Title

Segmentation and classification of MR brain images using artificial immune models

Date of Award

2005

Availability

Article

Degree Name

Doctor of Philosophy (Ph.D.)

Department

Electrical and Computer Engineering

First Committee Member

Akmal A. Younis, Committee Chair

Abstract

Many applications in brain imaging benefit from the ability to accurately and rapidly segment brain imaging data in an automated manner. Different segmentation approaches include the use of biologically inspired systems for the purpose of segmentation. The biological immune system is attractive due to the accuracy by which it differentiates between pathogens and normal body cells. The immune system cells are able to classify the proteins presented on the surface of cells as being self proteins, which belong to the body, or as non-self proteins. Accordingly, the cells carrying those proteins are either left to live, as in the former case, or attacked by lymphocytes, as in the later case.The MRI brain segmentation approaches presented in this dissertation are mainly inspired by the recognition process of the biological immune system. The approaches capture the main concepts by which the immune system recognizes pathogens and implements them in a numerical form. Although the biological mechanics of that process are far more complicated than their numerical representations, the presented approaches, which are mathematically formulated, were able to accurately and rapidly segment multi-spectral MRI brain scans.The MRI datasets that were used in the evaluation of the proposed models are the Harvard Morphometric Center database [1], and the BrainWeb database [2]. The accuracy was evaluated in terms of Tanimoto's overlap coefficient and Dice's similarity in comparison to manual segmentations established by an expert, in the case of the first source of data, or to ground truths established from simulator phantoms, in the case of the second source of data. Through extensive comparative studies with other MRI brain segmentation algorithms that were applied to the same datasets, the GAIN network developed in this dissertation exhibited comparable accuracy characteristics and enhanced performance characteristics in terms of segmentation.

Keywords

Engineering, Biomedical; Engineering, Electronics and Electrical; Health Sciences, Radiology

Link to Full Text

http://access.library.miami.edu/login?url=http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqdiss&rft_dat=xri:pqdiss:3198718