Single-shot T2* mapping with 3D compensation of ... - Semantic Scholar

Available online at www.sciencedirect.com R

NeuroImage 18 (2003) 390 – 400

www.elsevier.com/locate/ynimg

Single-shot T2* mapping with 3D compensation of local susceptibility gradients in multiple regions Stefan Posse,a,* Zhou Shen,a Valerij Kiselev,b and Lars J. Kemnac a

b

Department of Psychiatry and Behavioral Neuroscience, Wayne State University, Detroit, MI 48201, USA Section of Medical Physics, Department of Diagnostic Radiology, University Hospital Freiburg, Freiburg, Germany c Department of Diagnostic Radiology, University Hospital Freiburg, Freiburg, Germany Received 5 April 2002; revised 26 August 2002; accepted 13 September 2002

Abstract Macroscopic magnetic field inhomogeneities severely limit sensitivity of blood oxygenation level-dependent (BOLD) functional MRI (fMRI) in frontal and central brain regions close to brain stem. A single-shot multiecho echo-planar imaging method (TurboPEPSI) was developed that combines quantitative T2* mapping with gradient compensation of local susceptibility inhomogeneities in multiple volumes of interest (VOIs). Gradient compensation was optimized in individual subjects based on magnetic field mapping and applied at selected echo times, interleaved with acquisition of uncompensated echoes. Intrinsic T2* values from uncompensated echoes were obtained in real-time simultaneously with effective T2* values from gradient compensated echoes. It is demonstrated that up to three VOIs can be compensated in a single excitation, in addition to collecting uncompensated data, using 8-echo acquisition on a clinical 1.5 Tesla scanner. A theory was developed to optimize the sequence of uncompensated and compensated echoes to achieve maximum BOLD sensitivity. Gradient compensation increased effective T2* values in left and right amygdala on average by 18.8 ⫹/⫺ 7.5 ms, while maintaining sensitivity in uncompensated brain areas. In orbitofrontal cortex effective T2* values increased by 22.2 ⫹/⫺ 5.3 ms. A CO2 challenge paradigm was used to demonstrate that this gradient compensation method significantly enhances BOLD signal changes in amygdala as compared to conventional echo-planar imaging (EPI) and uncompensated TurboPEPSI. © 2003 Elsevier Science (USA). All rights reserved. Keywords: fMRI; BOLD; Susceptibility artifact; Sensitivity; T2*; Quantification

Introduction Susceptibility related signal losses due to air–tissue interfaces limit sensitivity of blood oxygenation level-dependent (BOLD) contrast functional MRI (fMRI) in frontal and central brain regions close to brain stem. Partial compensation of local susceptibility gradients in the slice direction has been achieved using thinner slices to reduce dephasing effects [1]. However, thinner slices reduce volume coverage and signal-to-noise ratio (SNR), thus elongating scan times. Techniques with gradient compensation in the slice direc* Corresponding author. Wayne State University School of Medicine, Department of Psychiatry and Behavioral Neurosciences, 4201 St. Antoine, University Health Center-9B-26, Detroit, MI 48201. Fax: ⫹1-313-5775900. E-mail address: [email protected] (S. Posse).

tion have been shown to be effective [2–7], but the spatial nonlinearity of local gradients requires multiple acquisitions with different compensation gradients, which may not be practical for certain applications, e.g., event-related fMRI. A different class of methods employs RF excitation with nonlinear phase profile to match the spatial nonlinearity of local gradients [8 –10]. Recently, these methods have been combined with gradient compensation [11]. Song developed a double-echo method with quadratic excitation to simultaneously acquire conventional and gradient compensated data sets [12], achieving a temporal resolution similar to that of single-echo fMRI. Consistent with our previous work on group spin echo selection [13], Deichmann et al. emphasized that local gradients along all three spatial axes, in particular the slice and phase encoding directions, can lead to changes in effective echo time (TE) (i.e., center of

1053-8119/03/$ – see front matter © 2003 Elsevier Science (USA). All rights reserved. PII: S 1 0 5 3 - 8 1 1 9 ( 0 2 ) 0 0 0 1 6 - 2

S. Posse et al. / NeuroImage 18 (2003) 390 – 400

k-space) in regions affected by local gradients and to local signal losses [14]. Yet another strategy to increase fMRI sensitivity in areas with susceptibility inhomogeneity is to use shorter TE to match reduced T2* values. We have previously noted that single-shot multiecho echo-planar imaging (EPI) (TurboPEPSI) allows acquisition of multiple images that are matched for a wide range of T2* values, thus optimizing overall BOLD sensitivity in the combined image [15]. Glover and Law have shown that combining a short TE acquisition using reverse spiral encoding with forward spiral at longer TE reduces susceptibility related signal losses in frontal lobes and increases functional sensitivity in an olfaction task [16]. In order to compare these methods, it is desirable to quantify the increase in BOLD sensitivity with susceptibility compensation. Multiecho fMRI allows direct quantification of T2* values, but combination of this technique with gradient compensation has not yet been reported. In this study we developed a multiecho EPI method based on our TurboPEPSI methodology [15] that combines quantitive T2* mapping with compensation of susceptibility related signal losses in multiple VOIs at different TEs, using VOI-specific compensation gradients along all three spatial axes. We specifically focus on the amygdala, which remains a challenge for fMRI of emotion, drug craving, and related paradigms [17]. Local gains in T2* values were quantified in several subjects and a CO2 challenge paradigm was used to assess local improvement in BOLD sensitivity. The CO2 challenge is superior to a specific task, as gray matter in the entire brain shows a homogeneous BOLD signal increase, facilitating comparisons. In addition, the signal increase is about 10% at 1.5 T compared to 0.5 to 3% in an activation task. Combination of compensated and uncompensated T2* maps, and methods to sum images obtained at different TEs, were investigated to maximize BOLD sensitivity.

Theory It is well known that the BOLD effect can be detected over a wide range of TEs around the optimal TE, which provides for considerable flexibility in applying compensation gradients at different echo times. The useful range of echo times can be estimated by subtracting the signal decays in the baseline and activated states. Several groups have shown that the resulting difference signal is bell-shaped and peaks at a time delay that is equal to T2* of tissue [15,18,19]. However, fMRI sensitivity may not decrease as rapidly beyond the peak as suggested by the difference signal, depending on the dominant noise source. In visual cortex, using EPI with 6-mm spatial resolution, we found that the spatial extent of activation plateaus between 60 and 110 ms, suggesting that in these data the instability in initial signal intensity may have been the dominant noise source, rather than thermal noise [20]. In addition, signal contributions from cerebrospinal fluid, which has long T2* values and surrounds large veins, may contribute to BOLD contrast

391

Fig. 1. Single-shot multiecho EPI (TurboPEPSI) pulse sequence with gradient compensation in multiple VOIs. VOI-0 (regions not affected by local gradients) is uncompensated. Compensation gradients and the corresponding rephasing gradients for three different VOIs are shown schematically using different shading for each gradient pair.

at long TE. Using TurboPEPSI and flicker-light stimulation, we found a similarly extended range of TE sensitivity in visual cortex and detected signal changes at TE as long as 213 ms [15]. Based on these data, adequate BOLD sensitivity can be obtained at TE as long as 140 ms at 1.5 T. Local gradients Gl,i in VOIi are compensated at the center of a selected k-space trajectory, taking into account compensating gradients Gc that were applied to previous EPI modules (Fig. 1): 3 G l,iⴱTEn 3 G c共TEn兲 ⫽ ⫺ ⫺ ⌬t

冘3G 共TE 兲, c

j

(1)

j⬍n

where ⌬t is the effective duration of the compensation gradient. Compensation gradients are applied between individual EPI readout modules and the corresponding rephasing gradients are applied immediately after encoding each compensated image. These rephasing gradients are combined with the compensation gradient for the subsequent EPI module. The choice of compensation gradients at a given TE depends on the total number of TEs and the number of VOIs that need to be compensated. In order to quantify T2*, assuming a monoexponential decay, it is necessary to collect at least two echo times, preferentially using a short TE and a TE in the vicinity of T2*. For gray matter in areas distant from air–tissue interfaces the second echo time should be on the order of 70 ms at 1.5 T, while CSF containing spaces are best measured with the longest echo time available. For our 8-echo sequence these were echoes 1, 4, and 8. Collecting these three echoes without gradient compensation provides reference data with optimum sensitivity for brain areas that are far from air-tissue interfaces. Collecting the shortest TE without gradient compensation is advantageous for movement correction [21]. Pairs of the remaining echoes can be used to compensate local gradients in selected VOIs to achieve simultaneous susceptibility compensation in multiple VOIs. These echo pairs should have similar echo spacing to assure robust fitting. As gradient compensation is less

392


efficient in areas with spatially nonlinear gradients, it is advantageous to compensate such areas using echo pairs with short TEs, while regions with more linear local gradients can be compensated using echo pairs with longer TEs. For our sequence, echoes 2 and 5 were compensated for VOI1, and echoes 3 and 6 were compensated for VOI2. The remaining echo 7 was collected uncompensated. We labeled this compensation scheme “01201200” method, where 0 stands for uncompensated, 1 for VOI 1 compensated, and 2 for VOI 2 compensated. We also investigated the 00120120 scheme. To display the increase in T2* with compensation we combined the T2* maps from uncompensated and compensated data by selecting the maximum from these maps in each voxel. This yields a T2* map with optimal global BOLD sensitivity. Alternatively, we used weighted summation of echo images using the expected BOLD contrast as weights, which is computationally efficient, to increase BOLD sensitivity as compared to single echo data [15]. Many different schemes can be designed, depending on the number of VOIs that need to be compensated and depending on the nonlinearity of the local gradients. Even more regions can be compensated when sacrificing the ability to fit T2*. For example, compensation of a single and three VOIs was investigated using the 00101010 and 00102030 schemes. To characterize the sensitivity gain more rigorously we designed a numerical optimization scheme. Optimizing contrast-to-noise ratio with weighted summation for a given gradient compensation scheme Let us denote the signal for the specific echo times as s1, s2, . . .sn. . .sM. Here the index n ⫽ 1,2. . .M counts the echoes, and M is the total number of echoes. Let us denote the echo times as tn. Each signal sn is acquired with a given compensation gradient. The compensation scheme is currently considered as predefined. The problem is to find the optimal weighting to combine the multiecho image data into a single image. To solve it, we first calculate the contrastto-noise ratio Q for a given weighting and then find the optimal weighting that maximizes the value of Q. The weighting is defined by the weights wn, such that the summed signal takes the form

冘w s . M

S⫽

(2)

n n

n⫽1

Let us denote the BOLD-contrast at different echo times as ⌬sn. These values can be found by differentiation of the signal for small signal changes: ⌬sn ⫽ ⫺⌬R2tnsn, where ⌬R2 is the variation in the relaxation rate due to the BOLD effect. The contrast C of the summed signal takes the form

冘w ⌬s . n

n⫽1

n

The above signal is considered as noise free. The noise components are added explicitly for the present analysis, such that the observed signal for a given echo time takes the form sn ⫹ gn. Here gn are the noise components at the nth echo time. These values are subject to statistical averaging. The assumed statistical properties are the zero mean of gn, the statistical independence of gn for different echo times, and their echo time independent standard deviation ␴. The latter is an assumption, which has previously been validated in vivo [22]. The additive noise N in the summed signal is the linear combination of the noise at each echo time:

冘w g . M

N⫽

(4)

n n

n⫽1

Here both N and gn are specific noise realizations. In order to estimate the signal-to-noise ratio in the summed signal, we have to perform averaging denoted hereafter with the angular brackets. It is clear that N has zero mean: 具N典 ⫽ 0. The square of its standard deviation takes the following form under the above assumptions about the noise:

冘w 具g 典 ⫽ ␴ 冘w . M

具N 2典 ⫽

M

2

n

(3)

2

2

2

n

n

n⫽1

(5)

n⫽1

Now we can write the contrast-to-noise ratio Q as Q⫽

C

冑具N 典 2

⫽

1 ␴

冘 w ⌬s . 冑冘 w n

n

n

2

n

(6)

n

Let us find the set wn that maximizes this quantity. This can be easily done taking into account the geometrical meaning of the above expression. Consider both wn and ⌬sn as components of M-dimensional vectors w and ⌬s. Then Q is proportional to the scalar product of the unitary vector in the direction of w with the fixed vector ⌬s (Fig. 2). In turn, it is proportional to the cosine of the angle between these two vectors. The maximum is reached when vector w is parallel to ⌬s, which implies that w ⫽ Const · ⌬s n.

M

C⫽

Fig. 2. An illustration showing the maximization of the contrast-to-noise ratio Q. Its value is proportional to the scalar product of a constant vector ⌬s with a variable unit vector w/w. The maximum is reached when both vectors are parallel.

(7)

The value of Const does not matter, as the weights are normalized. The maximal Q takes the form


1 Q max ⫽ 冑 ␴

冘 ⌬s n

冑冘 M

2 n

⫽

2

Qn ,

(8)

n⫽1

where in the last term we introduce Qn ⫽ ⌬sn/␴, which is the contrast-to-noise ratio for the individual echo times. This form of Q guarantees that its value is larger than any individual Qn. The individual Qn can be analyzed in the form Qn ⫽ ⫺

⌬R 2 ␴

冑

slightly less sensitive than the 01201200 scheme. The 01201200 and 00120120 schemes were chosen as a compromise with adequate echo spacing to ensure reliable fitting of T2*.

Materials and methods Subjects and data acquisition

⌬R 2t ns n . ␴

(9)

Here the denominator ␴ is a given constant; the change in the relaxation rate ⌬R2 does not depend on the echo time, it rather characterizes the subject response. The other two factors in the nominator tnsn form a bell-shaped curve, which depends on the gradient compensation scheme. It is convenient for the further optimization to rewrite Qmax in the following form: Q max ⫽

393

冘共t s 兲 . M

2

n n

(10)

n⫽1

Here the prefactor should be considered as a given constant, while the baseline signals sn are subjected to optimization of the gradient compensation scheme. Optimization of the gradient compensation scheme A numerical simulation for the 8-echo TurboPEPSI sequence (see Materials and methods) was performed to determine the optimal sequence of uncompensated and compensated echoes. We assumed a baseline T2* value of 70 ms in brain areas that are not affected by susceptibility artifacts (VOI-0) and computed the weights accordingly. Either one or two target VOIs (VOI-1 and VOI-2) with susceptibility reduced T2* values of 20, 30, 40, or 50 ms were modeled. Activation-induced changes in ⌬T2* were modeled to be 5, 10, 15, 20, 25, or 30 ms. We then computed the following Q values: Q0 for the 00000000 scheme in VOI-0, Q1 for the compensated scheme in VOI-0 with the constraint Q1 ⬎ 0.6 ⴱ Q0, and Q2 (Q3) for the compensated method in VOI-1 (VOI-2 with the constraint 0.5 ⬍ Q2/Q3 ⬍ 2). We also required that the first echo be uncompensated. For each parameter combination, the compensation methods with maximum Q2 (⫹Q3) were tabulated. In addition, the corresponding Q values for the 01201200 and 00120120 schemes were tabulated for each parameter combination. The simulation confirmed that compensation using early echoes yields the highest sensitivity. Depending on the parameter combination, the following schemes were found to be optimal: 01110000, 01111000, 01110100, etc. in the case of a single target VOI and 01221000, 02112000, 02121000, 01212000, 01122000, 02211000, 01120200, 02210100, etc. in the case of two target VOIs. Overall, the 01201200 scheme has similar Q2 ⫹ Q3 values as compared to the optimized methods. The 00120120 scheme is on average

Fourteen healthy volunteers gave institution-approved informed consent to participate in this study. Functional MR imaging was performed on a Vision 1.5T clinical scanner (Siemens Medical Systems, Erlangen, Germany) equipped with gradient booster and quadrature head coil. The broadband amplifier was used, because it has smaller gain instability than the standard tube amplifier. Double hearing protection (ear plugs and headphones) was provided. Foam padding was used to minimize motion artifacts. Minimization of head movement artifacts during the CO2 challenge paradigm was achieved using a customfitted head holder consisting of 2-component polyurethane foam (KGF Enterprise, Chesterfield, MI). The two components were mixed and filled into a double layer plastic bag, which was wrapped around the subject head inside the head coil. The plastic bag was placed inside a custom designed head coil liner to prevent foam buildup which would obstruct the head coil and removal of the head holder. The expanding foam solidified within a few minutes during which time it could be molded tightly around the entire head and neck, leaving only the face uncovered. All subjects tolerated this procedure without discomfort. Higher order shimming up to third order, which is part of the scanner software, was performed automatically using 3D B0 field mapping. High-resolution T1-weighted MPRAGE images were acquired to localize VOIs. TurboPEPSI [15] was performed with eight EPI encoding modules using equidistant TEs ranging from 12 to 140 ms (read-out window duration: 16 ms, interimage spacing: 18.3 ms, matrix size: 32 ⫻ 32 pixels). Sixteen slices of 7-mm thickness with 10% interslice spacing were typically measured using pixel size of 6.25 ⫻ 6.25 mm2 and field of view (FOV) of 200 mm. In some experiments, a slice thickness of 3 mm was used. The duration of the trapezoidal readout gradients was 500 ␮s with 160 ␮s ramp times. Nonlinear sampling was employed on the gradient ramps. Phase encoding was refocused between images to encode the same k-space trajectory in all images. Data acquisition time for 16 slices was 2.7 s. Susceptibility compensation gradients along all three spatial axes were inserted between EPI modules. VOIs with low T2* values were defined on uncompensated real-time T2* maps computed with the NumART2* algorithm (see below). VOI dimensions were typically 25 ⫻ 25 ⫻ 15 mm3. Local linear gradient vectors in selected VOIs were calculated based on 3D B0 field mapping, which is part of the

394


scanner software. Because the PARGEN pulse-programming software does not allow formula-based calculations, a custom software interface was developed to read output from 3D field mapping, compute compensation gradient amplitudes, edit the pulse sequence text file, and recompile the pulse sequence. Single-voxel water spectroscopy with manual gradient shimming was used in some experiments to confirm results from 3D B0 field mapping. Calibration of compensation gradients was verified on a spherical water phantom at long echo times using gradient offsets. Three subjects underwent a CO2 challenge that increases signal in all cortical areas. Details of the procedure have been described previously [23,24]. Briefly, a mixture of 10 –15% CO2 and room air was delivered at a rate of 15 ltr/min and alternated with room air delivered at 15 ltr/min. CO2 values were maintained at 28 ⫹/⫺ 5 mm Hg during baseline condition and reached 48 ⫹/⫺ 5 mm Hg during hypercapnic condition. The paradigm was: 52 s baseline, 60 s hypercapnia, 100 s baseline, 60 s hypercapnia, 60 s baseline. Conventional EPI (TE: 60 ms) and TurboPEPSI scans with and without 01201200 compensation scheme of both amygdalae were acquired using TR: 4 s and identical image resolution (6.25 ⫻ 6.25 ⫻ 7.7 mm3).

Offline data analysis Results from real-time fMRI analysis were confirmed using statistical parametric mapping SPM99 (29). Signal changes in amygdala during the CO2 challenge paradigm were quantified based on the following measurements after removal of baseline drifts: mean signal intensity in amygdala (Sa) and in a reference region (medial grey matter above the ventricles) (Sr), and % signal changes in amygdala (sa) and in the reference (sr). The normalized signal change in the amygdala ⌬S ⌬S ⫽

absolute signal change in amygdala absolute signal change in reference region (12)

was computed as ⌬S ⫽

s aS a . sr Sr

(13)

Multiecho multislice images were processed offline using custom-designed C programs. Nonequidistant least-squares exponential fitting of all compensated or uncompensated echoes was employed.

Real-time image reconstruction and data analysis

Results

Image reconstruction and computation of multiecho images in “Mosaic” format, T2* maps or combined images using weighted summation was performed on the scanner in real-time during the ongoing scan. T2* maps without gradient compensation were computed using NumART2* algorithm [25]. T2* maps with 01201200 gradient compensation were computed using linear regression of the logarithm of the first two echo images for each VOI (using more echo times was not feasible due to software limitations), and by taking the local maximum of the three T2* maps. For functional studies using CO2 challenge, images (e1– e8) acquired with the 01201200 scheme were combined in the echo-time domain

Fig. 3 shows typical signal decay in the entire head, using an uncompensated TurboPEPSI sequence. Significant signal recovery in the left and right amygdala, even at long TE, was obtained with 00120120 and 01201200 gradient compensation schemes (Fig. 4). Even at echo times as long as 120 ms, strong signal was obtained while uncompensated echoes retained sensitivity in other brain areas. Gradient compensation was validated with single voxel spectroscopy using manual gradient shimming, which resulted in typical water linewidth of 15 Hz FWHM (in magnitude mode). Baseline T2* values in the amygdala ranged from 35 to 71 ms in different subjects. With 01201200 gradient compensation the effective T2* values in amygdala in combined T2* maps increased in all subjects (mean: 18.8 ms, SD: 7.5 ms) (Fig. 5, Table 1). There was no significant change in T2* values in other brain areas. Baseline T2* values in orbitofrontal areas ranged from 25.7 to 34.7 ms in different subjects. Gradient compensation using the 00120120 scheme increased local effective T2* values between 14 and 30.4 ms (mean: 22.2 ms, SD: 5.3 ms) (Fig. 6, Table 2). There was no significant change in T2* values in other brain areas. Similar gains were obtained in temporal areas. Compensation of a single VOI (orbitofrontal cortex) and three VOIs (orbitofrontal and bilateral temporal lobes) was demonstrated using 00101010 and 00102030 schemes. Baseline signal intensity and signal changes due to CO2 challenge in amygdala and adjacent areas were enhanced with TurboPEPSI as compared to conventional EPI, result-

I ⫽ 0.5 ⴱ e1 ⫹ max (e2,e3,e4) ⫹ 0.8 ⴱ max (e5,e6,e7) ⫹ 0.5 ⴱ e8.

(11)

using weighted averaging, taking into account the bellshaped BOLD sensitivity curve for T2* ⫽ 50 ms, which was chosen as a compromise between different brain regions. Real-time fMRI analysis using Functional Imaging in REal time (FIRE) software package was performed as described previously [26 –28] to quantify BOLD contrast with CO2 challenge paradigm. Increase in effective T2* in combined maps was quantified by region-of-interest (ROI) analysis in orbitofrontal cortex, in temporal lobes, and in left and right amygdala.


395

Fig. 3. Uncompensated multiecho images at TE ranging from 12 to 140 ms. Rapid signal decay is seen in the amygdalae.

ing in increased mean correlation coefficient. Gradient compensation in both amygdalae further increased BOLD contrast in these and adjacent areas as compared to uncompensated TurboPEPSI and conventional EPI, in ac-

cordance with locally increased effective T2* values, while maintaining BOLD sensitivity in the rest of the brain (Fig. 7, Table 3). Experiments on spherical phantoms with long T2 values

396


Fig. 4. Susceptibility compensation (00120120 scheme) of both amygdalae in images 3 (left), 4 (right), 6 (left), 7 (right). Even at echo times as long as 122 ms, significant signal intensity remains in the amygdalae (see crosshair). Uncompensated images (1,2,5,8) are equivalent to those in Fig. 1.


Fig. 5. Axial T2* maps obtained without (left, intrinsic T2* in amygdalae: 33 ms) and with (right, compensated T2* in amygdalae: 55 ms) gradient compensation. Circles indicate areas for quantification of T2* increase.

using isotropic voxel size showed that, as expected, all spatial axes, including the read-out direction, were highly sensitive to shim gradient offsets, which resulted in complete signal loss beyond a TE threshold that was dependent on gradient offset. Gradient compensation resulted in complete signal recovery even at the longest TE used.

Discussion In this study we have shown that gradient compensation can be integrated into multiecho EPI to reduce local susceptibility induced signal losses in multiple VOIs in a single shot. Using 8-echo TurboPEPSI on a clinical 1.5 Tesla scanner with minimum TE of 12 ms, compensation of nonlinear local gradients in amygdala, frontal, and temporal brain areas was demonstrated at echo times up to 122 ms while sensitivity in other brain regions was preserved. Increased BOLD contrast in amygdala was demonstrated using a CO2 challenge. As expected, the efficiency of the compensation method varies between subjects, which is most likely due to interindividual differences in head and brain shape. Quantification of T2* allows us to predict sensitivity gains in individual subjects and to compare the efficiency of different compensation schemes. Gradient compensation is limited by the spatial nonlinearity of local gradients. Compensation in small VOIs is thus advantageous. Maximizing the number of echoes provides enhanced flexibility for compensating local gradients in multiple small and adjacent VOIs, thus increasing sensitivity in a larger total VOI. The choice of TEs to compensate a given VOI depends on the nonlinearity of local gradients. In areas with highly nonlinear local gradients, it is advantageous to apply compensation gradients at short TEs, and to compensate areas with more linear local gradients at longer TEs. Maximizing the number of echoes is also advantageous for quantifying possible multiexponential decay. In addition, combination of data in the echo time domain is an alternative to time-consuming fitting, which is suitable for realtime fMRI and computationally robust. However, the present methodology has certain inherent

397

limitations: in-plane spatial resolution is relatively low to accommodate a large number of echoes. Reducing voxel size would be advantageous to compensate for spatial nonlinearity of local gradients and thus increase efficiency of gradient compensation. The decreased spatial resolution may lead to partial volume effects in small structures, such as the amygdale. To overcome this limitation, we have previously implemented 64 ⫻ 32 matrix size with similar echo spacing [15]. Preliminary experiments using a 64 ⫻ 64 matrix indicate that a 1200 compensation scheme with four echoes (TEmin ⫽ 25.1, ⌬TE ⫽ 40.8 ms) enhances signal intensity bilaterally in the amygdala (results not shown). However, that compensation scheme does not allow quantification of T2* changes. Using high performance cardiac or head gradients it is possible to acquire a 64 ⫻ 64 matrix with much reduced EPI encoding time, without inducing physiological stimulation. Half-Fourier k-space encoding allows increased spatial resolution, but inherent asymmetry of gradient echo k space [13] may introduce local bias in BOLD contrast toward a particular local gradient orientation, which cannot be quantified easily. Parallel imaging using multiple receiver coils, such as SENSE, is a new technology, which allows decreasing EPI encoding time by factors of 2–3 [30]. Combination of this approach with multiecho EPI will significantly enhance spatial resolution and the performance of gradient compensation in multiple VOIs. On the other hand, fMRI sensitivity decreases with decreasing voxel size [28,31], which may limit applications to single trial and event related paradigms. Multiecho EPI with gradient compensation is suitable for applications at high field, but decrease of T2* values with field strength limits the range of echoes that can be collected in a single shot. Increases in gradient performance with field strength are necessary to minimize encoding times. Preliminary experiments on our experimental 4T scanner, which is equipped with Sonata cardiac gradients, indicate feasibility of multiecho EPI with 01201200 gradient compensation: It is possible to collect eight echoes with a 32 ⫻ 32 matrix and echo spacing of 11.6 ms in about 91 ms, which is about

Table 1 Increase in T2* value with gradient compensation in amygdala Subject

T2* [ms] Left

1 2 3 4 5 6 7 8 Mean (SD)

Right

Baseline

Increase

Baseline

Increase

44.1 50.1 55.9 71.3 57.2 34.9 39.0 36.3

12.8 16.4 14.0 14.1 19.3 16.8 9.7 24.5

46.6 65.3 54.8 65.4 53.6 45.5 57.6 36.0

18.3 11.3 31.2 15.5 15.1 21.9 19.7 39.5

16.0 (4.5)

18.8 (7.7)

398


Fig. 6. Typical examples of T2* maps without (a) and with (b) gradient compensation in medial prefrontal cortex. The difference image (c) highlights the area of increased T2* values.

2.5 ⴱ T2* at that field strength and thus roughly consistent with our imaging parameters at 1.5 T. A 64 ⫻ 64 matrix can be acquired with four echoes, TEmin ⫽ 18 ms, and ⌬TE ⫽ 30 ms. As parallel imaging becomes available at high field strength the restrictions on spatial resolution and number of

Table 2 Increase in T2* value with gradient compensation in orbitofrontal cortex Subject

9 10 11 12 13 14 Mean (SD)

T2* [ms] Baseline

Increase

33.3 28.6 34.7 25.7 26.4 32.2

14.0 30.4 21.8 23.7 22.3 21.0 22.2 (5.3)

echoes will be further alleviated. As with all multiecho techniques, temporal resolution and/or volume coverage is reduced as compared to conventional single-echo techniques. In this study, using low spatial resolution, we achieved spatial coverage of the entire brain in 2.7 s, which is adequate for many fMRI applications. Further improvement in BOLD sensitivity can be achieved in several ways: optimizing compensation gradients for each slice separately to minimize sensitivity loss in superior slices, combining gradient compensation with spatially tailored RF excitation using nonlinear phase profile [8 –10], and slice-dependent optimization to minimize loss of sensitivity in areas with more linear local gradients. Even higher flexibility may be gained from higher-order shim gradients (e.g., Z2, ZY) that can be switched simultaneously with imaging gradients. In summary, the methodology presented here is advantageous for event-related fMRI in large brain regions that include multiple areas with local susceptibility gradients,


399

Fig. 7. Real-time fMRI of BOLD contrast response to a CO2 challenge. (a) Conventional EPI displays reduced BOLD contrast in lower slices. (b) TurboPEPSI with gradient compensation in both amygdalae using a 01201200 scheme and weighted echo averaging strongly enhances BOLD contrast in these and adjacent regions (slices 12–15), without compromising BOLD sensitivity in the rest of the brain.

such as amygdala and orbitofrontal cortex. Using a large number of echoes increases flexibility as compared to previous methodology and allows quantification of localized sensitivity gains in terms of T2*. The optimization of the compensation scheme allows a controlled tradeoff between sensitivity gains in areas with local susceptibility gradients and moderate sensitivity decreases due to the compensation gradients in brain areas that are unaffected by local gradients. Because multiecho EPI has intrinsically higher BOLD sensitivity than conventional EPI, this tradeoff is tolerable. A current limitation due to the large number of echoes is limited spatial resolution. Promising strategies for increasing spatial resolution include parallel imaging [30] and the use of higher performance gradients. Despite the current limitations, this methodology is advantageous for eventrelated neuroimaging studies of amygdala, which remains a

challenge with conventional methods. The results of several neuroimaging studies focusing on the amygdala have recently been questioned and it has been suggested that fMRI of the amygdale requires decreasing the voxel size to 4 – 8 ␮l or less to overcome susceptibility related signal losses [17]. However, the resulting strong loss in SNR makes that approach unsuitable for event-related studies. Using gradient compensation in left and right amygdala we have recently demonstrated amygdala activation during single 30-s periods of sad mood induction [32]. The orbitofrontal and temporal cortex, the rostral part of the cingulate, and the amygdala play a major role in emotional networks [33]. A technique that allows imaging of all these regions with minimum susceptibility artifacts has the potential to greatly improve our understanding of emotional processes in the human brain.

Table 3 Normalized signal change in amygdala (average of left and right side) during CO2 challenge with different imaging methods

Acknowledgments

Subject 6 Subject 7 Subject 8

EPI

TurboPEPSI (uncompensated)

TurboPEPSI (compensated)

0.65 0.81 1.16

0.80 0.91 1.32

1.32 1.11 1.70

Note. Corresponding increases in effective T2* with gradient compensation are listed in Table 1.

We thank Zahid Latif for designing the foam head holder and for help with data acquisition. Mae Nordin, Tyler Cederlind, and Mark Greenwald helped with data collection. We thank Mary Jacintha for help with data analysis. Peter Stanchev suggested the nonequidistant exponential fit. This research was supported by a grant of the State of Michigan (Joe Young, Sr. Foundation).

400


References [1] Young, I.R., Cox, I.J., Bryant, D.J., Bydder, G.M., 1988. The benefits of increasing spatial resolution as a means of reducing artifacts due to field inhomogeneities. Magn. Reson. Imaging 6, 585–590. [2] Frahm, J., Merboldt, K.D., Hanicke, W., 1988. Direct FLASH MR imaging of magnetic field inhomogeneities by gradient compensation. Magn. Reson. Med. 6, 474 – 480. [3] Ordidge, R.J., Gorell, J.M., Deniau, J.C., Knight, R.A., Helpern, J.A., 1994. Assessment of relative brain iron concentrations using T2weighted and T2*-weighted MRI at 3 Tesla. Magn. Reson. Med. 32, 335–341. [4] Constable, R.T., 1995. Functional MR imaging using gradient-echo echo-planar imaging in the presence of large static field inhomogeneities. J. Magn. Reson. Imaging 5, 746 –752. [5] Yang, Q.X., Dardzinski, B.J., Li, S., Eslinger, P.J., Smith, M.B., 1997. Multi-gradient echo with susceptibility compensation (MGESIC): demonstration of fMRI in the olfactory cortex at 3T. Magn. Reson. Med. 37, 331–335. [6] Constable, R.T., Spencer, D.D., 1999. Composite image formation in z-shimmed functional MR imaging. Magn. Reson. Med. 42, 110 – 117. [7] Stenger, V.A., Boada, F.E., Noll, D.C., 1999. Gradient compensation method for the reduction of susceptibility artifacts for spiral fMRI data acquisition, in: Proceedings of the 7th Annual Meeting of ISMRM, Philadelphia. p. 538. [8] Cho, Z.M., Ro, Y.M., 1992. Reduction of susceptibility artifact in gradient-echo imaging. Magn. Reson. Med. 23, 193–196. [9] Glover, G., Lai, S., 1998. Reduction of susceptibility effects in BOLD fMRI using tailored RF pulses, in: Proceedings of the 6th Annual Meeting of ISMRM, Sydney, Australia. p. 298. [10] Chen, N.K., Wyrwicz, A.M., 1999. Removal of intravoxel dephasing artifact in gradient-echo images using a field-map based RF refocusing technique. Magn. Reson. Med. 42, 807– 812. [11] Mao, J., Song, A.W., 1999. Intravoxel rephasing of spins dephased by susceptibility effect for EPI sequences, in: Proceedings of the 7th Annual Meeting of ISMRM, Philadelphia. p. 1982. [12] Song, A.W., 2001. Single-shot EPI with signal recovery from the susceptibility-induced losses. Magn. Reson. Med. 46, 407– 411. [13] Posse, S., 1992. Direct imaging of magnetic field gradients by group spin echo selection. Magn. Reson. Med. 25, 12–29. [14] Deichmann, R., Josephs, O., Hutton, C., Corfield, D.R., Turner, R., 2002. Compensation of susceptibility-induced BOLD sensitivity losses in echo-planar fMRI imaging. NeuroImage 15, 120 –135. [15] Posse, S., Wiese, S., Gembris, D., Mathiak, K., Kessler, C., GrosseRuyken, M.L., Elghawaghi, B., Richards, T., Dager, S.R., Kiselev, V.G., 1999. Enhancement of BOLD-contrast sensitivity by singleshot multi-echo functional MR imaging. Magn. Reson. Med. 42 (1), 87–97. [16] Glover, G.H., Law, C.S., 2001. Spiral-in/out BOLD fMRI for increased SNR and reduced susceptibility artifacts. Magn. Reson. Med. 46, 515–522. [17] Merboldt, K.D., Fransson, P., Bruhn, H., Frahm, J., 2001. Comments and controversies: functional MRI of the human amygdala? NeuroImage 14, 253–257.

[18] Menon, R.S., Ogawa, S., Tank, D.W., Ugurbil, K., 1993. 4 Tesla gradient recalled echo characteristics of photic stimulation-induced signal changes in the human primary visual cortex. Magn. Reson. Med. 30, 380 –386. [19] Bandettini, P.A., Wong, E.C., Jesmanowicz, A., Hinks, R.S., Hyde, J.S., 1994. Spin-echo and gradient-echo EPI of human brain activation using BOLD contrast: a comparative study at 1.5T. NMR Biomed. 7, 12–20. [20] Jonsson, T., Wennerberg, A., Forssberg, H., Glover, G.H., Li, T.Q., 1999. An image registration strategy for multi-echo fMRI. J. Magn. Reson. Imaging 10, 154 –158. [21] Wiese, S., Grosse-Ryken, M.L., Kiselev, V.G., and Posse, S., 1999. Mismatch between T2* and echo time dependence of BOLD contrast fMRI in men and women, in: Proceedings of the 7th Annual Meeting of ISMRM, Philadelphia. p. 176. [22] Kiselev, V.G., Wiese, S., Posse, S., 1999. Distinction of activation and noise in fMRI by multi-echo sampling, in: Proceedings of the 7th Annual Meeting of ISMRM, Philadelphia. p. 541. [23] Posse, S., Kemna, L.J., Wiese, S., Elghahwagi, B., Kiselev, V.G., 2001. Dependence of functional MR imaging contrast on graded hypo- and hypercapnia using whole brain T2*-mapping, Magn. Reson. Med. 46, 264 –271. [24] Kemna, L.J., Posse, S., Tellmann, L., Schmitz, T., Herzog, H., 2001. Interdependence of regional and global cerebral blood flow during visual stimulation: an O-15 butanol positron emission tomography study. J. Cereb. Blood Flow Metab. 21, 664 – 670. [25] Hagberg, G.E., Indovina, I., Sanes, J.N., Posse, S., 2002. Real time quantification of T2* changes using multi-echo planar imaging and NumART2*. Magn. Reson. Med. 48, 877– 882. [26] Gembris, D., Taylor, J.G., Schor, S., Frings, W., Suter, D., Posse, S., 2000. Functional MR imaging in real-time using a sliding-window correlation technique. Magn. Reson. Med. 43, 259 –268. [27] Mathiak, K., Posse, S., 2001. Evaluation of motion and realignment for functional magnetic resonance imaging in realtime. Magn. Reson. Med. 45 (1), 167–171. [28] Posse, S., Binkofski, F., Schneider, F., Gembris, D., Frings, W., Habel, U., Salloum, J.B., Mathiak, K., Wiese, S., Kiselev, V., Graf, T., Elghawaghi, B., Eickermann, T., 2001. A new approach to measure single event related brain activity using real-time fMRI: feasibility of sensory, motor, and higher cognitive tasks. Hum. Brain Mapp. 12 (1), 25– 41. [29] Friston, K.J., Holmes, A.P., Worsley, K.J., Poline, J.P., Frith, C.D., Frackowiak, R.J.S., 1995. Statistical parametric maps in functional imaging: a general linear approach. Hum. Brain Mapp. 2, 189 –210. [30] Pruessmann, K.P., Weiger, M., Scheidegger, M.B., Boesiger, P., 1999. SENSE: sensitivity encoding for fast MRI. Magn. Reson. Med. 42 (5), 952–962. [31] Yoo, S.S., Guttmann, C.R.G., Panych, L.P., 2001. Multiresolution data acquisition and detection in functional MRI. NeuroImage 14, 1476 –1485. [32] Posse, S., Fitzgerald, D., Habel, U., Rosenberg, D., Moore, G.J., Schneider, F., Real-time fMRI of temporo-limbic regions detects amygdala activation during single trial self-induced sadness. NeuroImage, in press. [33] Davis, M., Whalen, P.J., 2001. The amygdala: vigilance and emotion. Mol. Psychiatr. 6, 13–34.