Multiplet resonances in MRS data from normal and cancerous prostate - Signal Processing in MRS with Biomedical Applications - page 369

real part of the complexvalued spectrum. The stability of convergence was
confirmed at longer partial signal lengths and at the total signal length for
the three problems under study. This has been true as well for all our other
applications of the FPT to MRS time signals as described in chapters 3, 6, 7,
9 and 10.
Prior to but fairly near convergence, at N P = 600 for the normal glandu
lar data and at N P = 500 for the normal stromal data, the phosphocholine
resonance had not yet been detected and the computed concentration of glyc
erophosphocholine was overestimated. For the malignant prostate data, even
though phosphocholine was resolved prior to convergence, the concentration
of GPC was still overestimated by about a factor of two.
In this light, mention should also be made here concerning the ratio of
GPC to PCho, in relation to the socalled “glycerophosphocholine to phos
phocholine switch” which has been suggested to be a marker of malignant
transformation for other tissues, notably breast [307], as discussed in the pre
vious chapter. These inaccuracies that appeared at intermediate stages of
calculation completely disappeared once convergence was attained. It there
fore can be considered of potential clinical importance for prostate cancer
diagnostics via MRSI to determine the exact number of true resonances and
the exact point of convergence.
In the present chapter, we have focused upon prostate cancer, which is the
most common cancer among men in the U.S. and in much of Europe, and
a major cause of cancer deaths worldwide. We have discussed the dilemmas
concerning screening, early detection and aspects of clinical decision making
regarding prostate cancer.
We have seen that relying upon conventional Fourierbased data analysis,
in vivo MRSI has made important strides in improving the accuracy with
which prostatic tumor and extracapsular extension are identified, as well as
helping to distinguish cancerous prostate from benign prostatic hypertrophy
[413].
In vivo MRSI has also contributed to various aspects of clinical management
of this malignancy. Nevertheless, major challenges and di culties remain, and
as of today, none of the existing noninvasive methods for prostate cancer di
agnostics have su cient sensitivity and specificity to be broadly recommended
for screening and other aspects of surveillance.
It would obviously be premature to render definitive conclusions about the
role the Padeoptimized MRSI in solving these dilemmas. From the input
data of Ref. [443] for a fairly small number of prostate cancer samples, it
cannot be stated with certainty which metabolites are best for identifying
prostate cancer and distinguishing this from normal stromal and glandular
prostate. Moreover, unfortunately, only the means and standard deviations
of the computed concentrations were given in Ref. [443] for the three types
of tissue.
Data from the individual patients, or the minimum and maximum of the
computed concentrations would have been valuable. Notwithstanding these