Conclusion and outlooks - Signal Processing in MRS with Biomedical Applications - page 398

can also handle encoded time signals. Further, we have shown that the FPT
can adequately solve the quantification problem in the presence of noise.
The clinicallyrelevant di culties within MRS are noise, as well as obtaining
the needed spectral parameters. These include the complex frequencies and
amplitudes, from which metabolite concentrations are reliably deduced. In
addition, there is a crucial fifth unknown parameter which is the number of
genuine resonances that are constituents of a given spectrum. Unless the
latter is determined unequivocally, underfitting or overfitting will inevitably
result, with peaks either being missing or falsely detected. Both of these severe
failures are totally unacceptable for clinical diagnostics and, moreover, they
are the ones that made physicians skeptical of all fittingtype data analyses
in MRS and MRSI. In sharp contrast, exploiting the powerful principle of
Froissart doublets, we have shown that exact signalnoiseseparation can be
achieved, such that all spurious resonances are unequivocally identified and
separated from the genuine information content.
In the thorough applications of the FPT presented in this topic, with di
rect relevance to cancer diagnostics, we used noisefree and noisecorrupted
theoreticallygenerated synthesized/simulated time signals to set up the stan
dard with a high level of control for obtaining the unambiguous solution
of the quantification problem. Having passed this most stringent test with
an unprecedented accuracy, robustness and clinical reliability, the FPT was
also applied with great success to clinically encoded time signals and noise
corrupted spectra that are abundant with overlapping resonances. We foresee
the subsequent step as application of the FPT in combined studies of malig
nant versus benign lesions, in which in vitro and in vivo time signals encoded
by MRS and MRSI are directly and comparatively analyzed, together with
histopathology for crossvalidation. This type of comprehensive and com
plementary imagehistopathology correlation is considered to be particularly
promising for improving the diagnostic accuracy of MRS in oncology.
It will be vital to apply the FPT to encoded in vivo MR time signals as
sociated with various cancers and nonmalignant tissue that have presented
differential diagnostic dilemmas, notably benign tumors, infectious or inflam
matory lesions. The FPT would yield unambiguous quantitative physical and
biochemical information. This could facilitate the development of normative
data bases for metabolite concentrations versus the corresponding findings
seen in malignancy. Such an opportunity would, in turn, provide needed stan
dards to aid in cancer diagnostics, identifying malignant versus benign disease
with specific patterns of departures from normal metabolite concentrations.
Overall, we anticipate that Pade-based optimization suggested in this topic
will be important for realizing the full potential of magnetic resonance spec-
troscopy and imaging in early cancer diagnostics and various aspects of cancer
treatment. Such a potential of these two modalities from magnetic resonance
physics is appreciated by biomedical researchers and clinical practitioners, as
a way to greatly improve both diagnostics and therapy.