Biomedical Engineering Reference
In-Depth Information
FIGURE 8.4
A field map showing typical static magnetic field vari-
ations inside the brain. Dark areas represent fields less
than the external field, while bright areas represent
fields higher than the external one. The total range
corresponds to approximately
0.5 ppm.
sequence used, where these parameters are chosen as a compromise between
speed, contrast, and other factors. Unfortunately, EPI sequences, which are
commonly used to acquire fMRI time series, are particularly susceptible to
this form of geometric distortion. This is most notable near regions where
there is a tissue
air interface, such as near the temporal lobes and the sinuses
(see Figure 8.4).
Fortunately, a simple method is available which reduces this geometric dis-
tortion to low levels.
7
It involves acquiring a field map by appropriately com-
bining additional images taken using different gradient echo weightings.
That is, an image is produced that shows the strength of the deviation in mag-
netic field at each voxel. This field deviation is proportional to the amount of
distortion, which principally occurs along the phase encode direction for EPI
data. Therefore, by calculating the magnitude of the distortion, the image can
be transformed by “warping” the distorted voxel positions to the nondis-
torted voxel positions.
It is also necessary to correct the intensity of unwarped voxels, as the original
geometric distortion affects intensities: compressed regions get brighter and
expanded regions get darker. By applying the concept of “conservation of
intensity,” an unwarped image can be intensity-corrected. However, there can
also be a severe loss of signal associated with magnetic field inhomogeneities,
due to local gradients causing spin dephasing; it is not a straightforward mat-
ter to recover this signal loss. The above considerations should ideally be taken
into account when generating statistical inference during fMRI analysis, as the
spatio-temporal noise structure of the images is affected by both the original
distortion and the correction methods.
The resulting images then contain minimal distortion and can be better
aligned with images taken with other, less distorting sequences (such as a
T1-weighted structural image). Figure 8.5 shows an example of unwarping,
using the field map shown in Figure 8.4; on the left is the original distorted
image, and on the right is the corrected image. Note the large effect that the
unwarping has on the frontal lobe. Using such geometric distortion correction,
any later combination of fMRI statistics with either the same subject's high res-
olution scan or with other subjects' data will give more accurate results.
