Recapitulation of Pad´e-optimized processing of biomedical time signals - Signal Processing in MRS with Biomedical Applications - page 384

method which has achieved this aim. The few published studies [334]-[339]
to date applying in vivo MRS to benign and malignant adnexal lesions have
been mired most notably by poor resolution and SNR, such that the results
are inconclusive. Thus, for this very small, moving pelvic organ, attempts to
glean diagnostically meaningful information from in vivo MRS using the con
ventional Fourier analysis have heretofore been modest. Yet, in vitro analysis
[340]-[343] reveals that there is a rich store of MRvisible metabolic informa
tion which distinguishes benign from malignant ovarian lesions.
Taken together, these considerations spurred us to perform the comparative
study [27, 29] of the resolution performance of the FPT and FFT for MRS
data as encoded in Ref. [343] from cancerous and noncancerous ovarian cyst
fluid. As reviewed in chapter 9, for these ovarian data, remarkably, all the
input spectral parameters were reconstructed exactly by the FPT using only
sixtyfour signal points, over two orders of magnitude fewer than with the
conventional FFT. Specifically, using the FFT, some 32768 time signal data
points were scanned (512 times more than needed in the FPT), and the FIDs
were zerofilled to 65536 data points, in order to obtain this same Pade level
of resolution in the Fourier absorption spectra. This provides direct evidence
of the powerful extrapolation features of the FPT.
For the benign and malignant ovarian cases, direct visual comparisons of
the reconstructed total shape spectra via the FPT versus the FFT at vari
ous partial signal lengths are also telling. For example, at a partial signal
length of N/32 = 32, the main spectral features are already well apparent
with Padereconstruction. In contrast, only the most rudimentary total shape
spectrum is generated by the FFT at that partial signal length. When the
total shape spectrum is fully converged at N/16 = 64 with the FPT, the
Fourierreconstructed spectrum is still completely uninformative.
Improved resolution and SNR as provided by the FPT could be of cru
cial help for improving the diagnostic yield of in vivo MRS in ovarian cancer
diagnostics and for malignancies of other deepseated, moving organs. Ap
plications of in vivo MRS and MRSI for neurooncology, breast cancer diag
nostics and elsewhere would also benefit greatly from this improvement in
resolution and SNR. Moreover, the capabilities of the FPT to clearly resolve
overlapping resonances would provide the opportunity to use short echo times,
thereby further enhancing resolution and SNR.
Regarding 2D MRS, our initial applications [439] indicate that the Pade
optimized 2D MRS also yield marked improvements in resolution relative to
the 2D FFT. We are currently exploring further this avenue via detailed com
putations on clinical 2D FIDs. This is an area which is attracting increasing
attention in clinical oncology, including brain tumors, breast and prostate
cancer [89, 387, 438].
The di culties of attempts at quantification when using Fourierbased meth
ods with postprocessing fitting, have often lead to the use of either a dichoto
mous variable of the Bernoulli type (e.g., presence or absence of a composite
choline peak) or to reliance upon metabolite ratios. Applications of in vivo 1 H