Date of Award

Summer 2009

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical & Computer Engineering

Program/Concentration

Electrical and Computer Engineering

Committee Director

Yuzhong Shen

Committee Member

Jiang Li

Committee Member

Zia-ur Rahman

Call Number for Print

Special Collections LD4331.E55 S535 2009

Abstract

Predicting and assessing tumor progression is important in brain tumor treatment. We attempt to use machine learning techniques to achieve consistency in assessing brain tumor progression. This thesis presents a prediction method of brain tumor progression by exploring a large MR database, which contains two patients ' complete records covering all their visits in the past two years. All ten MRI series, namely, apparent diffusion coefficient (ADC) , diffusion tensor imaging (DTI) , fractional anisotropy (FA), fluid attenuated inversion recovery (FLAIR), max eigenvalue (MAX), mid eigenvalue (MID), min eigenvalue (MIN) , post-contrast T1-weighted, T1- weighted, and T2-weighted, were co-registered to the corresponding DTI series at the first visit. Annotated normal and tumor regions were then overlaid to other registered MRI slices at the same visit. Intensity values of each pixel inside annotated regions of interest, at each visit, were extracted across all ten imaging series and concatenated to a 10-dimensional vector. Each of those feature vectors falls into one of three categories: normal, tumor, and normal but progressed to tumor at a later time. In this preliminary study, we focused on the trend of brain tumor progression during three consecutive visits, i.e., visit A, visit B, and visit C. A machine learning algorithm consisting of a multi-layer perceptron with an adaptive learning factor and a strategy for handling unbalanced data sets was trained using the data containing information from visit A to visit B, and the trained model was used to predict tumor progression from visit A to visit C. A floating search feature selection algorithm was also used to determine which MRI series is more effective for the prediction. Preliminary results showed that prediction for brain tumor progression is feasible. An average of 78.93% pixel-wise accuracy was achieved for tumor progression prediction at visit C. It was also found that some MRI series had only limited contributions to the prediction, increasing the accuracy insignificantly. In addition, this thesis also proposed a region growing method for brain tumor segmentation in T2-weighted MR images. The proposed region growing method makes use of two adaptive thresholds, which take into account the characteristics of the entire growing region and its growing front edge. Experiment results showed its superiority to some conventional region growing methods.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

DOI

10.25777/btjk-pw50

Download

Share

COinS