackle image segmentationproblem as a clustering problem. To improve the quality and the stabilityof the clustering results, we combine different clustering solutions. This approachis also known as cluster ensembles. We formulate the clustering problem as agraph partitioning problem. In order to obtain diverse and high quality clusteringresults quickly, we made modifications and improvements on the well-knownmultilevel graph partitioning scheme. Our method clusters medically meaningfulcomponents in tissue images into regions and obtains the final segmentation.Experiments showed that our multilevel cluster ensembling approach performedsignificantly better than existing segmentation algorithms used for genericand tissue images. Although most of the images used in experiments, containnoise and artifacts, the proposed algorithm produced high quality results.