Go to  Advanced Search

Estimation of the brain's hemodynamic response from fMRI images using the Eigenvector Based Algorithm for Multichannel Deconvolution

Show full item record

Files in this item

Files Size Format Description   View
ubc_2007-0406.pdf 7.862Mb Adobe Portable Document Format   View/Open
Title: Estimation of the brain's hemodynamic response from fMRI images using the Eigenvector Based Algorithm for Multichannel Deconvolution
Author: Gadala, Marwa
Degree Master of Science - MSc
Program Electrical and Computer Engineering
Copyright Date: 2007
Abstract: The ability to determine the brain's hemodynamic response without relying on an input function would be an extremely valuable asset in a large number of medical applications, and today, functional Magnetic Resonance Imaging (fMRI) is one of the leading methods in developing a better understanding of the human brain. Assuming the linear timeinvariant model for the observed fMRI response ([1], [2], [4]), this work provides an estimate of the hemodynamic brain response both on a regional and on an individual voxel level, as well as provides an estimate of the input signal that excited the brain's response. The solution to this problem is achieved using the Eigenvector-Based Algorithm for Multichannel Blind Deconvolution (EVAM) ([5], [6]) combined with Independent Component Analysis (ICA) [23]. The resulting estimate of the input signal produced by the proposed method could prove to be a valuable insight into the actual signal that triggered the brain during the experiment, and not the ideal signal that should have triggered it based on experimental observations. Also, contrary to previous works, no prior assumptions regarding the shape or order of the brain's response are made. When compared to non-blind identification algorithms traditionally used in the literature, the results show a significant improvement as the shape of the hemodynamic brain response conforms with current medical understandings. Furthermore, the estimated hemodynamic brain response is then used as a basis to determine active and inactive voxels. Two clustering methods, K-Means Clustering and Correlation-Based Clustering, are compared. Correlation-Based Clustering is found to be superior and is thus used to spatially map the active and inactive voxels. Spatial maps of important brain regions yield promising results where spatial sparseness is not characteristic of the images. Finally, a preliminary comparison between a healthy subject and a subject inflicted with Parkinson's disease yields promising differences, especially in the left primary cortex where very little activation was observed. Interestingly, symptoms of Parkinson's disease are thought to be a result of decreased stimulation of the motor cortex. Although no major medical conclusions can be made due to the risk of incorrectly attributing intersubject variability to differences due to Parkinson's disease, this preliminary comparison shows promising results that encourage future research in this area
URI: http://hdl.handle.net/2429/31753
Series/Report no. UBC Retrospective Theses Digitization Project [http://www.library.ubc.ca/archives/retro_theses/]
Scholarly Level: Graduate

This item appears in the following Collection(s)

Show full item record

All items in cIRcle are protected by copyright, with all rights reserved.

UBC Library
1961 East Mall
Vancouver, B.C.
Canada V6T 1Z1
Tel: 604-822-6375
Fax: 604-822-3893