Advanced Image Reconstruction Methods in MRI - Advance Image Processing in Magnetic Resonance Imaging - page 49

(a) (b) (c)

(d) (e) (f)

FIGURE 2.2 SENSE reconstructions from a real data set acquired with 4 coils and R = 4.
(a) Median-filtered SENSE and (d) corresponding regularized reconstructions; (b) low-
resolution reconstruction from autocalibration and (e) corresponding regularized reconstruc-
tion; and (c) GS reconstruction and (f ) corresponding regularized reconstruction.

rate, known as the autocalibration scan [17,18,19] and uses these data to
reconstruct a low-resolution regularization image [20]. A high-resolution reg-
ularization image can also be created from these data using the GS model.
Details of the algorithm can be found in Reference 21.
Figure 2.2 shows a set of regularized reconstructions with different regular-
ization images from real experimental data acquired with four receiver coils and
R 4. As can be seen, different regularization images can affect the final recon-

struction.

=

λ

2.4.2.2 Selection of

A straightforward way to select the regularization parameter is to set heuristically

as a constant over the entire image. This method is not effective because the
condition of S varies at different locations. A more elaborate way is to select

λ

λ
adaptively using traditional regularization methods such as the L -curve or the
generalized cross-validation (GCV) methods [22]. The L -curve method was used
in parallel imaging with some success [20]. The GCV method works well in general
but sometimes gives biased results if the noise

d

∆ is highly correlated [22].