Pad´e quantification of malignant and benign ovarian MRS data - Signal Processing in MRS with Biomedical Applications - page 301

highest early in the encoding. It is thus advantageous to encode the time
signal as rapidly as possible, i.e., to avoid long T when mainly noise will be
measured. This is particularly important for clinical signals encoded at lower
magnetic field strengths.
As we have extensively reviewed, the FPT provides not only the shape spec
tra, but also the vitally important parametric analysis, i.e., the quantification
from which the metabolite concentrations are obtained. Here, with only 64
data points out of 1024 sampled data, the FPT exactly reconstructed all the
spectral parameters to an accuracy of six decimal places for all the twelve
metabolite peaks. These parameters were then used to compute metabo
lite concentrations simply and unequivocally. With the standard Fourier ap
proach, metabolite concentrations are estimated from the shape spectra by
integrating the areas under the peaks or fitting the peaks to a subjectively
chosen number of Lorentzians and/or Gaussians.
Even for peaks that are well separated, as noted, this procedure of numerical
quadrature is subjective due to the uncertainty about the lower and upper
integration limits. However, when the peaks overlap, this standard method
for reconstructing metabolite concentrations is fraught with major di culties
and uncertainties; it is thus well recognized that “spectral crowding” creates
quantification problems [346]. This “spectral crowding” is not a problem for
the FPT, which via parametric analysis, yields reliable information about
the concentrations not only of isolated resonances, but also of those that are
overlapping [10, 11].
As discussed, several MRobservable compounds have been found to pro
vide some distinction between benign and cancerous ovarian lesions when in
vitro MRS is applied using high magnetic field strength and the conventional
laboratory specimen processing techniques. Notably, concentrations of adja
cent resonances such as threonine (1.33 ppm), lactate (1.41 ppm) and alanine
(1.51 ppm) and the nearly overlapping resonances isoleucine and valine in the
region of 1.02 ppm to 1.04 ppm differ significantly in these two types of lesions
[343]. The high concentrations of these branched chain amino acids are seen
as protein breakdown products due to necrosis and proteolysis.
However, none of these studies [341]-[343] reported any metabolite alone
or in combination with other metabolites that unequivocally distinguished
benign from cancerous ovaries. Even when the differences in concentrations
were statistically highly significant such as in Ref. [343], the ranges were still
overlapped.
One avenue for further investigation, and which is currently underway, is
to apply the FPT to largerscale in vitro experimental MRS data from benign
and malignant ovarian lesions. In this way, we are exploring the possibilities of
whether the FPT with its capacity to unequivocally yield exact quantifications
could also specify metabolite concentrations that more clearly distinguish can
cerous from nonneoplastic ovary. The FPT could thereby help establish the
standards of MRdetectable metabolite concentrations for normal versus spe
cific pathological entities of the ovary.