Recapitulation of Pad´e-optimized processing of biomedical time signals - Signal Processing in MRS with Biomedical Applications - page 385

MRS for breast cancer detection have frequently been based upon the presence
or absence of a composite choline peak. This can compromise diagnostic ac
curacy, since choline may also be observed in benign breast lesions, and in the
normal breast during lactation. Furthermore, choline is often undetected in
small tumors that are then misclassified as benign [22]. Metabolite ratios are
also problematic for a number of reasons, including their variability according
to different echo times [252]. Several other confounding factors including the
cancer treatment itself, can affect metabolite ratios. Overall, the cutpoints
used to define malignancy vary widely from author to author [7].
One of the key obstacles to a greater use of MRS in clinical oncology has
been the lack of a uniform approach to data analysis. Data compatibility is
needed in order to make multicenter comparisons, indispensable for wider
routine application of these methods [306]. Differences in data processing
methods, rather than real differences, have been considered as the major con-
tributor to deviations among various clinical results [259].
At odds with the need for data compatibility is the requirement within the
FFT for fitting. As discussed, this can lead both to overfitting (spurious
peaks) as well as underfitting (true metabolites being undetected). This is
not only unacceptable to diagnosticians, but renders interstudy comparisons
tenuous, at best, unless the same in vitro basis set is used in some fitting
codes to predetermine the number of metabolites. We have cited several
examples of contradictory findings with respect to brain tumor diagnostics
directly related to whether or not a given metabolite was included in a basis
set of a linear combination of model in vitro spectra from signals encoded
separately [144, 145].
The problems of Fourierbased signal processing with postprocessing fitting
are most pronounced with respect to overlapping resonances. As seen repeat
edly, closelylying or overlapping metabolites are often the most important
for clinical oncology. Sharply counterposed to these limitations of Fourier
based data analytical techniques applied to MRS, is Padeoptimized MRS.
As demonstrated, the FPT unequivocally indicates the number of metabo
lites, including those that are overlapping, and provides accurate and precise
parametric information needed to determine metabolite concentrations. For
the two other clinical areas in oncology, breast cancer and prostate cancer, it
was the comparison between the MR total shape spectra and the component
spectra, that most dramatically demonstrated the advantages of the FPT.
For both these areas, in order to clearly distinguish between normal/benign
and malignant tissue, absolute certainty is required about the number of res
onances and their chemical shifts.
As presented in chapter 10, the FPT provided exact reconstruction of all
the input spectral parameters for the time signals corresponding to the nor
mal, benign as well as to the malignant breast lesions. The Pade absorption
spectra yielded unequivocal resolution of all the extracted physical metabo
lites, even of those that were completely overlapping (phosphocholine and
phosphoethanolamine at 3.22 ppm). The capacity of the FPT to resolve the