Biomedical Engineering Reference
In-Depth Information
Although unsubstantiated claims were repeatedly put forward for objectivity, in
fact, the inherent subjectivity of fitting has conclusively been demonstrated. This
becomes particularly troublesome for overlapping resonances [
1
,
2
].
Using the FFT, a shape spectrum is obtained from pre-assigned frequencies
whose minimal separation is determined by the total acquisition time,
T:
The
FFT spectrum is defined only on the Fourier grid points
k=T .k D 0; 1; 2; :::/:
Attempts to improve resolution by increasing
T
and thereby decreasing the distance
1=T
between the grid points, lead to another problem, because clinical FIDs
become heavily corrupted with background noise for longer
Since FIDs fall off
exponentially, the larger signal intensities are observed early in the encoding. It is
thus advantageous to encode all FIDs as rapidly as possible, i.e., to avoid long
T:
T
at which mainly noise will be measured. Because of these two mutually exclusive
requirements within the FFT, attempts to improve resolution invariably lead to a
worsening of the signal to noise ratio (SNR). The FFT also lacks extrapolation
capabilities. Yet another reason for its low resolution is that the FFT is a linear
mapping, since its transformation coefficients or weights are independent of the
FID points [
1
,
2
].
25.3.1
Limitations of the FFT in radiation neuro-oncology
Several of the problems with current applications of MRSI in radiation neuro-
oncology are related to resolution and SNR. Attempts to improve the SNR have
usually entailed either increasing the acquisition time, or the volume of brain
tissue from which data is acquired. The latter leads to recording heterogeneous
voxels with a mixture of tissue types. Because of the importance of achieving
volumetric coverage of brain tumors, which themselves are often heterogeneous,
the SNR issues regarding MRSI are of major concern for radiation neuro-oncology.
Insufficient resolution and SNR limit the capability of MRSI to detect small foci of
brain tumors. Resolution and SNR for brain tumor diagnostics via MRSI have been
improved by using higher magnetic field strengths. However, detection of residual
brain tumor is troublesome even with 3T scanners. Another strategy has been to use
short echo times (TE) in attempts to capture clinically important metabolites that
decay rapidly [
9
]. However, at short TE the pitfalls of relying upon fitting can be
exacerbated. Moreover, since
T
2
relaxation times of various metabolites differ, peak
ratios can be affected by changes in TE [
2
]. Consequently, reliance upon metabolite
ratios becomes even more problematic.
In our systematic review [
2
], none of the metabolite ratios estimated via the FFT
unequivocally distinguished brain tumors from non-malignant brain pathology. Low
NAA can reflect loss of neurons with infiltration of brain tissue by tumor. However,
since it is a marker of neuronal viability, NAA can also be decreased in almost
any brain abnormality. On the other hand, choline can be low in very small or
necrotic brain tumors or tumors otherwise containing mixed tissues. The cutpoints
for metabolite ratios used to indicate the presence of malignant brain tissue are
