Prognosis of Glioblastoma Multiforme Using Textural Properties on MRI

• Author / Creator

  Heydari, Maysam

• This thesis addresses the challenge of prognosis, in terms of survival prediction, for patients with Glioblastoma Multiforme brain tumors. Glioblastoma is the most malignant brain tumor, which has a median survival time of no more than a year. Accurate assessment of prognostic factors is critical in deciding amongst different treatment options and in designing stratified clinical trials. This thesis is motivated by two observations. Firstly, clinicians often refer to properties of glioblastoma tumors based on magnetic resonance images when assessing prognosis. However, clinical data, along with histological and most recently, molecular and gene expression data, have been more widely and systematically studied and used in prognosis assessment than image based information. Secondly, patient survival times are often used along with clinical data to conduct population studies on brain tumor patients. Recursive Partitioning Analysis is typically used in these population studies. However, researchers validate and assess the predictive power of these models by measuring the statistical association between survival groups and survival times. In this thesis, we propose a learning approach that uses historical training data to produce a system that predicts patient survival. We introduce a classification model for predicting patient survival class, which uses texture based features extracted from magnetic resonance images as well as other patient properties. Our prognosis approach is novel as it is the first to use image-extracted textural characteristics of glioblastoma scans, in a classification model whose accuracy can be reliably validated by cross validation. We show that our approach is a promising new direction for prognosis in brain tumor patients.

• Subjects / Keywords

  • Texture
  • Survival
  • Decision tree
  • Machine learning
  • MRI
  • Glioblastoma
  • GBM
  • Prognosis

• Graduation date

  Fall 2009

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/R39G74

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.