Image Registration Methods in MRI - Advance Image Processing in Magnetic Resonance Imaging - page 204

Calculate similarity
along all image pairs
N (N

Calculate the
hierarchical tree

− 1)/2

G1

G5

G2

G6

G3

G4

FIGURE 7.7 A hierarchical algorithm for multiple image alignment.

The main advantage of this approach is to perform a large number of
registrations between image pairs that are yet almost aligned (i.e., the ones at
the low hierarchical level). This reduces the processing time and increases the
registration effectiveness. The small number of registrations between groups at
high levels in the hierarchy can be eventually performed by sophisticated
registration algorithms.

7.7 BRAIN IMAGES REGISTRATION

Image registration plays an important role in MR brain imaging, in particular, in
functional image analysis. In functional magnetic resonance imaging (f MRI), in
fact, the signal changes due to the hemodynamic response is small compared to
signal changes produced by subject movement, heartbeat, and respiration. It
should be noted that subject movement of just 100

m can generate a change in
the signal larger than the one caused by activation. Respiration causes bulk
movements as well as changes in blood oxygenation that affect the BOLD signal.
Subject motion and respiration in MRI scanners cannot be completely eliminated,
so the functional images' alignment must be performed in the postprocessing
phase. This operation usually involves rigid 3-D registration.
In a single fMRI study, the activation map obtained from a subject by
processing functional images is usually superimposed on a high-resolution struc-
tural MR image (typically, a T1-weighted MRI). The registration between the
two images requires a rigid registration, but in this case the gray-level distribution
along the two data sets can differ due to the different acquisition techniques.

µ