Magnetic resonance in neuro-oncology: Achievements and challenges - Signal Processing in MRS with Biomedical Applications - page 273

nostics. We now demonstrate how this methodology could be implemented in
practice, as a handson tool for clinicians.
In Fig. 8.2 we give an example of how this can be achieved, from the anal
ysis performed in chapter 3. This was for an MRS time signal that closely
matches FIDs encoded via proton MRS from the brain of a healthy volunteer
[88]. We first show the parameters (position, width and absolute value of the
amplitude) for each of the 25 reconstructed peaks on the upper left panel (i)
in Fig. 8.2. On the upper right panel (iv) in Fig. 8.2, we give the metabolite
assignments. The T 2 relaxation times and concentrations for each of these 25
metabolites are also shown, as computed from the Padereconstructed ampli
tudes. Here, the Larmor frequency is 63.864 MHz which corresponds to the
static magnetic field strength B 0 =1.5T. The fraction of the concentration of
a given metabolite is also displayed relative to that which is most abundant.
In this example, peak number 6 corresponding to NAA at 2.065 ppm has the
highest concentration. Thus, for instance, peak number 16 associated with
choline at 3.239 ppm, has a concentration of 6.240 mM/g ww and, therefore,
its fraction relative to NAA is 0.56.
The component spectrum is presented with the peak numbers on the left
middle panel (ii) in Fig. 8.2 and the total shape spectrum on the right middle
panel (v). It is important to compare these two middle panels, from which it
can be seen that many of the peaks are closely overlapping, and could not be
identified from the total shape spectrum. On the bottom left panel (iii), the
peak assignments plus peak numbers are given for each of the 25 components,
with the localization of the peaks. The absorption total shape spectrum (as
the sum of all the absorption component shape spectra with all the peak
assignments) is shown on the bottom right panel (vi).
The advantages of the Pade analysis are particularly apparent when com
paring the left and right bottom panels (iii) and (vi) in Fig. 8.2. There,
it can be seen how tenuous it is to make any attempt whatsoever to surmise
which components are actually hidden underneath a spectral structure. Thus,
rather than reconstructing the mobile lipids under the two broad structures
in the range 1 2 ppm, as done unambiguously by the FPT, equally accept
able (in the leastsquare sense) results of fitting by the usual methods from
MRS (e.g., VARPRO, AMARES, LCModel, etc.) could “reconstruct” two,
three, four or more peaks that would all give the same absorption total shape
spectrum from 1 ppm to 2 ppm, similarly to the Lanczos paradox [78, 148].
Even more serious problems with clinically unacceptable uncertainties stem
ming from fittings are found to occur in many other parts of the spectrum
from panel (vi) in Fig. 8.2. Notably, any attempts to use fitting in order to
ascertain that the peaks close to 2.7 ppm are, in fact, nearly degenerate (each
comprised to two components with exceedingly close chemical shifts differing
from each other by 0.001 ppm) would be virtually impossible.
We consider that Fig. 8.2 would be very helpful for clinicians, since it
provides both a graphic and a quantitative overview of MRS. By enabling
repeated crosschecking between these two presentations, the clinician can