Brain tumor segmentation of normal and lesion tissues using hybrid clustering and hierarchical centroid shape descriptor

Abstract

ABSTRACTRobust segmentation of the brain magnetic resonance (MR) images is extremely important for diagnosing the tissues quantitatively. It is crucial to detect the changes caused by the growth of edema and tumor in healthy tissues for better medical treatment planning. In order to increase the image quality, skull stripping or brain extraction is an essential pre-processing step in neuroimaging before the segmentation process. Hybrid algorithm made up of K-means clustering, and Fuzzy C-Means clustering (KFCM) algorithm offers advantages in the aspect of accuracy on soft tissues of brain MR images. KFCM algorithm clusters the images into the cerebrospinal fluid, white matter, gray matter and abnormal region. The segmented abnormal region has some false positive pixels which can not be removed by low order image processing techniques. In this study, we present the use of Hierarchical Centroid Shape Descriptor (HCSD) on the already segmented abnormal region by the above said method. The HCSD selects the region of interest only, i.e. abnormal region. Our algorithm offers considerable improvement in segmentation accuracy validated by the truth map. The quantitative evaluation and validation of experiments were carried out on 20 high-grade glioma suffering patient and 10 T1-weighted anatomical models of healthy brains.