Recapitulation of Pad´e-optimized processing of biomedical time signals - Signal Processing in MRS with Biomedical Applications - page 383

and what the FPT can offer to help solve these problems that are critical for
cancer diagnostics. As presented in chapters 9 through 11, the advantages of
Padeoptimization have clearly been demonstrated for MRS data from three
problem areas within oncology: ovarian, breast and prostate cancer. We chose
these problem areas because of their urgent clinical importance. For the first
time, we have applied the FPT to time signals that were generated accord
ing to in vitro MRS data as encoded from (a) malignant and benign ovarian
lesions, (b) breast cancer, fibroadenoma and normal breast tissue and (c) for
cancerous prostate, normal stromal and glandular prostate.
The FFT is a lowresolution spectral/image estimator. One of the major
problems for Fourierbased MR imaging and spectroscopy is the need for long
imaging times. For MRI, image artifacts with edge distortions occur due to
truncations in the FFT, which cannot supply images that are simultaneously
bright and sharp. This is because sharpness of images stems from improved
resolution which in the FFT is obtained by increasing the number of grid
points per axis.
However, larger data sets unavoidably contain more noise, which reduces
brightness. Due to its extrapolation feature, the FPT allows sampling at
shorter acquisition time, and this enhances both sharpness and brightness,
as well as diminishing the Gibbs phenomenon. Envelopes of time signals,
such as those observed in MRS, decay exponentially, so that the larger signal
intensities are found early in the recording. It is, therefore, advantageous to
encode the time signal as rapidly as possible, i.e., to avoid long acquisition
times at which mainly noise is recorded.
Within the FFT, attempts to improve resolution can only be made by in
creasing acquisition time T, thus usually leading to a worsening of the SNR,
since clinical MRS time signals encoded at 1.5T become heavily corrupted
with background noise at larger acquisition times. For breast cancer diagnos
tics using 1 H MRS, this is especially troublesome, due to the need for lipid
suppression. One of the current strategies has been to increase the echo time,
TE, which diminishes the overlap with the lipid signal, but this is achieved
by a diminution in signal intensity. Poor SNR was cited as one of the major
reasons for false negative findings using 1 H MRS to detect malignant breast
lesions [116]. Moreover, some of the potentially informative metabolites for
identifying breast cancer have short T 2 relaxation times, and will have decayed
at longer TE [22]. Problems related to Gibbs phenomena are particularly im
portant for MRSI.
Poor resolution and low SNR represent a major limitation for current ap
plications of MRS in oncology. This has been a key obstacle for the use of in
vivo MRS to identify cancers located in deepseated abdominal/pelvic organs
that are also in motion due to respiration and/or peristalsis.
Ovarian cancer is a case in point, with major public health implications.
In many developed countries it is the leading cause of death from gynecologi
cal malignancies. Noninvasive early detection of ovarian cancer would confer
a major survival advantage, but yet as of today, there still is no diagnostic