Biomedical Engineering Reference
In-Depth Information
14
) Also, if a template coordinate system is associ-
ated with atlas information such as coordinate-dependent brain structure
labels (e.g., “cerebellum”) or functional labels (e.g., “visual area V1”), this
allows automatic estimation of which brain areas have been activated.
However, due to subject-subject variability and registration errors, this
type of activation reporting is generally less safe than looking “by eye” at
the combined activation and same-subject high resolution images,
described in the previous section.
This type of registration is both an intersubject and intermodality problem.
However, for fMRI, the problem can be broken down into three stages:
Talairach and Tournoux.
1.
Register the raw functional image with the structural image (as
above)
2.
Register the structural image with the template
3.
Combine the two transforms and apply to the original functional
activation images
The reason for using a two-step registration with the structural image as
the intermediary, rather than a one-step registration of the raw functional
image directly to the template, is that functional images typically have poor
anatomical detail, as the sequence is tuned to be fast and give good bold con-
trast. This, together with the fact that the raw functional images are usually
low resolution, means that registrations with the template are often not very
accurate, especially when large DOF transformations are used. However,
when registering with the structural image of the same subject, a low DOF trans-
formation is sufficient (usually 6 to 12 DOF), as the anatomy should be exactly
the same (provided geometric scanning distortions are minimized). The struc-
tural image, which has higher resolution and better anatomical contrast, can then
be registered to the template with a high DOF transformation (12 DOF or more).
This allows a good match to be found despite the difference in anatomy between
the individual and the template, since there is sufficient detail for the registration
to utilize in the structural images.
Finally, the reason the two transforms are combined in the last step into a
single transform (which takes functional data into template space) rather than
applying two separate transforms is that any errors due to interpolation then
occur only once rather than twice. This transformation is applied to statistical
activation maps (which are originally in the same coordinate system as the
functional image) so that they can be combined with structural information
in standard (template) space; this allows the locations of activation sites to be
reported in this standard space.
8.4.3
Group Analysis
To answer many questions about brain function, it is common to ask about acti-
vation in a group of subjects rather than in an individual. This may be in order
to increase SNR, as the amount of data is increased, or, more likely, for the
