12169

11147

كارشناسي ارشد

مهندسي كامپيوتر - نرم افزار

اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر

1395

دوازده، [88]ص.: مصور، جدول، نمودار

ص. ع. به فارسي و انگليسي

شادرخ سماوي

نادر كريمي

محمدرضا احمدزاده، رسول اميرفتاحي

1395/11/26

كتابنامه

برق و كامپيوتر

مهندسي برق و كامپيوتر

ID11147

Segmentation Of Glioma Tumor In Brain Magnetic Resonance Images Mahsa Kadkhodaei m kadkhodaei@ec iut ac ir Date of Submission 2017 01 02 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisor Dr Shadrokh SamaviAbstractMedical images have a critical role in diagnosing the diseases and become an important assistingtool for doctors and experts Despite the aid of medical images there are still many difficultiesin detecting the abnormal tissues from healthy ones Manual segmentation method that is usedby experts can be prone to error not repeatable and time consuming due to fast growth of imagesin clinics during these years Computerized image processing methods represent a good solutionto many of the named problems Segmentation means to divide images into meaningful regionswhich have common features like image intensities As a result in this thesis we focus on Magneticresonance images of brain to segment tumor from normal tissues Existing methods for segmentingbrain tumors can be categorized into three broad family Learning based Generative and Combinedmethods Learning based methods obtain information from images to characterize the brain lesionsagainst other tissues Generative methods make use of prior information about physical structureand anatomy of brain Combined methods incorporate both the learning and generative methods Variation of human brain anatomy may lead to decrease accuracy of generative methods Alsothe learning based methods need a sufficiently large training data to generalize well into unseenimages In the end learning and combined based methods show promising results among the recentalgorithms The proposed methods in this thesis gain from learning techniques The 3D correlationand features of MR images play significant role in our methods In the first method supervoxelalgorithm creates meaningful groups of voxels Then feature extraction and classification are doneon supervoxels The second approach applies a two level classification on 3D patches where thefirst classifier extracts the main parts of tumor and the second classifier attempts to correct themisclassifications of the first one Finally the proposed methods for brain tumor segmentation havebetter or comparable results than some of the recent studies in terms of dice score Keywords 1 Image Segmentation 2 Magnetic Resonance Imaging 3 3D segmentation 4 Classification 5 Supervoxel 6 Multi Level Classification

شادرخ سماوي

نادر كريمي

محمدرضا احمدزاده، رسول اميرفتاحي