Biomedical Engineering Reference
In-Depth Information
10
which is capable of acquiring a full brain
image every few seconds, with a typical within-slice resolution of 2 to 4 mm, and
slice thickness of 3 to 8 mm.
The change in MR signal induced by the BOLD response to brain activity is
typically 0.5 to 5 percent of the average image intensity within the brain. In addi-
tion, there is considerable change in the signal that is unrelated to the processes
of interest (that is, not related to the given stimulus). Therefore, separation of the
signal of interest from this “noise” requires sensitive statistical analysis.
To illustrate the type of statistical analysis required, consider a simple
“block-design” fMRI experiment. In this example, a visual stimulation (e.g.,
a flashing checkerboard) is shown to the subject for a period of 10 image
acquisitions followed by an equal period of “rest,” where a reference stimu-
lation (or no stimulation) is shown. This 10 ON, 10 OFF pattern is repeated
several times to enable the signal-to-noise ratio (SNR) in the following statis-
tical analysis to be large enough for activation to be detected. Given that a
typical image acquisition takes 3 seconds, such a session might last 10 min-
utes in total.
A particular location in the brain is considered “activated” if the intensity
variation follows the same pattern as the stimulus. More precisely, the inten-
sity for each voxel, as a function of time, is compared with the stimulus func-
tion (which is 1 in the ON condition and 0 in the OFF condition) by some
method, such as correlation. If the intensity and stimulus functions have a
high correlation (that is, they are similar), this voxel will be classed as
“active.” This is quantified by converting the correlation score into a proba-
bility of the observed intensity signal occurring purely by chance (the null
hypothesis); the regions with very low probability of having occurred by
chance are considered “active.”
in fMRI is echo planar imaging (EPI),
8.1.2
fMRI Analysis Overview
In fMRI analysis a large number of processing stages are required before the
final activation results are obtained. In-depth coverage of these stages is
beyond the scope of this chapter (see Reference 11 for more detail), but it is
helpful to have a broad understanding of the full analysis to appreciate the
general context for the issues of motion correction and registration in fMRI.
Although no standard analysis protocol is universally accepted, a typical
sequence of analysis steps for a single-session fMRI experiment is:
1.
Acquire and reconstruct the individual images.
2.
Phase-correct all time series for variations in timings of scanning
slices within the volume scan time (TR).
3.
Apply motion correction to correct for head motion.
4.
Spatially smooth the data to increase SNR and precondition later
statistics.
