Dedicated sequences for auditory fMRI

Summary of Chapter 5: Results - Continuous EPI:
The aim of the work presented in this chapter was to characterize the spatial
and temporal imaging characteristics of the Continuous EPI sequence
compared to the standard (conventional) EPI sequence, and to compare
and optimize its performance in relation to conventional EPI for the use in
auditory fMRI experiments.
Characterization Two characteristic image artifacts which appear in the
Continuous EPI sequence were identified and characterized. One, based on
the irregular echo spacing of the readout train, results in a multiple ghosting
artifact, which is hardly noticeable under practical conditions. The other,
resulting from gradient shape imperfections in the water excitation slice
select gradients, leads to an intensity modulation over multi-slice images
and can be corrected by a system-specific bias on the second of the two slice
selection gradients.
Further, the fat suppression using these water selective binomial pulses
was shown to be equally effective when applied to conventional and Continuous
EPI sequences.
For fMRI applications, the temporal signal stability of a sequence is a
crucial factor influencing the detection efficiency of functional activation.
Based on histograms of the voxel-wise relative standard deviation, the temporal
stability of Continuous EPI could be shown to be equal to that of
matched conventional EPI sequences at 1.5, 3 and 7 T. Finally, based on
sequence sound recordings, the use of “soft”, sinusoidally shaped gradient
ramps (which occupy a large part of the gradient duration) are beneficial
for the suppression of the pulsating sound component.
fMRI experiments The second part of this chapter consisted in the results
of seven auditory fMRI experiments in which conventional and Continuous
EPI sequences were compared.
In the first experiment, a 5 min pulsed sine tone block design experiment
of four volunteers measured on the 1.5T Sonata system, consistently showed
a larger BOLD response amplitude for the Continuous EPI sequence in all
subjects, on average by 40%.
The second experiment was a single volunteer 3 min proof-of-concept
experiment of an interesting extension of the Continuous EPI, the auditory
stimulation evoked by the periodic re-introduction of the pulsating gradient
sound component.
The third through fifth experiments compared the BOLD response amplitude
using conventional and Continuous EPI at 1.5, 3 and 7T field
strength using a pulsed sine tone stimulus of three different frequencies,
which were presented in a pseudo-randomized order over a total duration of
15 min. While in the 7T experiment, a single run on one volunteer showed a
50% increased BOLD response amplitude for Continuous EPI for two of the
stimulation frequencies, but no difference for the third, the predictive power
of this result was severely limited by the measurement of only one single session.
At 3 T, four volunteers were measured in two sessions per subject. In
both sessions both a conventional and a Continuous EPI run was measured,
with the order of those runs interchanged for the second session. While the
inter-session and inter-individual differences were larger than the difference
between sequences, the average BOLD response was greater for the Continuous
EPI sequence at all three frequencies by 12 to 25%. At 1.5 T, the
observed difference between sequences was even smaller, after complications
(RF artifacts and the exclusion of one subject) in the first trial, a second
trial on 6 volunteers (again measured in two sessions with interchanged sequence
order) gave average improvements for Continuous over conventional
EPI of only 3 to 11%, with even larger inter-individual and inter-session
differences. These large variations were also reflected by the failure to observe
consistently high significance levels in the activation patterns of the
individual runs. While some of the variance may be explained by the lower
contrast to noise level at 1.5T and the lower statistical power of the short
stimulation epochs (compared to fMRI experiment 1), a part of the more unusual
activation patterns could also be traced back to subject motion (both
stimulus-correlated and uncorrelated motion was observed).
The last two experiments aimed at exploring some of the sequence parameters
in the Continuous EPI sequence that could be expected to influence
the acoustic performance as well as testing some of the assumptions that had
been used in its implementation. Both experiments, even though measuring
6 and 8 volunteers, respectively, were plagued by insufficient statistical
power due to the even more extreme variance between sessions and subjects.
They both used 16 s epoch block designs which are known to yield optimum
statistical power [169, chapter 1]. Especially with the clear effect seen in the
block design fMRI experiment 1 in mind, these experiment could have been
expected to yield statistically significant results. Different factors could have
contributed to the failure to do so. On the one hand, rather long sessions of
around 45min were used (compared to about 25 min in exp. 1). Although
in these experiments an attempt was made to control subject attention by
simultaneously projecting a cartoon movie, varying attention levels could be
a part of the problem (or the distraction by the movie itself). On the other
hand, the backwardspeech stimulus which was used here might not have
been an ideal stimulus even though it resulted in consistent activation patterns
of high statistical significance. As a third factor, the scanner system
itself might have had some unexpected influence on the sequence differences
measured in these experiment (and the preceding experiment 4). Compared
to the Sonata system, the Avanto system used here is much better damped
acoustically (an important selling point, and easily detectable when listening
to the same sequence on both scanners). Here, it could be suspected
that the general increase in acoustic contrast between stimulus and scanner
sound might have reduced the difference in activation measured with Continuous
versus conventional EPI sequences, even though the trends from a
volume graded acoustic stimulation experiment do not support (nor contradict)
this, and stimulus volume is known to have no direct influence on
BOLD response amplitude ([117]). Of the non-significant trends found in
these experiment, maybe the most robust one is that of both sequences resulting
in the same BOLD response amplitude in a visual task (something
that has been shown clearly for the original implementation of Continuous
EPI [152]). Another trend, reproducible between the two experiments, was
that of an increase in BOLD response for increasing fundamental frequency
of the Continuous EPI sequence sound beyond the psycho-acoustically motivated
“critical point” of about 30 Hz. However, for the highest frequency
setting of 110 Hz measured in experiment 7, very low response amplitudes
were measured.
Summary of Chapter 8: Results - BURST:
In the course of the present work, the steady state behavior of gradient echo
BURST (URGE) was examined in detail. An approximate relation for the
optimal flip angle and the achievable signal amplitude could be given and
compared to more exact numerical simulations and phantom measurements.
Also, the size of the spoiler gradient necessary to obtain efficient dephasing
of remaining transverse magnetization has been determined to be at least
as large as the encoding gradient. As an alternative, gradient spoiling along
a different gradient axis has been proposed and found to be beneficial in
practical situations.
An URGE-EVI sequence was implemented on two clinical 1.5T and 3T
systems using the above findings. Sound levels during scanning were only
6 dB(A) above the ambient noise level, compared to about 30-35 dB(A) for
EPI sequences. Temporal signal stability was found to be reduced, compared
to the standard EPI sequence at 3 T, but still in a range where reliable
BOLD fMRI measurements should be possible. Also, the signal to noise ratio
was found to be comparable to an EPI sequence with similar parameters
(which were certainly not optimal for the EPI). In an fMRI experiment
combining auditory stimulation with a simultaneous finger-tapping task,
the BOLD sensitivity of the URGE sequence was established, as highly
significant activation was found in the expected location for the motor cortex
and the supplementary motor area. However, this experiment has failed to
show activation in the expected location of the primary auditory cortex.
This was very likely a result of pronounced distortion and signal drop-out
artifacts present in the functional image volumes.
As currently implemented, the usefulness of URGE-EVI for auditory
fMRI is therefore seen to be severely limited. The main obstacles are pronounced
distortion artifacts similar to those expected for a relatively lowbandwidth
EVI sequence and signal drop-outs due to the rather large voxel
size. For the currently used 3D segmented phase encoding scheme and given
the limitations imposed by the gradient hardware – and especially by a cumbersome
limitation in the scanner’s current image reconstruction software –
these issues are not easy to resolve. The removal of that software limitation
in the upcoming version opens prospects of making higher temporal and
spatial resolution possible, especially in combination with parallel imaging
techniques. This might also reduce the signal drop-out artifacts. While the
distortion artifacts might be relieved by using 3D phase encoding scheme
using only 2D segments (a stack-of-EPI trajectory), such schemes (which
multiply refocus the echoes of a BURST excitation) can be expected to
give rise to ghosting artifacts based on phase or magnitude errors in the
individual echoes.