Biomedical Engineering Reference
In-Depth Information
FIGURE 4.2
Transverse T
weighted multislice MRI images of the brain of a normal volunteer acquired
with a Gaussian slice profile and contiguous slices. (a) First examination; (b) second examination
displaced by approximately half the separation between adjacent slices; (c) image (b) resliced
after rigid-body registration to match (a); (d) subtraction image (c) minus (a); (e) as (d) but
using slices overlapped by 25%; (f) as (d), but using a 50% slice overlap. There are substantial
residual signals in (c) arising from incorrectly resliced pixel values because of undersampling
in the slice direction. These artifacts are reduced in (d) and virtually absent in (e).
2
When two datasets with different degrees of undersampling are to be aligned,
it may be valuable to consider keeping the least well sampled dataset static and
reslicing the other to match the former. Related issues of missing data concern
the edges of the region of support of the data, where otherwise adequately sam-
pled images will nevertheless suffer from intensity errors upon reslicing.
Finally, in this context it is significant that MRI data are generally presented
and stored in magnitude form although they are intrinsically complex, so
that each pixel actually has both a signal magnitude and a phase associated
with it. Use of magnitude reconstruction is expedient because it is both con-
gruent with the properties of most image display methods in that only one
scalar quality need be presented for each pixel, and because MRI phase varies
for instrumental and other reasons and is difficult to keep under control. How-
ever, the nonlinear process of forming magnitude images generates aliased
