|
nif:isString
|
-
Two different groups of 7 healthy volunteers each were scanned on a 3 T scanner (4 male, 3 female, age 28±7 years) and on a 7 T scanner (6 male, 1 female, age 33±9 years). All participants were right-handed according to a minimal score of 6 on the Edinburgh Handedness Inventory [52]. Study protocols were approved by the Ethics Committees of the Medical Faculty of the RWTH Aachen University and of the University of Pennsylvania. All participants gave written informed consent and were paid an allowance at the end of their participation. The experimental protocol which was used to compare real-time GE EPI and real-time SE SVS acquisitions consisted of the 6 following runs (Figure 1): a GE EPI and a SE SVS based functional localizer of the primary motor cortex (PMC loc), a SE SVS and a GE EPI based neurofeedback run targeting the primary motor cortex (PMC NF), and a GE EPI and a SE SVS based localizer of the visual cortex (VC loc).
Figure data removed from full text. Figure identifier and caption: 10.1371/journal.pone.0091620.g001 Sequence of data acquisition.GE – gradient-echo, EPI – echo-planar imaging, SE – spin-echo, SVS – single voxel spectroscopy, PMC – primary motor cortex, VC – visual cortex, NF – neurofeedback, loc – localizer. The primary motor cortex functional localizer runs consisted of finger tapping and baseline blocks, and the visual cortex functional localizer runs consisted of the presentation of a flickering visual checkerboard and baseline blocks. For the neurofeedback runs, the participants were instructed to adjust the speed and strength of their finger tapping so that a green horizontal bar would move up to the level of a predefined red horizontal target bar. All functional runs comprised 5 blocks of activation (i.e. finger tapping or visual stimulation, respectively), interleaved with 5 baseline blocks. Each block lasted 30 seconds, resulting in total run duration of 5 min.
Functional GE EPI and SE SVS data were acquired on a 3 T and a 7 T MR scanner (Siemens Medical Solutions, Erlangen, Germany) equipped with a transmit body coil and a 12-channel phased array head receive coil at 3 T, or a birdcage single-channel head coil (quadrature) at 7 T. EPI images were obtained with a single-shot gradient-echo T2*-weighted sequence with 300 repetitions (TR = 1000 ms, 16 slices, volumes matrix size 64×64, voxel size = 3×3×3.75 mm3, flip angle α = 77°, bandwidth = 2.23 kHz/pixel, TE = 30 ms at 3 T; TE = 28 ms at 7 T). At 3 T, the water spectra were acquired using a spin-echo PRESS protocol with 300 repetitions (TE/TR = 30/1000 ms, flip angle α = 90°–180°–180°, bandwidth = 1 kHz, acquisition duration = 512 ms). At 7 T, the acquisition protocol was slightly different with TE = 20 ms, bandwidth = 2 kHz, acquisition duration = 256 ms. Spectroscopic voxels were chosen as isotropic as possible based on the individual GE EPI brain activation maps for the motor and visual conditions (approximately 1×1×1 cm3). On both scanners we performed a manual calibration of the transmitter amplitude and optimized the gradient shim currents using Siemens manual shimming adjustments in order to improve the spectroscopic signal quality. Acquisition parameters were selected to obtain a robust T2* estimate and the BOLD effect. Note that for the SE SVS pulse sequence, the TE was selected as short as possible for the given PRESS protocols, i.e. we targeted the T2* contrast. This was also done to reduce T2 weighting, and to acquire early-echo data for a more accurate T2* approximation. The SE SVS estimates of the T2* contrast were barely affected by the flip angle for the transversal magnetization. This was because the inversion pulses contributed to the T1 saturation, because the post-acquisition delay time was long compared to the T2 of the tissue, and because the T2* estimates were calculated from the free induction decay function (FID) directly. Neither water nor fat suppression was applied for the spectroscopy protocols. The first 10 acquisitions were discarded to avoid T1 saturation effects. The visual instructions and feedback were shown to the subjects via MR-compatible goggles (Resonance Technology Inc. Northridge USA) on the 3 T scanner, and projected to an MR-compatible screen on the 7 T scanner. The data were exported in real-time to the local PC and processed with the custom-made software as described in [43], [53].
Data Processing and Feedback Signal Extraction: Immediately after acquiring the data from the primary motor and visual cortex localizer runs, the images were pre-processed with SPM8 functions (Wellcome Trust Centre for Neuroimaging, Queen Square, London, UK), i.e. realigned to the first scan of the respective localizer run, and smoothed with an isotropic Gaussian kernel with 4 mm full-width-at-half-maximum (FWHM). Next, we specified a general linear model (GLM) with regressors for the experimental conditions, and acquired thresholded t-maps. For the SE SVS acquisitions, the single-voxel was defined so that it covered those voxels that exhibited a significant positive BOLD response to the left hand finger tapping or visual stimulation, respectively [43]. The ROI for the GE EPI acquisitions was restricted to the SE SVS ROI (Figure 2).
Figure data removed from full text. Figure identifier and caption: 10.1371/journal.pone.0091620.g002 Illustration of the GE EPI and SE SVS ROIs.A GE EPI activation map and a single-voxel PMC ROI (blue) of a representative participant are shown on sagittal, transverse, and coronal planes of this participant’s structural scan. The PMC SE SVS ROI (size approximately 1×1×1 cm3) was defined to cover the voxels that exhibited a significant positive BOLD response to the left hand finger tapping. The GE EPI ROI was restricted to the SE SVS ROI.
During the neurofeedback primary motor cortex GE EPI runs, the EPI volumes were first realigned to the first volume of the primary motor cortex functional localizer run. The feedback signal, which corresponded to the average activity within the ROI, was then calculated as soon as a new volume was acquired. During the neurofeedback SE SVS runs, the acquired water spectra were shifted to zero, filtered with a Gaussian filter, the eddy currents were compensated, and the water spectra were phase-corrected [54]. The feedback was provided after each FID acquisition as an absolute T2* measure which was estimated with the statistically optimized linear regression approach [43]. The optimal linear regression length was estimated based on the SE SVS primary motor and visual cortex runs. After the feedback signal was extracted from either SE SVS or GE EPI acquisitions, the signal was processed in order to reduce noise and to remove spike-like artifacts using our custom-made real-time software [53]. For the GE EPI acquisitions, the head motion parameters were taken into account, but head motion parameters were not available for the SE SVS acquisitions. To ensure that motion artifacts did not cause significant SE SVS signal distortions, we located relatively small ROIs within large active zones revealed by the primary motor and visual cortex localizer runs. Inter-run head movements between the SE SVS runs were controlled by acquiring GE EPI scans before and after the SE SVS runs; they were less than 1 mm.
Time Courses Quality Measures and Comparison Analysis: The comparison analysis between GE EPI and SE SVS acquisitions was based on their CNR, percent signal change, and t-statistics, and was performed separately for data acquired at 3 T and at 7 T. For SE SVS, statistically optimized linear regression was applied to the acquired FID in the time domain. The natural logarithm of the FID can be simplified assuming that the water signal is the dominating component in the acquired FID, and that all other proton sources of the signal are negligible [43]:(1)with water time constant T2* and amplitude A. The linear regression was subsequently applied to the absolute logarithmic curve (ln(|FID|)) and determined by the optimal linear regression length (OLR) [43]. To compensate for line-broadening caused by applied Gaussian filtration, T2* estimation function was weighted with the correspondent filter coefficients [42]. The optimal linear regression length was estimated in the sense of a statistical measure, i.e. the maximum t-value in the distribution of t-values of time series computed for a set of linear regression lengths (Figure 3; red curves). The processed signal in Equation [1] may still have a large non-linear component because of the inadequate shimming conditions (Figure 3; blue curves), which can complicate the regression analysis of the acquired data. However, despite its simplicity, the proposed OLR approach has been shown to provide reliable T2* estimations at high and ultra-high magnetic fields [42], [43].
Figure data removed from full text. Figure identifier and caption: 10.1371/journal.pone.0091620.g003 T2* approximation of the SE SVS data acquired at 3 T and at 7 T.The linear regression fits (red) are shown for single PMC ln|FID|’s (blue) for representative participants at 3 T (A) and at 7 T (B). Optimal linearization lengths are 0.18 s at 3 T (t = 41.6, p<0.001), and 0.052 s at 7 T (t = 16.7, p<0.001).
Because the GE EPI voxel intensity is proportional to exp(−TE⋅R2*), the T2* values estimated from SE SVS time courses were transformed to using the applied echo time TE and arbitrary scaling. This allowed for a direct comparison between GE EPI and SE SVS acquisitions. Block-related averages were averaged across the time-course condition/baseline periods. The percent signal changes (Δ%) were estimated as an average from block-related condition/baseline 30-point plateaus. For the statistical analysis of the BOLD signal changes, we specified general linear models (GLM) with regressors for the experimental conditions defined in SPM8 (Welcome Trust Centre for Neuroimaging, UK). Each participant’s fMRI motion parameters were included into the GLM as nuisance regressors. Effects on the time-course quality ratings were analyzed in repeated-measures ANOVA for all data sets with functional run (PMC, PMC NF, and VC), MR scanner (3 T and 7 T) and acquisition technique (GE EPI and SE SVS) as within-subject factors. To further evaluate the difference between two samples, standard two-sample t-tests were used (t- and p- values; one-tailed). To calculate the CNR, we estimated differences between signal means during baseline and activation blocks, and their residual variances:(2)where condition/baseline is the time course of the ROI in the functional localizer condition and baseline, respectively. All computations were carried out on a standard PC in Matlab 7.10 (The Mathworks, Natick, MA). The custom-made neurofeedback toolbox is available on request from the corresponding author.
|
![[This page as RDF]](http://data.gesis.org/somesci/static/rdf-icon.gif){kind=icon}