Biomedical Engineering Reference
In-Depth Information
5.
Filter each voxel's time series, to remove slow temporal drifts and
high frequency noise.
6.
Perform the statistical analysis (creating a statistical parametric
map—an SPM).
7.
Threshold the SPM to find the “significant activated regions.”
Therefore, at the end of such an analysis, the result is a map showing the
“activated” voxels or clusters of voxels (for example, see Color Figure 8.6*).
These activation maps are often then subject to some higher-level analysis,
such as combining the low resolution results with the subject's high resolu-
tion image or combining results across a group of subjects.
To combine images effectively, an accurate linear or nonlinear (warping)
registration method is required. The transformations involved in linear reg-
istration (rigid-body or affine) have been described in Chapters 2 and 3 of
this topic, and nonlinear, or nonrigid, transformations are described in
Chapter 13. Registration is necessary for combining low resolution statisti-
cal images with a high resolution structural image, for combining statistical
results across several subjects in a group, or for transforming the results
into a standard coordinate system. Moreover, for best results, the fMRI
images should be corrected for geometric distortion that occurred during
the functional experiment, prior to registration. In general, as the raw func-
tional images do not have particularly good spatial resolution or contrast,
accurately registering these images with standard anatomical images is
quite a challenging problem.
8.2
Motion Correction
The first and most important registration issue associated with fMRI is
motion correction, which attempts to eliminate intensity changes in an fMRI
data set that are the result of subject motion during the experiment. This is
particularly important, as all subsequent analysis requires that each voxel
corresponds to a fixed location in the brain at every point in time. Since sub-
jects often move in the scanner during the course of an experiment (which
can take as long as an hour) the position of the head varies from image to
image. This is particularly true for clinical patients (as opposed to “normal”
subjects). Consequently, all images need to be registered to a consistent co-
ordinate system to enable further analysis.
The motion can vary over time from small, subvoxel motion to large, obvi-
ous motion. However, even subvoxel motion has a detrimental effect on the
statistical analysis—especially for smaller activations. Therefore, motion cor-
rection is required for almost all fMRI experiments.
* Color figures follow page 22.
