Magnetic resonance in neuro-oncology: Achievements and challenges - Signal Processing in MRS with Biomedical Applications - page 271

The authors of Ref. [144] found a resonance at 3.8 ppm at long TE, to
which they did not give an assignment, but which was also characteristic of
meningiomas, and was absent from other brain tumors. Further complicating
the situation is that when patients are treated with mannitol, a resonance at
3.8 ppm will often appear, although in Ref. [144] only two of the nineteen
patients with meningiomas were receiving mannitol at the time of the MRS
examination. Contrary to claims of objectivity [76], the subjectivity of fitting
procedures has been clearly demonstrated and this can undermine diagnostic
accuracy in neurooncology.
Another important example of the problem of fitting when overlapping
resonances are present is illustrated in the study of Kaminogo et al. [252]. As
noted, these authors reported that the NAA to choline ratio at short TE was
of limited value in grading gliomas. They suggest that in gliomas the peak
around 2 ppm could contain other metabolites besides NAA, namely lipids at
2.05 ppm, and glutamine - glutamate at 2.1 ppm.
The data concerning myoinositol are contradictory for distinguishing tu
mors from normal brain tissue and from nonneoplastic processes, as well as
for brain tumor grading and histopathological typing. Most authors have as
signed the peak at 3.56 ppm to myoinositol alone, while some have viewed
this as a combined myoinositolglycine peak. Glycine is an inhibitory neu
rotransmitter, which also has a proton signal at this position. After release,
glycine is taken up by astrocytes, the site where myoinositol is also located
[211, 304]. It remains to be determined whether better distinction between
these two overlapping resonances of glycine and myoinositol would help clarify
some of the abovementioned diagnostic dilemmas.
Fitting procedures such as the LCModel have been noted to be particularly
problematic in the presence of large amounts of mobile lipids. Auer et al.
[298] point out that prominent broad resonances at 0.9 ppm and 1.3 ppm
are not fully modeled by the baseline spline functions of the LCModel and,
as such, lead to incorrect estimates of lactate and alanine. Inappropriate fit
of the entire spectrum with substantial phasing errors can in some cases be
the final, and wrong, result. These authors [298] suggest some procedures to
tackle this latter problem, with the aim of improving the accuracy of tumor
grading using in vivo proton MRS. They note, however, that limited spectral
resolution and SNR, as well as a lack of prior knowledge about the specific
lipid constituents of various disease states, presently limit the accuracy of
quantifying lipid and macromolecular constituents.
In their study using a long TE (272 ms), Kuznetsov et al. [305] omitted
the peak at 0.9 ppm attributed to lipids, because they considered it could be
biased by baseline fluctuations. Furthermore, they stated that limitations in
resolution prevented them from directly ascertaining the contribution of lipid
versus lactate to the 1.3 ppm peak. In Ref. [144] mixed forms of lipid plus
lactate doublet at 1.3 ppm were reported, using an even longer TE (288 ms).
This peak was prominent in the patients with metastatic lesions. As discussed
in chapter 2 (section 2.6), the offdiagonal FPT can adequately describe this