SEARCH

SEARCH BY CITATION

Keywords:

  • echo volumar imaging;
  • EVI;
  • fMRI;
  • single shot;
  • 3D

Abstract

  1. Top of page
  2. Abstract
  3. THEORY
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

Echo volumar imaging (EVI) is a 3D modification of echo-planar imaging (EPI) that allows data from an entire volume to be acquired following a single RF excitation. EVI provides a high rate of volumar data acquisition, which is advantageous for functional MRI (fMRI). However, few studies to date have applied EVI to fMRI, since because of gradient hardware limitations EVI generally has to be used with long sampling times, resulting in high sensitivity to susceptibility-induced distortions. In this study we modified the EVI sequence to improve its suitability for fMRI. The sampling time is reduced by the use of a high gradient-switching frequency, a small number of echoes, and outer volume suppression (OVS); rewind gradients ameliorate Nyquist ghosting; and phase correction via a calibration scan reduces ghosting and distortion. It is shown that the modified EVI sequence allows fMRI data to be acquired with a temporal resolution of 167 ms. Magn Reson Med, 2006. © 2006 Wiley-Liss, Inc.

Echo volumar imaging (EVI) (1) is a 3D extension of echo-planar imaging (EPI). In this sequence, a slice-selective RF pulse is used to excite signal from a thick slab. The signal is then spatially encoded using a switched gradient applied in the readout direction, in conjunction with phase encoding via two blipped gradients applied in orthogonal directions (Fig. 1a). Although EVI is an excellent technique for acquiring Tmath image-weighted volumar data at high speed (2–5), it has not been widely used for functional MRI (fMRI) studies. This is mainly a consequence of the heavy demands imposed on the gradient hardware by the need to encode 3D information in a single free induction decay (FID). In EVI this encoding is accomplished by generating a large number of gradient echoes (equal to the product of the number of pixels spanning two dimensions of the data set). Because of limitations on the achievable gradient strength and rise time, the sampled echo train generally has a long duration. The resulting low voxel bandwidth in the Fourier-transformed data means that EVI suffers from a high sensitivity to image distortion due to magnetic field inhomogeneity. This distortion occurs in the direction of the less frequently applied blipped gradient (Fig. 1a), which is typically the slice-selection direction, in which the image matrix dimension is also generally the smallest. The long echo train used in EVI also limits the value of the minimum echo time (TE) when full k-space data are acquired. This can limit the blood oxygen level-dependent (BOLD) sensitivity of fMRI, since the TE can become longer than the Tmath image of gray matter (GM), particularly at high field.

thumbnail image

Figure 1. a: Timing diagram for the conventional EVI sequence. b: Timing diagram for the modified EVI sequence. Rewind gradients (gray) are applied along the x-axis at the same time as the phase-encoding gradient pulses are applied along the z-direction. Selective RF pulses are played out in conjunction with an x-gradient followed by crusher pulses (gray) to suppress the signal from outside the image matrix.

Download figure to PowerPoint

In the standard implementation of EVI, alternate planes in k-space are acquired using a different polarity of the frequently applied blipped gradient, as shown in Fig. 1a. This can result in a significant Nyquist ghost in the slice-select direction due to mismatching of data in successively acquired planes of k-space. The resulting ghost generally has a more serious impact on image quality than the Nyquist ghost formed in EPI. In EPI the ghost artifact appears offset by half the field of view (FOV) in the blipped gradient direction, whereas in EVI (Fig. 1a) the Nyquist ghost is separated from the image by half the FOV in the slice direction. Since in EVI data the object typically spans the whole FOV in the slice direction, while in the blipped direction of EPI data this is not generally the case, the overlap of ghost and image is more severe in EVI data. In addition to increasing the deleterious impact of the ghost artifact, this does not yield a region in which ghost-only signal can be found, thus limiting the use of image-based ghost correction (6).

In this paper we describe a number of modifications to the EVI sequence that are directed toward alleviating these problems and allowing EVI to be more straightforwardly used in fMRI studies. We demonstrated the efficacy of the modified EVI sequence by applying it to somatosensory fMRI studies in which a high rate of volumar acquisition (TR = 167 ms) was achieved. The use of an EVI readout combined with a variety of spin-preparation prepulses, including those used for inversion recovery (IR) and arterial spin labeling (ASL), is also briefly described.

THEORY

  1. Top of page
  2. Abstract
  3. THEORY
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

In the EVI sequence shown in Fig. 1a, data are acquired from Nz, kx-ky planes of k-space. Each plane is scanned using Nx gradient echoes that are generated using the rapidly switched gradient that is applied in the readout (y-) direction, and each echo is sampled using Ny data points. The echoes are phase-encoded by the application of a blipped x-gradient during each reversal of the switched gradient. The blipped gradient in the slice-select (z-) direction is applied less frequently, being pulsed on only after acquisition of each kx-ky plane. Fourier transformation of the sampled data yields an image matrix of size Nx × Ny × Nz.

In the original EVI sequence (1) the sign of the blipped x-gradient is reversed after acquisition of each kx-ky plane. Alternate planes in k-space are thus acquired under opposite polarity x-gradients and have to be “time-reversed” prior to Fourier transformation (1). As a result of this reversal, phase evolution due to field inhomogeneities occurs in the opposite sense in alternate planes, giving rise to a mismatch between adjacent data points in the kz-direction. After Fourier transformation, this is manifested as a Nyquist ghost that appears shifted by half the FOV in the slowly sampled slice-select (z-) direction, particularly in regions with significant field inhomogeneity. In the modified EVI sequence (Fig. 1b), a “rewind” gradient pulse (7) of duration TRW is played out in the x-direction between acquisitions of successive kx-ky planes. This gradient pulse has an area that is equal and opposite to the total area of the blipped gradients used in traversing one kx-ky plane. Scanning of each plane therefore begins from the same point in k-space, and the blipped x-gradient is always positive. Since all k-space planes are sampled with the same polarity x-gradient, there is a smooth evolution of phase, which reduces Nyquist ghosting and also simplifies the data reordering that is required prior to Fourier transformation.

The duration of the echo train, TEC, of the EVI sequence shown in Fig. 1b is

  • equation image(1)

where τ is the duration of each echo acquired under the switched y-gradient, and TRW is the length of the rewind gradient pulse, which is usually small compared to Nxτ. In EVI, the spatial distortion in the z-direction due to a frequency offset, δω is proportional to δωTEC. Therefore, to achieve an acceptable level of spatial distortion, the number of echoes sampled (Nx × Nz) and/or the time per echo (τ) must be limited. The choice of τ is a compromise between limiting field-inhomogeneity-induced distortions and maximizing the signal-to-noise-ratio (SNR), which is dependent on equation image at fixed TE. The minimum value of τ is also restricted by the achievable gradient strength and rise time.

The duration of the echo train, TEC, also influences the minimum TE (TEm) that can be achieved in EVI. TEm depends on TSL, which is the duration of the slice-selection module and the gradient pre-excursion pulses that are needed to ensure data sampling begins in one corner of k-space, plus the time then needed to reach the center of k-space in the EVI echo train:

  • equation image(2)

Large values of TEC extend the minimum achievable TE, thus leading to increased Tmath image decay and potentially to reduced BOLD sensitivity.

In the experiments described in this paper, a total of 256 echoes (Nx = 32, Nz = 8) were acquired per image using TRW = 1.04 ms. For all experiments, a high gradient switching frequency of 1.9 kHz (τ = 260μs) was used, and 64 points were sampled in the y-direction. These parameters resulted in a value of TEC of 75 ms, which gave a voxel bandwidth of 13 Hz in the z-direction. TSL was limited to 6.5 ms, giving a TEm of 44 ms. Relatively high spatial resolution was achieved within the small matrix by using a surface coil and outer volume suppression (OVS) (7) to limit the detected signal to the FOV, thus avoiding wraparound artifacts. Two double-slice-selective RF pulses, each of which were followed by spoiler gradient pulses, were applied in conjunction with an x-gradient prior to each z-slice selective 90° pulse. These were designed to saturate two slabs (each 3.4 cm wide and separated by 9 cm) so as to prevent aliasing of signal from outside the FOV in the x-direction (Fig. 1b) (8). The volume sampled in the z-direction was limited by tailoring the thickness of the slab excited by the selective 90° RF pulse.

The calibration scan method, which is commonly used in EPI (9), was then adapted for use with EVI to reduce Nyquist ghosting and image distortion further. A single calibration data set was acquired with no blipped gradient applied in the z-direction. Fourier transformation of these data with respect to kx and ky yields a set of Nz, 2D projections along z, which can then be used to form maps of the average phase evolution due to local field inhomogeneities, ϕn(x,y), within the excited z-slab at times TSL + (n − ½)TEC/Nz for n = 1 to Nz. One can then phase-correct the 3D data acquired with z-gradient blips by subtracting ϕn(x,y) from the phase of the image data in the nth plane after Fourier transformation with respect to kx and ky, but prior to transformation with respect to kz.. This removes phase mismatches between alternate planes of k-space data and eliminates the distortion due to the average field offset across the slab at each x-y location. Distortion due to field variation with z-position remains. Nyquist ghosting that occurs in the x-direction in EVI data due to mismatch of echoes acquired under positive and negative read (y-) gradient lobes can be reduced in the conventional manner by acquiring an echo train with both x- and z-blipped gradients switched off and then using these data to phase-correct alternate echoes (9). Such a correction did not prove necessary in the work described here because we carefully adjusted the switched (y-) gradient waveform and used a small interecho spacing.

Interleaved imaging approaches employing segmented k-space acquisitions (10) can be advantageously combined with EVI. When spatial resolution and matrix size are kept constant, interleaving allows TEC to be significantly reduced and the voxel bandwidth to be correspondingly increased, although temporal resolution is compromised. Figure 2 shows the k-space trajectory for one implementation of interleaved EVI, in which data from alternate kx lines are acquired from two separate excitations. The number of echoes acquired per excitation is halved, which reduces the distortion by approximately a factor of 2 when τ and spatial resolution are kept constant. When alternate kx lines are sampled in different excitations, it is necessary to employ an echo time TE shift (ETS) of duration τ/2 in one of the acquisitions to ensure smooth evolution of phase and amplitude in the final interleaved data set (11). In this case the ghost due to a mismatch between data acquired in different segments is shifted by half the FOV in the x-direction.

thumbnail image

Figure 2. k-Space trajectories for an interleaved EVI sequence in which alternate k-lines are sampled in the two passes. The trajectories for the first and second passes are shown with a thick line and a thin line, respectively.

Download figure to PowerPoint

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. THEORY
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

The subjects and phantoms were scanned using a custom-built 3-Tesla scanner that employed an insert head-gradient coil (12). A 27-cm diameter TEM volume RF coil was used for RF excitation, and a 4-cm-diameter surface coil was used for signal reception. A spherical gel phantom that consisted of four equally sized segments, each filled with different concentrations of agar gel doped with Gd-DTPA to simulate approximately the T1 and T2 of brain tissues, was used in the initial experiments. Subsequent experiments on human subjects were approved by the local ethics committee, and all subjects gave informed consent.

Images were acquired in sagittal orientation, with the switched gradient applied in the superior–inferior direction and the slice-select gradient applied in the mediolateral direction. The z-gradient coil was driven in resonant mode so as to allow generation of a large, rapidly varying gradient with the limited available amplifier drive voltage (300 V). The gradient coil, which had an inductance of 280 μH, was connected in series with a 25 μF capacitor, yielding a resonant frequency of 1.9 kHz. Nonlinear sampling of the echoes produced under the sinusoidal switched gradient waveform (13) was used to produce a uniformly spaced grid of sample points in k-space. Phase reference maps, with the blipped gradient pulses in the slice-select direction switched off, were acquired at the end of each scan session. Calibration scan correction using these reference maps was applied to all images shown, unless indicated otherwise.

EVI Sequence Development and Assessment

To evaluate improvements in image quality provided by the modified sequence, we acquired phantom images using both a conventional EVI sequence (Fig. 1a) and the modified sequence (Fig. 1b) incorporating rewind gradients. Both data sets had a resolution of 3 × 3 mm2 in-plane and 1.5 mm in the slice-select direction, and a matrix size of 64 × 32 × 8. To illustrate the effect of applying a calibration scan correction to EVI data collected with the modified EVI sequence (Fig. 1b), we acquired five 3D volumes from a brain, along with a reference phase map, and averaged the images formed both with and without correction. The interleaved implementation of the sequence was also tested on the gel phantom.

Applications of the Modified EVI Sequence

BOLD fMRI Using EVI

EVI was used to measure activation of somatosensory cortex with high temporal resolution in three subjects. The 4-cm-diameter receiver-only surface coil was positioned over the left somatosensory cortex. Single-shot EVI data were acquired with a resolution of 3 × 3 mm2 in plane and 1.5 mm in the slice-select direction, over a 64 × 32 × 8 matrix, using a flip angle of 30°. Six volumes, each of which comprised eight sagittal slices, were sampled per second (TR = 167 ms). A block paradigm using vibrotactile stimulation of the tip of the right thumb via a piezoelectric bender element with an 8-mm-diameter contactor was employed. This paradigm consisted of 12 cycles, each of 30-s duration, with an ON period of 6 s, during which the thumb was subjected to vibrotactile stimulation at a frequency of 35 Hz, with a peak-to-peak amplitude of 400 μm.

Six-parameter motion correction was applied using the AIR algorithm in MEDx (Sensor Systems, Sterling, VA, USA). The data were spatially smoothed with a Gaussian of 4.5 mm FWHM, and a high-pass filter was used to remove drift effects from the data (cutoff frequency = 0.006 Hz) while a notch filter (0.27–0.46 Hz) was used to remove respiration-induced fluctuations. We calculated activation maps from the resulting data sets in SPM99 using a model formed by convolving the paradigm time course with a canonical hemodynamic response function (HRF), with the motion parameters used as confounds. Activation maps were thresholded at a corrected probability of 0.005.

EVI With Prepulses Used for Spin Preparation

EVI data were acquired from human subjects, using two different prepulses to generate contrast.

First, T1-weighted IR images were acquired following each fMRI experiment for use in identifying cortical sulci. These images were produced by applying an 180° hyperbolic secant pulse to invert magnetization prior to the EVI sequence and using a recovery time, TI, of 1200 ms to null the signal from GM. IR-EVI data were acquired with a resolution of 3 × 3 × 3 mm3 over a 64 × 32 × 8 matrix.

Second, signal targeting with alternating radiofrequency labeling (STAR) (14, 15) was applied in conjunction with the EVI signal readout in order to produce perfusion-sensitive, 3D images. This involved preceding the EVI sequence with a slice-selective inversion pulse applied to a slice oriented parallel to the imaged slab, but displaced so as to overlie arteries carrying blood into the imaged region. By subtracting the resulting image data from control images acquired with the inverted slice positioned on the opposite side of the imaged region (so as not to contain in-flowing arterial blood), we produced perfusion-sensitive images (14). In the implementation described here, sagittal echo volumar images encompassing the left motor cortex were acquired with the magnetization inverted in a 50-mm-wide slice positioned to produce a 16-mm gap between the inverted and imaged regions. Inversion was accomplished using a hyperbolic secant pulse, which was applied 2200 ms before imaging. Labeled and control images were acquired alternately with a 3-s interexperimental delay. Vascular crushing, with a b-factor of 1.5 mm2s−1 and critical velocity of 27 mm s−1 (16), was used to attenuate the signal from blood flowing in the larger vessels of the vascular network within each voxel, and the magnetization within the imaged slab was saturated immediately prior to the inversion. Thirty labeled/control image pairs with 3-mm isotropic resolution were acquired using a matrix size of 64 × 32 × 8.

RESULTS

  1. Top of page
  2. Abstract
  3. THEORY
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

EVI Sequence Development and Assessment

Figure 3 shows data acquired with the original (a) and modified (b) EVI sequences from the phantom. In these images the read direction is horizontal and the slices are numbered such that slice 1 is the most lateral, and thus is closest to the surface coil, while slice 8 is the most medial. For comparison, Fig. 3c shows a magnetization-prepared rapid gradient-echo (MPRAGE) (17) image of a similar region of the gel phantom with 1.5-mm slice thickness and an FOV similar to that used for the EVI data.

thumbnail image

Figure 3. Echo volumar images of the gel phantom generated using (a) the conventional EVI sequence and (b) the modified sequence. For both a and b the image matrix size is 64 × 32 × 8, and the FOV is 192 × 96 × 12 mm3. The arrow points to a region of significant artifact in the image acquired using conventional EVI. c: For comparison, an MPRAGE image showing a similar region of the gel phantom is shown.

Download figure to PowerPoint

Example calibration scan data from a 12-mm-thick slab in the human brain are shown in Fig. 4. Figure 4a shows the evolution of the phase map obtained from each of the eight sets of 32 echoes obtained without application of the blipped z-gradient. The effective TE increases by 9.4 ms from one map to the next. A steady evolution of the phase accumulated due to the effects of magnetic field inhomogeneity can be seen in these data. Figure 4b shows the corresponding modulus images obtained from the 12-mm slab. As expected, signal dropout due to through-slab dephasing increases at the longer TEs. The result of correcting an EVI brain image data set using the reference scan method is shown in Fig. 5. Figure 5a shows data reconstructed without phase correction, while Fig. 5b shows the same data reconstructed using the phase information from a calibration scan.

thumbnail image

Figure 4. (a) Phase maps and (b) modulus images generated from a calibration scan acquired from the brain without the blipped z-gradient. Sagittal images were obtained from the Fourier transform of successive kx-ky planes acquired at 9.4-ms intervals. Each image has an FOV of 192 × 96 mm2 and a matrix size of 64 × 32, and was produced from the 12-mm-thick slab.

Download figure to PowerPoint

thumbnail image

Figure 5. EVI image data set (matrix = 64 × 32 × 8; FOV = 192 × 96 × 12 mm3) acquired from the left hemisphere of the brain (NEX = 8). a: Images reconstructed without using the calibration scan. b: The same data reconstructed after phase correction based on the calibration scan data.

Download figure to PowerPoint

Figure 6 shows a comparison of EVI data acquired from the phantom using the modified single-shot sequence and the two-shot interleaved approach shown in Fig. 2. It was found that halving the number of echoes acquired in each plane of k-space at fixed τ in the interleaved approach allowed the minimum TE to be reduced from 44 to 26 ms, and also increased the voxel bandwidth from 13 to 24 Hz. Consequently, the field inhomogeneity-induced image distortion and signal dropout are reduced in the interleaved data of Fig. 6b.

thumbnail image

Figure 6. A comparison of EVI data generated from the gel phantom using interleaved and noninterleaved acquisitions. a: Image data (matrix = 64 × 32 × 8; FOV = 192 × 96 × 12 mm3) acquired in a single shot using an echo train of 75-ms duration, and a 44-ms TE. b: Image data (matrix = 64 × 32 × 8; FOV = 192 × 96 × 12 mm3) acquired in two shots with an echo train of 42-ms duration, and a 26-ms TE, using the segmented acquisition scheme shown in Fig. 2a. It can be seen that the field inhomogeneity-induced distortions are less severe in the interleaved images. The distortion is most pronounced on the right-hand side of images 3–5, where the phantom contains an air bubble that results in large field inhomogeneity.

Download figure to PowerPoint

Applications of the Modified EVI Sequence

BOLD fMRI Using EVI

Activation in response to vibrotactile stimulation was found in both primary and secondary somatosensory cortex (SI and SII, respectively) in all data sets. Figure 7 shows a representative activation map overlaid on an IR-EVI image. To demonstrate the high temporal resolution achieved in this experiment, example average time courses from clusters in SI and SII are shown with no low-pass temporal filtering applied. For comparison, the canonical HRF from SPM99 is also shown. Both time courses correlate well with the shape of the HRF, though the post-stimulus undershoot appears more pronounced in the experimental data.

thumbnail image

Figure 7. a: Activation due to vibrotactile stimulation of the right thumb measured using EVI. Data were thresholded at a corrected probability of 0.005, and are shown overlaid on an IR-EVI image. b: Time courses of the responses in SI and SII to vibrotactile stimulation. The black and gray lines depict the average time course from 5 voxels in SI and 12 voxels in SII, respectively. Both time courses were averaged across the 10 cycles of the fMRI experiment. No low-pass temporal filtering was applied to the data, to highlight the high temporal resolution of EVI for functional studies. For comparison, the canonical HRF from SPM99 is shown as a smooth continuous line.

Download figure to PowerPoint

EVI With Prepulses Used for Spin Preparation

Figure 8a shows sagittal EVI data acquired using the sequence of Fig. 1b with 3-mm isotropic resolution. These images display Tmath image contrast resulting from the 44-ms TE employed. Figure 8b shows corresponding IR-EVI data acquired using a TI of 1200 ms. The resulting nulling of GM signal is evident, and leads to a clear depiction of the sulci and gyri. Figure 8c shows EVI-STAR perfusion-sensitive difference data formed by averaging and subtracting the 30 labeled/control image pairs. This is shown as a color map (thresholded at 0.7% signal change) that is overlaid on IR-EVI images acquired with GM nulled. As expected, high perfusion difference values are found in regions of GM.

thumbnail image

Figure 8. a: Five slices from EVI image data. To indicate the image orientation, superior (S), inferior (I), anterior (A), and posterior (P) labels are shown. b: The same five slices acquired using an IR EVI sequence with TI chosen so as to null the GM signal. The arrow shows the position of the central sulcus (CS). c: Perfusion-sensitive difference images that were acquired using EVI in conjunction with STAR with a delay time TI = 2200 ms. Areas of higher perfusion can be seen to overlie the GM areas evident in b.

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. THEORY
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

EVI offers an excellent approach to the simultaneous acquisition of multislice data, which allows very high volumar data acquisition rates to be achieved. However, the use of EVI has been limited as a consequence of its sensitivity to artifacts due to field-inhomogeneity-induced image distortion and Nyquist ghosting. These result from the long signal sampling times that are needed to generate a sufficient number of gradient echoes to span a 3D region of k-space in a single FID. The modifications to the EVI sequence described here, which include the use of rewind gradients, OVS techniques, and phase correction via a calibration scan, allow these artifacts to be greatly reduced. This is evident from Fig. 3, which shows images of the gel phantom that were obtained using the original and modified EVI sequences. A significant Nyquist ghost appears shifted by four slices in the data acquired using the original EVI sequence (Fig. 3a), but ghosting is not obviously evident in the images acquired using the modified sequence (Fig. 3b). Destructive interference between the ghost and the image is particularly clear in slice 6 of Fig. 3a. In acquiring these images, the thickness of the excited slab was made 3 mm smaller than the width of the image matrix in the z-direction, as can be seen from the absence of signal in slices 1 and 8 in Fig. 3b. Nyquist ghosting and signal shifting due to resonance offsets causes signal spillage into these slices in the data acquired using the conventional EVI sequence, which, in contrast to the data of Fig. 3b, were processed without phase correction using a reference scan. The reduced level of artifact achieved using the modified sequence made the process of image optimization at the scanner via adjustment of gradient timings more straightforward and less time-consuming. A slight disadvantage of the introduction of the rewind gradients is the resulting increase in duration of the echo train and consequent reduction in the frequency separation of points in the through-slab direction. In the experiments described here, the inclusion of rewind gradients increased the echo train length by 12.5%.

The reduction of distortion and ghosting that is achieved specifically by the use of the calibration-scan correction is evident from the data shown in Fig. 5. This correction should eliminate distortion due to spatially varying fields that do not change significantly across the slab thickness. The calibration-scan correction was facilitated by the shape of the image matrix used in our studies, in which the slab thickness was generally much smaller than the width of the FOV. It is likely that the calibration-scan approach to correcting EVI data can also be extended to include the acquisition of multiple data sets that are each subjected to a constant phase-encoding in the through-slab direction, as has been done in the case of EPI (18).

The use of segmented acquisition in conjunction with EVI has been successfully demonstrated, and in the implementation described here (in which k-space was spanned over two acquisitions while the time per echo was kept constant) was shown to provide a useful reduction in the minimum achievable TE and the field-inhomogeneity-induced distortion (Fig. 6). Reduced distortion would be particularly advantageous for studying brain areas with large magnetic field inhomogeneity, such as the orbitofrontal cortex. If the focus is not on reducing image distortion, interleaving can also be used to increase spatial resolution or the extent of the FOV in either of the phase-encoding directions while keeping the length of the echo train the same as in a single-shot acquisition. This involves increasing either the time per echo or the number of echoes acquired. In this study we evaluated only the segmented acquisition scheme shown in Fig. 2, in which alternate kx lines are acquired from separate excitations. A scheme in which alternate kz planes are acquired from separate excitations could also be employed, and would provide a greater reduction of the echo train length used in each acquisition, since the number of rewind gradient pulses would also be reduced. However, this would require the use of a larger ETS.

Despite the advantages of segmented acquisition, we did not use this approach in the EVI applications described in this paper, because of the factor of 2 loss in temporal resolution that results from interleaving the sequence. The use of parallel imaging (19, 20) in conjunction with EVI would allow the advantages of segmented acquisition to be realized without any penalty in temporal resolution. Unfortunately, the 3T scanner used in this work had only a single RF channel for reception, so it was not possible to employ parallel imaging. Successful combination of the modified sequence with parallel imaging is likely to have a strong impact on the future utility of EVI.

The modified EVI sequence was used to acquire fMRI data with a temporal resolution of 167 ms while maintaining a reasonable spatial resolution (voxel size of 3 × 3 × 1.5 mm3) over an FOV of 192 × 96 × 12 mm3 that spanned the somatosensory and motor cortex of one hemisphere. The timing of the EVI sequence implemented here would allow a minimum TR of about 85 ms; however, this was not experimentally feasible because of high acoustic noise and concerns about excessive heating of the gradient coils (which are not water-cooled) and the scanner's limited data acquisition rate. In order to achieve a temporal resolution of 167 ms with conventional multislice EPI over a similarly sized data matrix, each 2D image would have to be acquired in about 21 ms. This would preclude the use of a TE ∼ Tmath image, which is required for generation of optimal BOLD contrast. Neglecting the effect of different Tmath image weightings, the intrinsic SNR of images acquired using conventional EPI would be reduced by approximately the square root of the number of slices compared to the EVI data, as a result of the shorter acquisition window. However, the sensitivity to distortion in the EPI data would be significantly lower. Temporal and spatial resolution and BOLD sensitivity similar to those achieved with EVI could be achieved with the use of principles of echo-shifting with a train of observations (PRESTO) (21). Corresponding PRESTO data would also have a much lower sensitivity to distortion, which is a significant advantage, but would be more sensitive to motion artifacts and also display different contrast compared to that produced in most EPI-based fMRI experiments (21). This is a consequence of the short TR that is employed, and the consequent generation of steady-state transverse magnetization that persists in a strongly dephased form between successive excitations.

Increasing the temporal resolution with which fMRI data are acquired via the use of EVI could provide a number of advantages. First, a higher image acquisition rate leads to improved characterization of the hemodynamic response and reduces problems in analysis caused when spatially separate regions of the brain are sampled at different times. This would be particularly beneficial for event-related fMRI studies in which short interstimulus intervals (22) are used, or in which small differences in the timing of activation across trials have to be monitored (23). Second, increasing the rate of acquisition of images in fMRI experiments by reducing the TR can increase the efficiency with which activation is detected. Reducing TR leads to a decrease in SNR in each image due to saturation; however, the number of images acquired in a fixed experimental time increases, and with an optimal choice of flip angle, this leads to an overall increase in image SNR per unit time. Third, increasing the sampling rate in an fMRI study facilitates the detection and subsequent elimination of physiological noise due to cardiac and respiratory effects (24). Finally, increasing the volumar acquisition rate potentially allows more accurate correction of subject motion. When 3D data are acquired via phase-encoding across multiple excitations, movement during one volumar acquisition leads to blurring and formation of artifacts that cannot be corrected by simple motion-correction algorithms. In the case of multislice fMRI data, movement during the acquisition of a multislice set means that different slices are acquired with the head at different positions, and thus the resulting data will not conform to the model of rigid body motion that is often assumed in motion-correction algorithms.

In addition to the use of EVI for fMRI, we have shown that additional contrast can be obtained in EVI by the use of prepulses. We obtained EVI data with GM nulled by preceding the EVI acquisition with an IR, which allowed a good depiction of the sulcal anatomy. ASL via the STAR approach (14) was combined with an EVI readout to allow simultaneous acquisition of multislice, perfusion-sensitive images. This offers a potentially valuable new approach to quantitation of perfusion, which will be more fully described in a future publication.

CONCLUSIONS

  1. Top of page
  2. Abstract
  3. THEORY
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

We have shown that the use of rewind gradients, OVS, and calibration maps greatly improves image quality in EVI, and allows a temporal resolution of 167 ms at a spatial resolution of 3 × 3 mm in-plane and 1.5-mm slice thickness to be achieved. With these modifications and the future use of parallel imaging, the potential advantages of EVI as a method for generating fMRI data with high temporal resolution can be realized.

Acknowledgements

  1. Top of page
  2. Abstract
  3. THEORY
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

We thank Penny Gowland for useful contributions to this work. W. van der Zwaag received a PhD studentship from the Biotechnology and Biological Sciences Research Council and Unilever plc.

REFERENCES

  1. Top of page
  2. Abstract
  3. THEORY
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES