Strategies for improving the detection of fMRI activation in trigeminal pathways with cardiac gating

www.elsevier.com/locate/ynimg NeuroImage 31 (2006) 1506 – 1512

Strategies for improving the detection of fMRI activation in trigeminal pathways with cardiac gating Wei-Ting Zhang,* Caterina Mainero, Ashok Kumar, Christopher J. Wiggins, Thomas Benner, Patrick L. Purdon, Divya S. Bolar, Kenneth K. Kwong, and A. Gregory Sorensen Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, and Harvard Medical School, Bldg 149 (2301), 13th Street, Charlestown, MA 02129, USA Received 21 June 2005; revised 14 December 2005; accepted 20 February 2006 Available online 19 April 2006 Functional magnetic resonance imaging (fMRI) has become a powerful tool for studying the normal and diseased human brain. The application of fMRI in detecting neuronal signals in the trigeminal system, however, has been hindered by low detection sensitivity due to activation artifacts caused by cardiac pulse-induced brain and brainstem movement. A variety of cardiac gating techniques have been proposed to overcome this issue, typically by phase locking the sampling to a particular time point during each cardiac cycle. We sought to compare different cardiac gating strategies for trigeminal system fMRI. In the present study, we used tactile stimuli to elicit brainstem and thalamus activation and compared the fMRI results obtained without cardiac gating and with three different cardiac gating strategies: single-echo with TR of 3 or 9 heartbeats (HBs) and dualecho T2*-mapping EPI (TR = 2 HBs, TE = 21/55 ms). The dual-echo T2* mapping and the single-echo with TR of 2 and 3 HBs cardiac-gated fMRI techniques both increased detection rate of fMRI activation in brainstem. Activation in the brainstem and the thalamus was best detected by cardiac-gated dual-echo EPI. D 2006 Elsevier Inc. All rights reserved. Keywords: fMRI; Trigeminal; Brainstem; Pons; Thalamus; Cardiac gating; Dual-echo

Introduction and background Functional imaging of the trigeminal system in humans is important to increase our understanding of its physiological and pathological role in diseases such as migraine and neuropathic pain. It is however more difficult to achieve than imaging of the cortex because of the presence of numerous potential artifacts. Only a few functional studies have examined the trigeminal system, either during headache using positron emission tomography (PET) (Hsieh et al., 1999; Matharu et al., 2004) or during both

* Corresponding author. Fax: +1 617 726 7422. E-mail address: [email protected] (W.-T. Zhang). Available online on ScienceDirect (www.sciencedirect.com). 1053-8119/$ - see front matter D 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2006.02.033

painful or innocuous sensory stimuli using blood oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) (Weiller et al., 1995; DaSilva et al., 2002; Komisaruk et al., 2002; Fitzek et al., 2004). The detection ratio of trigeminal nuclei activation in brainstem in these fMRI studies was, however, not very high. In one typical report, tactile stimulation evoked fMRI pontine activation in only 33% of cases (Komisaruk et al., 2002). Even electrical pain stimulation on trigeminal nerves elicited brainstem activation only in 25% of cases (Fitzek et al., 2004). Most investigators believe that the main reason for such a low detection rate is that the brainstem moves rapidly in the systolic phase of the cardiac cycle, highlighting the well-known vulnerability of BOLD fMRI to movement-induced artifacts. It has been shown that velocities of pulsatile movement are as high as 2 mm/s caudally in the brainstem and 1.5 mm/s medially in the thalamus (Poncelet et al., 1992), providing one reason why fMRI is less sensitive in brainstem and thalamus than in cortex. One strategy to improve fMRI sensitivity in those areas is to use statistical methods such as frequency-based filtering (Biswal et al., 1996; Corfield et al., 1999) or retrospective gating (Le and Hu, 1996; Glover et al., 2000) to remove the cardiac-related movement. However, the low sampling frequency of many fMRI studies (usually 0.3 – 0.5 Hz with TR of 2 – 3 s) precludes detecting and filtering components with the frequency of cardiac-related wave (1 – 1.33 Hz with 60 – 80 beats/min). Retrospective gating requires a large number of samples and exact synchronization of the fMRI acquisition with external cardiac/respiratory recording, which limits its application and makes it inappropriate to study response to painful stimuli. Triggering fMRI acquisition phase locked to a particular time point of the subject’s cardiac cycle (cardiac-gated fMRI) is an alternative approach commonly used to minimize the pulsatile movement-induced artifact and has been used in auditory (Guimaraes et al., 1998; Melcher et al., 2000; Griffiths et al., 2001; Backes and van Dijk, 2002), visual (DuBois and Cohen, 2000), and cervical spinal cord (Backes et al., 2001) fMRI studies. Based on the assumption that the velocity and magnitude of brain movement are relatively constant between heartbeats, cardiac-gated fMRI takes samples at the same point after the ECG triggering

W.-T. Zhang et al. / NeuroImage 31 (2006) 1506 – 1512

signal. Consequently, the repetition time (TR) varies depending on the length of each cardiac cycle. This variability in TR therefore translates into different residual longitudinal magnetization following T1 relaxation, which must be corrected (T1 correction, Guimaraes et al., 1998) before statistical comparisons. The T1 correction approach used by Guimaraes et al. (1998), however, has its own variance and therefore may not fully correct for the variable TR. T1 correction might be bypassed by using a TR long enough (e.g., 9 – 10 s) to reach nearly full T1 relaxation in each measurement (Griffiths et al., 2001; Backes and van Dijk, 2002). Statistical power, however, is sacrificed due to reduced sampling rate for a given amount of scan time. Yet another possible approach to collecting gated data is to remove completely the T1 effect in the signal intensity by acquiring measurements at multiple echo times to create a T2* map at each voxel (Wiggins and Norris, 1998; Speck and Hennig, 1998; Posse et al., 1999; Schulte et al., 2001). Previous ungated fMRI data have shown that the BOLD contrast is enhanced using single-shot multi-echo sequence (Posse et al., 1999). In the present study, we used tactile stimuli-induced activation in the trigeminal system as a model to study cardiac gating. We specifically investigated whether cardiac gating might improve our ability to detect neural activity in the principle sensory trigeminal nucleus (PSTN) in the pons and ventroposterior medial (VPM) nucleus in the thalamus, which are two major relay nuclei in the trigeminal pathway mediating tactile sensation. We compared three different cardiac-gated fMRI strategies: (1) gated with T1 correction; (2) gated with long TR and without T1 correction; and (3) gated with dual-echo EPI. Finally, we validated T1 values derived from gated fMRI by comparing them to the direct T1 mapping.

Materials and methods Ten right-handed healthy subjects (age: 31.9 T 5.6 years, mean T SD; 6 females) were enrolled and gave informed written consent. None of the subjects reported a history of or was suffering from acute or chronic trigeminal or cervical pain. The study protocol was approved by the Institutional Review Board of our institution. All MRI scans were performed on a Siemens 3-T Trio MR scanner (Siemens Medical Systems, Erlangen, Germany) with an 8-channel head array coil. Each subject was wired to a fourchannel electrocardiogram (ECG) recorder, which transmitted real-time ECG data to trigger the fMRI scanning. To ensure the best accuracy of cardiac gating, we made every effort to collect a clear ECG, and we did not observe any large ECG fluctuations

1507

during our experiments. A high-resolution isotropic (1  1  1 mm3) T1-weighted sagittal anatomical scan was first acquired for superposition of fMRI activation maps. Then each subject went through four randomly ordered functional MR scans, one ungated (TR/TE = 3000/30 ms) and three gated, during which multi-slice sagittal planes were acquired with gradient EPI sequences varying in different TRs or TEs (see below and Table 1). Of the three gated fMRI scans, two were using conventional gradient echo EPI sequence with TR of 3 heartbeats (HBs) and 9 HBs respectively (TE = 30 ms). The third gated fMRI scan used dual-echo EPI sequence which measured the signal at two echo times (21 and 55 ms) for each single excitation every 2 HBs. We chose 2 HBs to investigate a possible effect of TR in the cardiac-gated fMRI by comparing the single-echo data with different TRs (2, 3, and 9 HBs). Parallel acceleration technique (iPAT) with GRAPPA (generalized auto-calibrating partially parallel acquisition) reconstruction with acceleration factor of 2 was used to acquire images in all single echo protocols. iPAT was not used in the dual-echo protocol due to the severe ghost artifact. In eight subjects, a rapid Inversion Recovery Echo Planar Imaging (IR-EPI) (Clare and Jezzard, 2001) sequence was run right after the fMRI sequence with TR of 3 HBs. The IR-EPI sequence (TR/TE = 5220/22 ms) cycled through 20 inversion times in the same slices and same matrix as gated EPI, so that a more accurate T1 map could be fitted and used as a standard to compare the T1 values fitted from gated EPI data. An innocuous brush stimulus was given to the right ophthalmic division (V1) of the trigeminal nerve during each fMRI scan, using a brush attached to an MRI-compatible mechanical transducer, alternating 6 cycles of 60 s off and 45 s on. In the three gated fMRI scans, the ECG R-wave was used to trigger the measurement. The exact acquisition time for a given image could be read from computer’s internal clock. This allowed the interval between two consecutive measurements to be calculated. The voxel size was 3  3  3 mm3 with zero interslice gap (interleaved acquisition order; sagittal in-plane matrix 64  64; FOV 192  192 mm2). The maximum number of slices that could be measured within one heartbeat was 11 to 18 slices using conventional fMRI and 7 to 11 slices using the dual-echo fMRI and was limited by the subject’s R – R interval. All fMRI scans were accompanied with the same task paradigm and had the same duration. However, since different TRs were applied in different methods, the total number of measurements was different with as a consequence various statistical power in the data analysis. For each subject, a total of seven time series were analyzed (Table 1): (1) ungated with TR of 3 s; (2) gated with TR of 3 HBs but without T1 correction; (3) gated with TR of 3 HBs and T1 corrected; (4) first echo (TE1); and (5) second echo (TE2) of gated

Table 1 Summary of data acquisition and processing Scan no.a

Series no.

Time series

Cardiac gating

T1 correction

TR

TE (ms)

Number of time points

1 2

1 2 3 4 5 6

Ungated Gated and Gated and Dual-echo Dual-echo Dual-echo

No Yes Yes Yes Yes Yes

no no yes yes yes no

3 3 3 2 2 2

210 ¨210 ¨210 ¨300 ¨300 ¨300

7

Long TR

Yes

no

9 HBs

30 30 30 21 55 21 55 30

3

4

uncorrected corrected (TE1) (TE2) (T2*)

HB: heartbeat; TE: echo time; TR: repetition time. a The order of the four scans was randomized.

s HBs HBs HBs HBs HBs

¨75

1508

W.-T. Zhang et al. / NeuroImage 31 (2006) 1506 – 1512

dual-echo EPI, both T1 corrected; (6) T2* derived from the first and the second echo of gated dual-echo data; (7) gated with TR of 9 HBs. AFNI (Cox, 1996) and several in-house programs were used in data analysis. All time series were corrected for motion using AFNI’s standard correction algorithm. Dual-echo data (time series 4, 5, and 6) of one subject were excluded from further analysis because its spatial translation was bigger than 1 voxel size (3 mm). In time series 3, 4, and 5, T1 correction was performed according to the method outlined by Guimaraes et al. (1998): Sðn;

iÞ

¼ Aðn; iÞ ½1  expðT Ri =T1n Þ

ð1Þ

Eq. (1) can be inverted as: Aðn;

iÞ

¼ Sðn; iÞ ½1  expðT Ri =T1n Þ1

ð2Þ

where S (n, i) is the measured signal for the nth voxel at ith acquisition, A (n, i) is the maximum signal amplitude for the nth voxel at the ith acquisition without T1 weighting. The fitted T1 in nth voxel is the value that minimizes the fluctuation (or variance) of A(i) time series. Then signal intensity was corrected to the averaged TR according to Eq. (1). To validate these fitted T1 values from fMRI data, a standard T1 map from IR-EPI data in the same slice prescription was also calculated according to the Eq. (3): Si ¼ S0 ð1  2 expðTIi =T1Þ þ expðTR=T1ÞÞ

ð3Þ

Where the subscript 1. . . i labels the ith repetition of TI. S 0 and S i represent the signal intensity at full T1 relaxation and that measured at each TI. The reconstruction of the dual-echo EPI sequence produced T2* and S 0 images derived from following equations:
S1ðn; iÞ ¼ S0 exp TE1 =T24ðn; iÞ ð4Þ S2ðn;

iÞ

T24ðn;

¼ S0 exp TE2 =T24ðn; iÞ

iÞ




¼ ðTE2  T E1 Þ= ln S1ðn; iÞ =S2ðn;

ð5Þ iÞ





Fig. 1. Illustration of an axial section of the pons at the level of principle trigeminal sensory nucleus (PSTN) (Duvernoy, 1995).

ð6Þ

where the subscript 1. . . n labels the nth voxel and subscript 1. . . i labels the ith measurement. S 1(n, i) and S 2(n, i) stand for the measured signal amplitude in the first and second echo time. S 0 stands for the initial signal intensity. All time series were then high-pass filtered (1/210 Hz) and smoothed (full width of half maximum (FWHM) = 4 mm). Deconvolution was used to evaluate the correlation of the signal intensity changes with a predicted hemodynamic response model. To correct for the different statistical power induced by the different number of measurements, the t statistics were transformed into Z statistics. Z scores have a mean of zero and a standard deviation of one, have a Gaussian distribution, and link to no degree of freedom. The threshold was set to P < 0.0001 (Z = 3.88, uncorrected) to determine an individual subject’s activation. Averaged group analysis was also performed using each individual’s Talairach-transformed statistical maps and the threshold was set to P < 0.01 (Z = 2.33, uncorrected). Ipsilateral dorsal pons and contralateral posterior ventral thalamus were outlined manually using the high-resolution anatomical images and used as two regions of interest (ROI). In the ipsilateral dorsal pons, we encompassed two slices above, two slices below, and the slice in which trigeminal roots traverse, where the principle sensory trigeminal nucleus (PSTN) is located

closely (Fig. 1, Duvernoy, 1995). The contralateral posterior ventral thalamus was defined as a quarter of the full dimension of the thalamus. Since PSTN in the pons and ventroposterior medial (VPM) nucleus in the thalamus are not identifiable in the anatomical MR images, activation overlapping with these defined regions was attributed to PSTN and VPM, respectively. For each ROI, the mean, the maximum, the standard deviation (SD) of the Z scores, and the SD of baseline signal intensities were calculated. Both SDs were divided by their means to eliminate the effect of different units and amplitude. One-way analysis of variance (ANOVA) was used for comparison between the groups. A Tukey post hoc comparison was performed between all pairs of groups.

Results The ratios of subjects who showed activation in the ipsilateral principle sensory trigeminal nucleus (PSTN) and ventroposterior medial nucleus (VPM) in the contralateral thalamus are summarized in Table 2. Four out of 10 (40%) ungated time series showed activation in PSTN. The detection rate increased to 70% for the gated single echo (TR = 3 HBs) and to 67% for the second echo (TE2) of dual-echo EPI time series. The detection rate of T2* time series calculated from dual-echo images was 89%, the highest among all groups. Only 20% of the gated time series with long TR showed PSTN activation, half lower than that in ungated time series. At a lower threshold of P < 0.01, three more subjects (in total 50%) showed activation in long TR time series and had consistent time courses. In the contralateral VPM of thalamus, 50% of ungated time series showed activation. The detection rate increased in one gated and T1-corrected time series (second echo of dual-echo sequence, 78%) and T2* time series (89%), but not in the first echo of dual-echo sequence (44%), gated single echo and corrected (50%), or gated but uncorrected time series (20%).

W.-T. Zhang et al. / NeuroImage 31 (2006) 1506 – 1512

1509

Table 2 The detection rate, the mean, maximum, and standard deviation (SD) of Z scores and averaged SD of baseline signal intensities (SI) in the pons and thalamus Detection rate

Mean Z

Max Z

SD of Z

SD of baseline SI

PSTN in the pons Ungated Gated and uncorrected Gated and corrected Dual-echo (TE1) Dual-echo (TE2) Dual-echo (T2* ) Long TR

4/10 3/10 7/10 4/9 6/9 8/9 2/10

(40%) (30%) (70%) (44%) (67%) (89%) (20%)

1.3 0.8 1.5 1.4 1.7 1.8 0.7

T T T T T T T

0.5 0.3** 0.6 0.9 1.0 0.5* 0.2*

3.7 2.9 5.0 4.5 5.7 6.8 2.9

T T T T T T T

1.0 1.3 3.1 3.7 3.1 3.8* 0.9

0.8 0.7 0.7 0.6 0.6 0.7 0.7

T T T T T T T

0.1 0.1 0.2 0.3 0.2* 0.1 0.1

2.4 2.7 2.3 2.2 1.2 6.5 2.3

T T T T T T T

0.3 0.5 0.3 0.4 0.2* 1.7*** 0.3

VPM in the Thalamus Ungated Gated and uncorrected Gated and corrected Dual-echo (TE1) Dual-echo (TE2) Dual-echo (T2* ) Long TR

5/10 2/10 5/10 4/9 7/9 8/9 1/10

(50%) (20%) (50%) (44%) (78%) (89%) (10%)

1.4 0.9 1.4 1.8 2.0 2.0 0.7

T T T T T T T

0.7 0.7 0.8 1.0 0.9 0.6 0.4*

4.4 2.0 3.8 4.2 5.0 5.4 2.0

T T T T T T T

2.1 1.7 1.6 1.8 1.1 1.2 1.0

0.7 0.5 0.6 0.7 0.7 0.7 0.7

T T T T T T T

0.2 0.1 0.2 0.2 0.2 0.1 0.2

2.1 2.5 2.1 1.9 0.8 3.3 2.2

T T T T T T T

0.2 0.5 0.2 0.5 0.2***,**** 1.4** 0.2

Data presented as mean T SD. * P < 0.05. ** P < 0.01. *** P < 0.001 compared with ungated time series. **** P < 0.05 compared with TE1. PSTN: principle sensory trigeminal nucleus; VPM: ventroposterior medial nucleus.

Again, the gated time series with the long TR failed to show improvement in detecting activation in the thalamic VPM (10%) compared to the ungated fMRI. One subject showed no activation in both ROIs in all time series for no clear reason. Analysis of the S 0 map from each subject’s dual-echo data revealed no activation in PSTN and VPM, implying that effects due to inflow were small. Activation in PSTN in representative subjects is shown in Fig. 2. Averaged maps showing the activation in PSTN and VPM are

shown in Fig. 3. In contrast, as shown in Figs. 2 and 3 either the contralateral dorsal pons or the ipsilateral medial thalamus was not activated by the brush stimulus, and we consider this strong evidence that the activation in our ROIs was a real activation rather than an artifact. The mean, maximum, and standard deviation (SD) of Z scores and averaged SD of signal intensities in each ROI are also reported as mean T SD in Table 2 (column 3 through 6). In the PSTN in the

Fig. 2. fMRI Z statistic maps of representative subjects. Both calculated T2* and T1-corrected time series successfully showed tactile stimuli-induced activation in the principle sensory trigeminal nucleus (PSTN) in the brainstem. Images are shown in the sagittal and axial planes. A: anterior, P: posterior; L: left; R: right.

1510

W.-T. Zhang et al. / NeuroImage 31 (2006) 1506 – 1512

Fig. 3. Averaged Z statistic maps of PSTN and VPM in pons and thalamus with different strategies. Activation in thalamus was not found in long TR and ungated time series hence not shown here. Images are shown in the sagittal and axial planes. A: anterior, P: posterior; L: left; R: right.

pons (Table 2, top part), the mean (1.8 T 0.5) and maximal Z scores (6.8 T 3.8) of the T2* time series were significantly higher than those of the ungated data (1.3 T 0.5 and 3.7 T 1.0, respectively). We found that all the three gated and the T1-corrected time series resulted in higher maximal Z scores than the ungated approach for activation in PSTN, and this approached statistical significance. The SD of Z scores of the TE2 time series (0.6 T 0.2) was significantly lower than that of the ungated (0.8 T 0.1), suggesting more stability after gating and T1 correction. Long TR data showed the lowest mean Z score (0.7 T 0.2), which is consistent to its low detection ratio. The SD of the baseline signal intensities reflect the amplitude of the noise. We found that the SD of the signal intensities of T1corrected time series (TE2, 1.2 T 0.2) was significantly smaller than the SD of the ungated time series (2.4 T 0.3). However, the SD of T2* series (6.5 T 1.7) is significantly higher than that of both the ungated and the gated then T1-corrected series, presumably due to fitting errors. In the VPM of the thalamus (Table 2, bottom part), the mean Z score of the long TR data (0.7 T 0.4) was significantly lower than that of the ungated data (1.4 T 0.7). We did not detect any other significance in the mean, maximum, or SD of Z scores among the different techniques. The SD of the baseline signal intensities of T1-corrected time series (TE2, 0.8 T 0.2) was significantly lower than the SD of the ungated (2.1 T 0.2) and TE1 (1.9 T 0.5) time series. Similar to the brainstem, the SD of T2* series (3.3 T 1.4) was significantly higher than that of both the ungated and the gated then T1-corrected series. T1 maps were fitted from IR-EPI and gated EPI data with TR of 3 HBs and 2 HBs. It appears that T1 map from IR-EPI data has less noise and better contrast compared to T1 maps fitted from gated EPI. Quantitatively, T1 values from IR-EPI are 2668 T 119 ms, 1325 T 10 ms, and 910 T 10 ms (mean T SEM, n = 8) in cerebrospinal fluid, gray and white matter, respectively, which compare well to the values reported by Wansapura et al. (1999).

These T1 values, however, are not significantly different from T1 fitted from gated EPI with TR of 3 HBs and 2 HBs. The activation results described above were not changed by using the IR-EPIbased T1 maps for correction (data not shown).

Discussion The present study systematically compared different cardiac gating methods to conventional ungated fMRI in their ability to detect activation in the trigeminal pathway. Brainstem studies clearly require novel approaches: unlike primary sensory fMRI activation, where discussions focus around how to best quantify the robust activation present in nearly every subject, simply finding any activation in these small regions of the moving brain is considered an accomplishment, hence, the strong motivation to find improved approaches such as cardiac gating. Our results are consistent with earlier reports that did not use gating: in our experiments, 40% of the ungated time series acquired during tactile stimulation to the face showed activation in PSTN, which is similar to the detection rate of 33% reported by Komisaruk et al. (2002) during the same sensory stimuli. Activation in the VPM of the thalamus was observed in 50% of ungated cases. Our work further confirms the earlier reported approach of applying cardiac gating and T1 correction as suggested by Guimaraes et al. (1998); in our study, the ratio of detected activation increased to 70% in the brainstem and to 78% in the thalamus. This advantage was gained from decreased standard deviation in regions of interest, which is consistent with Guimaraes et al.’s conclusion. In contrast, gated but uncorrected time series did not gain such advantage; this underscores the efficacy of T1 correction for variable R – R intervals. Table 2 clearly shows that the SD of T1-corrected gated fMRI time series was significantly lower than ungated and gated but uncorrected because it eliminated both pulsatile movementinduced noise and the noise due to variable TRs.

W.-T. Zhang et al. / NeuroImage 31 (2006) 1506 – 1512

Our work extends these earlier results, however, by exploring additional gating methods that demonstrate superior results. First, our data show that a straightforward shortening of the TR to 2Hbs results in a decrease of SD of the data in the brainstem and in the thalamus, suggesting that T1 correction is more efficient with TR of 2 HBs. This might be due to the spins in the TR of 2 HBs data being less saturated than those with TR of 3 HBs, which made the T1 fitting and T1 correction more accurate and efficient. This is further confirmed by the fact that there was no significant difference between the SDs before T1 correction of the 2 HBs data and 3 HBs data (3.0 T 1.3 vs. 2.7 T 0.5 in the brainstem and 2.7 T 1.4 vs. 2.5 T 0.5 in the thalamus). The use of the parallel imaging in the TR of 3 HB, but not in the 2 HB dual-echo sequence, is not likely the reason for this difference in SD since it has been shown that parallel imaging actually increases signal to noise ratio (Frederick et al., 1999; Bellgowan et al., 2006). Our results showed that the gain obtained from shortening the TR surpassed the loss of not using iPAT in the dual-echo single series. We also sought to further extend these earlier approaches by examining two other strategies. Firstly, the dual-echo technique combined with cardiac gating provides an enticingly effective way to detect brainstem activation, at least in theory. Conventional fMRI EPI sequence measures the signal decay at a single echo time, which is typically chosen to be close to the averaged tissue T2* in order to provide good contrast. Considering the regional and inter-subject variability of tissue T2* and intrinsic hardware instability, as well as possible variability of initial signal intensity (S 0), this T2* contrast may not be fully optimized with typical fMRI approaches. Posse et al. (1999) reported that BOLD contrast in visual cortex had been enhanced using single-shot multi-echo EPI sequence. To the best of our knowledge, there is no previous work combining multi-echo EPI with cardiac gating. We found that using the T2* time series approach, the expected activation in brainstem and thalamus could be identified in 8/9 subjects (89%). Furthermore, in the T2* series, the mean and maximal Z scores in the brainstem were significantly higher than those of the ungated series, suggesting, at least in brainstem, its superiority to other gated fMRI methods. The improvement was gained most likely because gated dual-echo fMRI eliminates not only the effect from pulsatile motion but also from S 0-related magnet drift over time. Additionally, Posse et al. (1999) have reported that the contrast to noise ratio of T2* maps is theoretically higher compared to conventional single echo fMRI. There were other advantages of the dual-echo sequence as well. These include (1) the initial signal intensity (S 0) could be monitored, thus allowing to estimate possible inflow effects in regions close to large vessels, which is especially advantageous for gated fMRI; (2) it might be possible to quantify fMRI signal change using calculated T2* change, thus enabling direct comparisons of results from different regions and individuals; (3) the effect of S 0-related magnetic drift over time could potentially be eliminated after getting T2*, which might have particular advantage for fMRI scans lasting longer than 10 min. There are a few disadvantages to the multi-echo T2* approach. Most noticeable is that the measurement time per slice approximately doubled in multi-echo BOLD fMRI with respect to conventional single-echo fMRI (Posse et al., 1999). In the present study, we used a variant of Posse et al.’s methodology: only two echo images to calculate the T2* map so that the acquisition time per slice did not increase greatly. Nevertheless, with our current hardware and software settings, the total number of slices for the

1511

dual-echo EPI sequence is reduced: about 7 – 11 slices per HB, fewer than the 11 – 18 slices per HB of other gated fMRI methods. Another limitation we did not explore is how accurate the T2* map might be by using only 2 different echo times. Indeed, we found that the SD in the T2* series significantly increased, which partly counteracted the detection benefit. We believe, however, that with further optimization, this approach may become the most appropriate method in the long term. Cardiac gating with long TR (¨9 s) initially appeared to be attractive because of its simplicity; however, the results obtained were not encouraging. One reason for the poor results of this approach is that we could acquire only about one-third (75 measurements) of the time points that could be acquired with a TR of 3 HBs. While some paradigms can compensate for this by extending the length of stimulus exposure, somatosensory habituation (Pubols, 1982; Milne et al., 1991; Stancak et al., 2003) precludes such an approach in our paradigms. Although previous work using cardiac-gated fMRI with long TR and acquiring a similar number of volumes (e.g., 60 and 96 measurements) has successfully showed BOLD changes in the inferior colliculus (Griffiths et al., 2001; Backes and van Dijk, 2002) and cochlear nucleus in the brainstem (Griffiths et al., 2001), we believe that different stimulation modality used (brush vs. pitch) and different target nuclei might explain such a difference between these and our study. Gated fMRI with long TR may be uniquely beneficial for event-related auditory study because it prevents the interaction between sound stimulus and scanner noises by using a sparse sampling approach. Different research problems have different priorities and tradeoffs must be made when deciding upon a cardiac-gated fMRI strategy to use. Conventional single-echo gated fMRI allows broader brain coverage, while at least in this study dual-echo gated fMRI results in better sensitivity. If broad brain coverage is required and long-time scanning is feasible, retrospective gating might be an alternative method. We conclude that cardiac-gated fMRI with dual-echo EPI provides the best detection of brainstem and thalamus activity during tactile stimuli.

Acknowledgments This work was supported by PHS grants 5P01 NS35611-07, National Center for Research Resources General Clinical Resource Centers Program (M01-RR-01066), NCRR Center for Functional Neuroimaging Technologies (5P41RR014075), Mental Illness and Neuroscience Discovery (MIND) Institute, and Federazione Italiana Sclerosi Multipla (FISM) 2003/B/8. We also want to thank Dr. Nouchine Hadjikhani, Dr. Vitaly Napadow and Dr. Kathleen Hui for the constructive discussions.

References Backes, W.H., van Dijk, P., 2002. Simultaneous sampling of event-related BOLD responses in auditory cortex and brainstem. Magn. Reson. Med. 47 (1), 90 – 96. Backes, W.H., Mess, W.H., Wilmink, J.T., 2001. Functional MR imaging of the cervical spinal cord by use of median nerve stimulation and fist clenching. Am. J. Neuroradiol. 22 (10), 1854 – 1859. Bellgowan, P.S., Bandettini, P.A., van Gelderen, P., Martin, A., Bodurka, J., 2006. Improved BOLD detection in the medial temporal region using

1512

W.-T. Zhang et al. / NeuroImage 31 (2006) 1506 – 1512

parallel imaging and voxel volume reduction. NeuroImage 29 (4), 1244 – 1251. Biswal, B., DeYoe, A.E., Hyde, J.S., 1996. Reduction of physiological fluctuations in fMRI using digital filters. Magn. Reson. Med. 35, 107 – 113. Clare, S., Jezzard, P., 2001. Rapid T(1) mapping using multislice echo planar imaging. Magn. Reson. Med. 45 (4), 630 – 634. Corfield, D.R., Murphy, K., Josephs, O., Fink, G.R., Frackowiak, R.S.J., Guz, A., Adams, L., Turner, R., 1999. Cortical and subcortical control of tongue movement in humans: a functional neuroimaging study using fMRI. J. Appl. Physiol. 86 (5), 1468 – 1477. Cox, R.W., 1996. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput. Biomed. Res. 29 (3), 162 – 173. DaSilva, A.F., Becerra, L., Makris, N., Strassman, A.M., Gonzalez, R.G., Geatrakis, N., Borsook, D., 2002. Somatotopic activation in the human trigeminal pain pathway. J. Neurosci. 22 (18), 8183 – 8192. DuBois, R.M., Cohen, M.S., 2000. Spatiotopic organization in human superior colliculus observed with fMRI. NeuroImage 12 (1), 63 – 70. Duvernoy, H.M., 1995. The Human Brain Stem and Cerebellum: Surface, Structure, Vascularization, and Three-dimensional Sectional Anatomy with MRI. Springer-Verlag Wien, New York, pp. 64. Fitzek, S., Fitzek, C., Huonker, R., Reichenbach, J.R., Mentzel, H.J., Witte, O.W., Kaiser, W.A., 2004. Event-related fMRI with painful electrical stimulation of the trigeminal nerve. Magn. Reson. Imaging 22 (2), 205 – 209. Frederick, B.deB., Wald, L.L., Maas, L.C. III, Renshaw, P.F., 1999. A phased array echoplanar imaging system for fMRI. Magn. Reson. Imaging 17 (1), 121 – 129. Glover, G.H., Li, T.Q., Ress, D., 2000. Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Magn. Reson. Med. 44 (1), 162 – 167. Griffiths, T.D., Uppenkamp, S., Johnsrude, I., Josephs, O., Patterson, R.D., 2001. Encoding of the temporal regularity of sound in the human brainstem. Nat. Neurosci. 4 (6), 633 – 637. Guimaraes, A.R., Melcher, J.R., Talavage, T.M., Baker, J.R., Ledden, P., Rosen, B.R., Kiang, N.Y., Fullerton, B.C., Weisskoff, R.M., 1998. Imaging subcortical auditory activity in humans. Hum. Brain Mapp. 6 (1), 33 – 41. Hsieh, J.C., Meyerson, B.A., Ingvar, M., 1999. PET study on central processing of pain in trigeminal neuropathy. Eur. J. Pain 3 (1), 51 – 65. Komisaruk, B.R., Mosier, K.M., Liu, W.C., Criminale, C., Zaborszky, L., Whipple, B., Kalnin, A., 2002. Functional localization of brainstem and cervical spinal cord nuclei in humans with fMRI. Am. J. Neuroradiol. 23 (4), 609 – 617.

Le, T.H., Hu, X., 1996. Retrospective estimation and correction of physiological artifacts in fMRI by direct extraction of physiological activity from MR data. Magn. Reson. Med. 35 (3), 290 – 298. Matharu, M.S., Cohen, A.S., McGonigle, D.J., Ward, N., Frackowiak, R.S., Goadsby, P.J., 2004. Posterior hypothalamic and brainstem activation in hemicrania continua. Headache 44 (8), 747 – 761. Melcher, J.R., Sigalovsky, I.S., Guinan, J.J.Jr., Levine, R.A., 2000. Lateralized tinnitus studied with functional magnetic resonance imaging: abnormal inferior colliculus activation. J. Neurophysiol. 83 (2), 1058 – 1072. Milne, R.J., Kay, N.E., Irwin, R.J., 1991. Habituation to repeated painful and non-painful cutaneous stimuli: a quantitative psychophysical study. Exp. Brain Res. 87 (2), 438 – 444. Poncelet, B.P., Wedeen, V.J., Weisskoff, R.M., Cohen, M.S., 1992. Brain parenchyma motion: measurement with cine echo-planar MR imaging. Radiology 185 (3), 645 – 651. Posse, S., Wiese, S., Gembris, D., Mathiak, K., Kessler, C., GrosseRuyken, M.L., Elghahwagi, 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, 87 – 97. Pubols, Jr., B.H., 1982. Factors affecting cutaneous mechanoreceptor response: II. Changes in mechanical properties of skin with repeated stimulation. J. Neurophysiol. 47 (3), 530 – 542. Schulte, A.C., Speck, O., Oesterle, C., Hennig, J., 2001. Separation and quantification of perfusion and BOLD effects by simultaneous acquisition of functional I(0)- and T2(*)-parameter maps. Magn. Reson. Med. 45 (5), 811 – 816. Speck, O., Hennig, J., 1998. Functional imaging by I0- and T2*parameter mapping using multi-image EPI. Magn. Reson. Med. 40 (2), 243 – 248. Stancak, A., Svoboda, J., Rachmanova, R., Vrana, J., Kralik, J., Tintera, J., 2003. Desynchronization of cortical rhythms following cutaneous stimulation: effects of stimulus repetition and intensity, and of the size of corpus callosum. Clin. Neurophysiol. 114 (10), 1936 – 1947. Wansapura, J.P., Holland, S.K., Dunn, R.S., Ball, W.S.Jr., 1999. NMR relaxation times in the human brain at 3.0 Tesla. J. Magn. Reson. Imaging 9 (4), 531 – 538. Weiller, C., May, A., Limmroth, V., Juptner, M., Kaube, H., Schayck, R.V., Coenen, H.H., Diener, H.C., 1995. Brain stem activation in spontaneous human migraine attacks. Nat. Med. 1 (7), 658 – 660. Wiggins, C.J., Norris, D.G., 1998. Dual-echo EPI: An fMRI method for regions of varying B0 homogeneity. Proceedings: 6th Annual Meeting of the International Society for Magnetic Resonance in Medicine (ISMRM) p. 1449.