Image Processing Reference
In-Depth Information
250
200
150
100
50
0
10
20
30
40
50
60
70
80
Encode count
(a)
(b)
FIGURE 3.13
Non-uniform 3x-“hat” sampling pattern and spatial-domain aliasing result.
profile — densely sampled (1x-acceleration) close to the center of k-space (zero
frequency), less-densely sampled (2x-acceleration) farther away from the center
of k-space, and sparsely sampled (4x-acceleration) at the edges of k-space. The
ratio of the total number of selected lines to the number of lines in the FOV
determines the final acquisition acceleration.
Note that changing from a uniform to an irregular sampling pattern compli-
cates the spatial-domain aliasing pattern. Consequently, the SENSE equations no
longer solve the parallel MR signal equation in the least-squares sense. This led
to the development of reconstruction algorithms, e.g., SPACE RIP, Generalized
SMASH, and GRAPPA that do consider irregular subsampling patterns.
Figure 3.14 presents both a GRAPPA and SPACE RIP reconstruction of
3x-hat accelerated data. As seen in the figure, the aliasing artifacts visible in
Figure 3.14(b) are visibly suppressed. For comparison, a 3x-uniform accelerated
SENSE reconstruction is shown in Figure 3.14(c) with significant artifacts in
the reconstruction.
While both GRAPPA and SPACE RIP provide accurate reconstructions, there
are significant differences between them. On the sub-sampling side, SPACE RIP
(a) GRAPPA (3x hat)
b) SPACE RIP (3x hat)
c) SENSE (3x uniform)
FIGURE 3.14
Reconstructions of 3x-“hat”-accelerated acquisition data using (a)
GRAPPA and (b) SPACERIP. For comparison, a SENSE reconstruction of uniform-3x-
acceleration data is shown in (c).
Search WWH ::
Custom Search