An Automatic Glioma Detection and Extraction Framework for the Removal of Shrinkage in Segmentation

Abstract

Medical image analysis is an essential part of clinical and laboratory research that has led to development of techniques for various medical applications. These techniques improve the sensitivity, specificity and provide complete characterization and grading of tumors. One of the tumor type that is a major threat to human population is Glioma found in adults. A lot of registered cases of malignant and benign tumor have been found. The technological advancement in images analysis techniques has gain popularity, as they provide a second opinion to the experts for making a better decision in treatment planning. They also help in avoiding biopsy that causes patient to go through post physiological stress. Image segmentation and classification of glioma tumors is an eminent research area that provides assistance to radiologist that aid in early detection and treatment planning of patient. For tumor extraction from MR Images, Graph Cut technique is a ubiquitous method in the medical image segmentation field. It is an energy based technique that is determined from region based statistics of the posterior probabilities. Graph cut energy minimization framework is popularly adopted for tumor segmentation in MRI. newline newlineIn previous years the initialization of this technique has remained manual due to lack of priori information. This priori information is the most essential requirement for constructing accurate seed values in order to extract tumor region through an automatic process. MRIs are constructed with various spatial parameters, topology and pixel intensities. Different regions in brain consists of different properties and their identification is mostly provided by the expert radiologist. Hence, understanding, selection and identification of these values becomes difficult with the complex brain structure. In this thesis, we propose that the quality of image segmentation is improved by developing an intelligent framework that is able to internally feed this prior information. newline newlineThe tumor extraction completely depends on th