Biomedical Engineering Reference
In-Depth Information
CHAPTER 12
Applications of High-Performance
Computing to Functional Magnetic
Resonance Imaging (fMRI) Data
Rahul Garg
12.1 Introduction
Magnetic resonance imaging can be used to view the internal activity in a subject's
brain [1--21]. This is accomplished by measuring changes in the blood oxygenation
level in different parts of the brain [15]. It is the underlying neural activity that
determines blood flow to different regions of the brain. Such measurements are
performed while the subject is performing specific functions or mental tasks, giving
rise to the term functional magnetic resonance imaging.
There are many techniques to measure activity in a subject's brain, including di-
rect electrical recordings, and electroencephalography (EEG) [13]. Each technique
has its advantages and disadvantages. Obtaining electrical recordings is an inva-
sive procedure, but gives high-resolution spatial and temporal information. EEG
recordings are noninvasive and provide high-temporal resolution but poor spatial
resolution. Amongst the noninvasive techniques, fMRI recordings provide high
spatial resolution. The temporal resolution is about 2 seconds, which is reasonable
for capturing a variety of responses in a subject's brain during different tasks.
In a typical fMRI experiment, subjects are presented with some external stimu-
lus while they are being scanned inside an fMRI machine. Some examples of exter-
nal stimulus consist of instructions to perform certain tasks (such as finger tapping),
flashing of images of certain types (tools, buildings, pleasant or unpleasent images,
familiar faces, and so forth), and different types of sounds. In such experiments,
the subjects may also be expected to respond (say, by pressing a button) to the
external stimulus during the course of fMRI scan. The timings as well as details
of the external stimuli and responses are recorded along with the fMRI data.
12.1.1 fMRI Image Analysis Using the General Linear Model (GLM)
The fMRI data is analyzed to test a specific hypothesis about the brain function,
such as which areas in the brain become active during a finger tapping task. The
analysis of fMRI data has traditionally been done using the general linear model
method [9]. In this method, a suitable design matrix is constructed to answer
questions about the hypothesis. The design matrix consists of the time series of
different external stimuli and responses. The voxel time series of the fMRI data are
correlated with the design matrix. The brain areas that respond to specific patterns
263
Search WWH ::
Custom Search