Abstract
Experimental data up to 7.0 T show that the blood oxygenation level-dependent (BOLD) signal of functional magnetic resonance imaging (fMRI) increases with higher magnetic field strength. Although several studies at 11.7 T report higher BOLD signal compared with studies at 7.0 T, no direct comparison at these two field strengths has been performed under the exact same conditions. It therefore remains unclear whether the expected increase of BOLD effect with field strength will still continue to hold for fields >7.0 T. To examine this issue, we compared the BOLD activation signal at 7.0 and 11.7 T with the two common sequences, spin-echo (SE) and gradient-echo (GE) echo planar imaging (EPI). We chose the physiologically well controlled rat model of electrical forepaw stimulation under medetomidine sedation. While a linear to superlinear increase in activation with field strengths up to 7.0 T was reported in the literature, we observed no significant activation difference between 7.0 and 11.7 T with either SE or GE. Discussing the results in light of the four-component model of the BOLD signal, we showed that at high field only two extravascular contributions remain relevant, while both intravascular components vanish. Constancy of the BOLD effect is discussed due to motional narrowing, i.e., susceptibility gradients become so strong that phase variance of diffusing spins decreases and therefore the BOLD signal also decreases. This finding will be of high significance for the planning of future human and animal fMRI studies at high fields and their quantitative analysis.
Introduction
In 1990, Ogawa and colleagues reported the effect of blood oxygenation on the magnetic resonance imaging (MRI) signal behavior, known as blood oxygenation level-dependent (BOLD) MRI signal, in rats at 4.7 tesla (T) and 7.0 T (Ogawa et al., 1990). This BOLD signal was rapidly exploited for the investigation of functional brain activation with functional magnetic resonance imaging (fMRI), a technique now widely applied as the dominant imaging method to study cognitive processes under both normal and pathophysiological conditions. Early on, it had been noted that the BOLD effect increases with higher magnetic fields (Ogawa et al., 1990, 1993). Field dependence studies reported five times higher BOLD signal at 4.0 T than at 1.5 T, suggesting that the dependence of the deoxygenation contrast on the static field is greater than linear, the exponent being between 1.5 and 1.8 (Turner et al., 1993), with an increase of 70% in activated voxels (Yang et al., 1999). Motivated by this observation, cognitive scientists today aim to increase the magnetic field strength continuously in the expectation of further increasing the sensitivity of the BOLD fMRI method.
Later reports have introduced models with the goal of better describing the BOLD effect. According to these, the BOLD effect consists of four components with their own behavior, depending on measurement sequence as well as field strength. This more complex view of BOLD signal behavior motivated us to formulate the hypothesis that the expectation of a continuous increase of the BOLD effect with ever higher magnetic fields may not be justified in that the BOLD effect may actually reach saturation with an increasing magnetic field or even show a maximum, depending on the behavior of the individual components and their relative contributions at very high fields.
Experimental BOLD data have already been described using the four-compartment BOLD model containing both intravascular and extravascular contributions from single veins and capillaries (Hoogenraad et al., 2001). At present, however, there are no experimental data to demonstrate continuing increase of the BOLD signal for magnetic fields higher than 7.0, 8.0, or 9.4 T, the upper field strengths presently accessible for human studies (Vaughan et al., 2006; Shmuel et al., 2007; Walter et al., 2008).
To test our hypothesis, we examined the BOLD signal at the two high magnetic field strengths of 7.0 and 11.7 T in a well established rat model of forepaw stimulation (Hyder et al., 1994; Brinker et al., 1999; Gsell et al., 2006). Using a recently developed protocol of medetomidine sedation (Ramos-Cabrer et al., 2005; Weber et al., 2006) together with detailed monitoring and regulation of the physiological status, this model allows extremely high stability and reproducibility of the BOLD behavior upon stimulation. We analyzed the field dependence of the BOLD signal at both field strengths. Additionally, spin-echo (SE) and gradient-echo (GE) echo planar imaging (EPI) BOLD imaging was compared to distinguish and quantify directly the intravascular and extravascular components at both field strengths (Hoogenraad et al., 2001; Norris, 2006; Uludag et al., 2009).
Materials and Methods
Animals, anesthesia, and physiological monitoring.
All experiments were performed in accordance with the German Animal Welfare Act and approved by the local authorities. Twelve male Wistar rats, weighing between 320 and 455 g, were examined. Animals were housed in cages with controlled temperature (21 ± 1°C), humidity (55 ± 10%), a 12 h light/12 h dark cycle, and food ad libitum.
The sedation used for BOLD experiments follows the recently established and validated medetomidine protocol (Ramos-Cabrer et al., 2005; Weber et al., 2006). It consists of continuous subcutaneous administration of medetomidine (Domitor; Pfizer) after initial halothane anesthesia (2%) as described previously (Weber et al., 2008). Respiration rate was monitored continuously using DASYLab (version 9.0; MeasX) at 11.7 T, while PC-SAM (version 4.04; SA Instruments) was used at 7.0 T. Body temperature was kept constant at 37 ± 1°C by using a combination of a rectal temperature probe and a feedback-controlled heating system (medres medical research).
Experimental design.
In the first study (field comparison study using spin-echo sequence), eight adult male Wistar rats were examined at three different time points after arrival, 1–2 weeks (320–390 g), 3–4 weeks (370–420 g), and 5–6 weeks (400–455 g), to study the behavior of the BOLD response with SE-EPI at 7.0 and 11.7 T. BOLD fMRI scans were conducted four times within one session on each scanner, and animals were allowed to rest for 10 min between stimulation sessions. Note that fMRI was performed at both field strengths within a single experimental session. After completion of the protocol at one scanner, the animal—still in sedation—was transferred in the holder to the second scanner. The protocol was repeated on this system after an additional resting period of 15 min. Animals were divided into two groups. Group 1 was scanned at 7.0 T first and subsequently transferred to the 11.7 T system. Group 2 was scanned at 11.7 T first and then at 7.0 T (Fig. 1A). Sessions were repeated three times for each animal.
Scheme of experimental protocols (A, B) and data analysis strategy (C). A, Experimental protocol of the SE field comparison study. In group 1, the four animals were scanned at 11.7 T (red) first and then, within the same anesthesia session, scanned at 7.0 T (blue). Group 2 was scanned at 7.0 T first, followed by scans at 11.7 T. Each session consisted of four SE-EPI scans for BOLD activation at each field strength, and each session was repeated three times. B, Experimental protocol of the GE-SE comparison study. Animals in group 3 were scanned at 11.7 T with alternating SE- (red) and GE-EPI (gray) scans. After 2 weeks the animals were scanned at 7.0 T (blue for SE; black for GE). Group 4 was scanned at 7.0 T first, and 2 weeks later at 11.7 T. Each session consisted of 8–15 EPI scans in total. C, For data analysis, the CS of the activation area in the S1 was determined. This area was used to determine minimum, maximum, and average BOLD intensity parameters.
In the second study (comparison study of spin-echo and gradient-echo), another four animals (350–450 g) were examined to study the difference between SE-EPI and GE-EPI in separate sessions at the two field strengths (Fig. 1B).
Magnetic resonance imaging.
MRI experiments were conducted on dedicated high-field animal scanners (BioSpec; Bruker BioSpin MRI) at 11.7 and 7.0 T, with horizontal bore magnets equipped with actively shielded gradient systems. The 11.7 T system had a 9 cm gradient (750 mT/m) with an integrated 12 channel shim system, a 72 mm quadrature resonator for radio frequency (RF) transmission, and a quadrature surface coil of 3.6 cm diameter for RF reception. The 7.0 T system had a 29 cm shim system, a 20 cm gradient (200 mT/m), custom-built actively decoupled RF coils for transmission (14 cm Helmholtz coil), and a 2.8 cm linear surface coil for reception.
The animal holder was designed for use with both systems to maintain sedation while the animal was transferred between the two magnets. Animals were fixed in the cradle using a tooth bar, ear bars, and adhesive tape, and coils were positioned over the head. The animal's head was positioned so that the primary somatosensory cortex (S1) (4.7 mm caudal to the rhinal fissure) was located at the magnet's isocenter.
GE pilot scans were used for accurate positioning. Local shimming was performed using an automated shimming method for linear shims. The region of interest (ROI) was centered that way so that the lower three-quarters were filled with signal from brain. Quantitative Carr-Purcell-Meiboom-Gill multi-slice multi-echo (MSME) and multi-gradient-echo (MGE) scans were performed to assess T2 and T2* relaxation times of brain tissues. T2 and T2* maps were measured in all eight animals of the first study during their first session. Scan parameters were as follows: field of view (FOV) = 2.56 × 2.56 cm2, matrix = 128 × 128, isotropic in-plane resolution = 200 × 200 μm2, 16 slices at 1 mm thickness, repetition time (TR) = 3000 ms, 16 echoes with echo time (TE) = 12, 24, 36, …, 192 ms for MSME and TE = 5, 10, 15, …, 80 ms for MGE. MSME and MGE images were processed using ImageJ (W. S. Rasband, National Institutes of Health; http://rsb.info.nih.gov/ij). T1 relaxation times were determined by varying TR in SE-EPI scans.
Functional MRI protocol.
Functional MRI was achieved using BOLD signal. Coronal multi-slice single-shot EPI images with the following parameters were acquired: FOV = 2.56 × 2.56 cm2, matrix = 64 × 64, resolution = 400 × 400 μm2, five consecutive slices of 2 mm thickness, TR = 3000 ms, partial Fourier factor = 1.43 (echo position 35%), bandwidth = 150 kHz.
Before the field comparison study with SE, the TE of SE-EPI was varied: 20, 23, 25, 27, 30, and 40 ms at 11.7 T, and 25 and 30 ms at 7.0 T. Based on the results that no significant difference in BOLD signal and size of activation was observed, the SE-EPI echo time was set to TE = 25 ms with a readout window of <30 ms for both field strengths, and thereby the diffusion and motional narrowing regime was also kept constant at both field strengths. Before the SE-GE comparison study, the TE of GE-EPI was varied: 12, 14, 16, 18, 20, and 30 ms at 11.7 T, and 14, 16, 18, and 20 ms at 7.0 T. TE for GE-EPI was then set to TE = 16 ms, being the best compromise between image quality, signal intensity, BOLD signal, and activation size, while TE for SE-EPI was kept at 25 ms, as in the first study. In both studies the chosen values were a compromise of increased image quality [high signal-to-noise ratio (SNR), less signal dropout] at shorter TE and increased BOLD sensitivity (higher BOLD signal, larger activation cluster) at higher TE. Both sequences (SE is known to resolve small structures, while GE shows larger clusters and higher activations) allowed recording BOLD fMRI reliably at both field strengths.
Electrical forepaw stimulation was performed with two electrodes under the skin in the paw, using rectangular pulses (1.5 mA, 6 Hz, 0.3 ms, home-built stimulation unit). The paradigm consisted of five blocks (45 s resting period and 15 s activation period) and ended with an additional 45 s resting period, thus resulting in 115 repetitions in a total experimental time of 5 min 45 s. The stimulated paw remained unchanged across all sessions.
Image and temporal signal-to-noise ratio.
SE scans (MSME: FOV = 4 × 8 cm2; matrix = 128 × 256; resolution = 312 × 312 μm2; 1 slice of 0.5 mm thickness; TR = 500 ms; TE = 15 ms, bandwidth = 100 kHz) were used to measure the image SNR using the ratio of the signal intensity of the brain in a coronal slice and the SD of the noise outside the brain. SNR was also calculated for SE-EPI and GE-EPI. Signal intensity was determined in a ROI set in the S1, contralateral to the activation area.
Temporal SNR was measured in the central slice in the hemisphere, contralateral to the activation, in all SE-EPI scans.
BOLD threshold analysis.
Statistical parametric activation maps were constructed with the software STIMULATE (Strupp, 1996). The time course of each pixel during forepaw stimulation was examined using a paired Student's t test (p = 0.01), with period zero mean variance baseline correction activated, i.e., the calculation of the distribution variance is based on data where the period mean is subtracted from each point within its respective period. No background threshold, masking, filtering, motion correction or hemodynamic response function modeling was used. All image series were checked for motion, and those with severe artifacts were excluded. Individual scans were tested further for motion or drifts using FSL software (Smith et al., 2004); detected frequency and temperature drifts were <0.5 pixels.
Only clusters including at least three adjacent activated pixels in the slice through the center of the activated cortex region were considered as true activation clusters. In some cases, the first scan revealed only one or two pixels in that area. However, these pixels were accepted as a cluster when the S1 cluster of consecutive scans showed activation in >3 pixels. For each scan the following four cluster parameters were calculated (Fig. 1C): (1) cluster size (CS), i.e., the absolute number of activated pixels in the cluster; (2) minimum BOLD signal (MIN), i.e., the BOLD signal of the lowest amplitude pixel, which represents the detection limit (constraining MIN consequently changes the other cluster parameters); (3) average BOLD signal (AVG), which is the representative value for the activation amplitude of the whole cluster; and, finally, (4) maximum BOLD signal (MAX), i.e., the BOLD signal of the highest amplitude pixel. Furthermore, lower thresholds of 0 (no threshold), 1, and 2% on the BOLD signal were applied to all pixels. This permitted comparison of the unrestricted activation cluster (MIN > 0%) to the activation cluster completely in S1 (MIN > 1%) and to the center of the activation (MIN > 2%).
BOLD histogram analysis.
While the threshold analysis described above has the advantage of obtaining individual BOLD parameters for every animal, it has the disadvantage that detailed information on the distribution of activated voxels is lost. Therefore, data were reanalyzed using a second analysis approach: BOLD percentage change (BOLD%) and functional contrast-to-noise ratio (fCNR) were calculated on a pixel-by-pixel basis without any threshold. ROIs were drawn manually, each covering the S1 areas on the ipsilateral and contralateral sides of activation. The ROI-based analysis was used to calculate the average BOLD% and fCNR for every session as well as to fill a histogram of BOLD% and fCNR distribution using data from all animals and all sessions. Compared with the previous threshold analysis, histograms show all pixels from all animals without any threshold, and therefore small and negative values on BOLD% and fCNR are possible.
Statistical analysis.
To evaluate the SE study results, a matched pairs t test and a post hoc power analysis were performed with the power analysis program G*Power3 (Erdfelder et al., 1996; Faul et al., 2007), using an α error probability of 0.05 and total sample size of n = 18, after summing up all scans within one session. The effect size (Cohen's d) and the power (1 − β error probability) were calculated for the maximum BOLD contrast for a theoretical increase of 67%, as expected from literature reports at lower field strengths. Also, the maximum possible BOLD increase in agreement with our results was calculated [power = 1 − β (set to) ! = 0.90].
Results
Image and temporal signal-to-noise ratio
SNR measurements were performed to compare image quality between systems and to ensure reproducibility across sessions. Volume-normalized image SNRs of SE (MSME) images were SNR/mm3 = 913 ± 63 at 11.7 T and SNR/mm3 = 514 ± 95 at 7.0 T, showing an SNR increase of 78% for SE. The theoretical, linear increase of SNR with field strength is expected to be 67%; the additional increase of ∼11% is due to different coils at both field strengths.
The average temporal SNR (tSNR), determined in the nonactivated cortex on the EPI image series, was 43 ± 9 at 7.0 T and 57 ± 10 at 11.7 T, corresponding to an increase in tSNR of 33% at the higher field strength.
Relaxation times T1, T2, and T2*
The longitudinal relaxation time (T1) of the rat brain tissue is prolonged at higher field strengths. T1, as measured with SE-EPI using variable TR, resulted in T1 (7.0 T) ≈ 1.7 s and T1 (11.7 T) ≈ 2.0 s, which is consistent with previous findings (de Graaf et al., 2006). The TR of 3000 ms is a compromise between sufficiently relaxed magnetization and the need of an adequate temporal sampling for the chosen paradigm in an experimental scan window of ∼6 min.
Transversal relaxation rates (T2 and T2*) are shorter at a higher field. T2 and T2* of the rat brain tissue were measured with a MSME sequence and MGE sequence, respectively (Fig. 2A). T2 maps were quite homogeneous but showed some distortions at 7.0 tesla, most likely due to an imperfect shim at this system, whereas T2* maps also exhibited a median-distal decay of values in the cortex produced by edge artifacts between brain tissue and skull. T2 and T2* relaxation times were measured in ROIs of gray matter and in the complete corpus callosum, as well as in a small ROI in the primary somatosensory (S1) region remote from significant edge effect (Table 1). T2 relaxation times were 40–50 ms and T2* relaxation times were 20–30 ms at 7.0 T. Corresponding T2 and T2* values for 11.7 T were 30–40 ms and 10–20 ms, respectively. While T2 and T2* values at 11.7 T correspond well to expectations from other laboratories (de Graaf et al., 2006; Uludag et al., 2009), values at 7.0 T are slightly lower than reported ones, which results in a constant fraction of T2/T2* ≈ 2 for both field strengths.
T2 and T2* maps (n = 8, first study) (A), BOLD activation maps of SE- and GE-EPI sequences (B), and corresponding signal intensity time curves (C). A, Relaxation time maps with T2 and T2* values indicated by the color bars below the images. T2 maps are very homogeneous, while T2* maps show a somewhat larger gradient of T2* values of the cortex with values decreasing radially through the cortical surface. In the center of S1, T2* is 23 ± 3 ms at 7.0 T and 19 ± 3 ms at 11.7 T. B, SE-EPI and GE-EPI activation at 7.0 and 11.7 T with percentage BOLD signal indicated by the color bar below the images. Size and magnitude of BOLD activation were similar at 7.0 and 11.7 T, while a clear difference in activation size was observed between GE-EPI and SE-EPI. C, Corresponding signal intensity time curves for the activation areas depicted in B. The small bars beneath the signal time curve indicate the stimulation period while the longer bars mark the resting periods. For orientation, the baseline and the BOLD intensity at 2% are marked in color in the graphs.
T2 and T2* relaxation times in milliseconds for different regions of interest in the rat brain (n = 8, first study)
The largest differences in transversal relaxation times were observed in the corpus callosum, where T2 and T2* changed ∼40% with field strength, whereas changes in the center of S1 were only 27% for T2 and 17% for T2*.
Due to their long (single-shot) acquisition window, even SE-EPI images are prone to field inhomogeneities and may be partially T2* weighted. For this reason, EPI images were compared with classical SE and GE images and corresponding T2 and T2* maps. From their overall shape and lack of T2*-like signal dropouts, we conclude that the SE-EPI images obtained contain only very little T2* weighting, noticeable only in case of imperfect shimming at 7.0 T. The suitability for the GE/SE comparison at hand should not be affected by this, and BOLD activation and cluster size were indeed extremely different for SE- and GE-EPI.
Field comparison study with spin-echo EPI
All animals showed BOLD activation in each stimulation session. Figure 2B shows representative activation maps using SE-EPI (together with GE-EPI) for 7.0 and 11.7 T. Figure 2C presents the corresponding time courses. BOLD activation varied within a session as well as between sessions and between animals. Data of the field comparison (7.0 vs 11.7 T) of the BOLD response with SE-EPI within a single anesthesia session were averaged across animals and sessions. The values of the four variables (defined in Fig. 1C), MIN, AVG, MAX, and CS, are compiled in Table 2 for both field strengths and for different minimum thresholds on the BOLD signal, together with the corresponding value ratios (i.e., value at 11.7 T/value at 7.0 T). All neighboring pixels above the threshold (confidence level) in the S1 area contributed to the cluster. Three different thresholds on the pixel BOLD signal were applied, followed by corresponding data analysis, 0, 1, and 2%, to focus on different sizes of activation.
BOLD cluster parameters dependent on the threshold for BOLD signal in SE-EPI at 7.0 and 11.7 T
Without any threshold (0%), all pixels in the cluster contribute, and the following values have been obtained for the four cluster parameters: (1) CS = 13 ± 13 pixels (2.1 mm2) at 7.0 T and 18 ± 14 pixels (2.9 mm2) at 11.7 T, i.e., an increase of 44% with increasing field. The large SD was likely due to a non-Gaussian distribution, since clusters of 30–40 pixels were observed in some individual cases (Figs. 3 and 4A). This difference in CS, namely pixels that were part of the cluster at 11.7 T but not at 7.0 T, had a low BOLD signal <1% and lay at the border of the cluster. (2) MIN = 1.0 ± 0.3% at 7.0 T and 0.8 ± 0.4% at 11.7 T, i.e., a decrease of 22%. Whereas at 11.7 T SNR was good enough to detect small changes of 0.8%, at 7.0 T the SNR was too low to detect these pixels reliably. (3) AVG changed from 1.7 ± 0.5% at 7.0 T to 1.5 ± 0.6% at 11.7 T, i.e., a decrease of 10%. This is because AVG includes more pixels with low BOLD signal at 11.7 T due to the higher sensitivity. (4) MAX increased from 2.7 ± 1.6% at 7.0 T to 2.8 ± 1.6% at 11.7 T, i.e., an increase of only 4%.
Analysis of the BOLD response in the S1 of the rat brain using the largest cluster (at least four neighboring pixels) with activation (t test, 99% confidence level) in one slice (2 mm thick). Histograms of BOLD cluster parameters at 7.0 (top) and 11.7 T (bottom) are shown. MIN, AVG, and MAX BOLD signals as well as CS were averaged over all animals, sessions, and scans. Note: MIN and AVG at 11.7 T were clearly smaller than those at 7.0 T, while MAX and CS were similar.
Comparison between 7.0 and 11.7 T for SE-EPI (A) and the ratio of SE-EPI to GE-EPI (B). A, Cluster parameters MIN, AVG, MAX, and CS for 7.0 and 11.7 T with a lower threshold of 1% of the BOLD signal. Note: there was no significant difference between BOLD activation between field strengths. B, Ratio of cluster parameters obtained with SE-EPI and GE-EPI sequences. The ratio for all four parameters was similar at both field strengths. GE-EPI is more sensitive to low signal pixels (MIN ratio ≪ 1), and produces higher BOLD signal (MAX ratio ≫ 1); also, the cluster is approximately three times bigger (CS ratio = 3).
In most cases the AVG value was found to be slightly lower than the arithmetic mean of MIN and MAX, which implies that MIN and MAX, which are represented by a single voxel, are meaningful, separate parameters. Therefore, none of the three parameters MIN, AVG, and MAX, alone provides a good characterization of the BOLD signal across field strengths; only together are they useful (Fig. 3).
When using a BOLD signal threshold of 1% (which corresponds to an activation area approximately limited to the S1) or a threshold of 2% (corresponding to the center of the cluster activation), the cluster size is reduced at both field strengths. No significant differences for CS, MIN, AVG, or MAX between field strengths were found; no value showed a difference >10% when comparing 7.0 and 11.7 T.
As no difference was found between both field strengths using threshold analysis, the data were reanalyzed using a different approach (histogram analysis). BOLD% and fCNR were calculated for each pixel in a ROI in the cortex in both hemispheres, without any threshold. All pixels from all studies and all animals were filled into histograms (Fig. 5). While there is no clear difference between fCNR at 7.0 and 11.7 T for the ipsilateral and contralateral side, the distribution contralateral to the activation is much wider for BOLD% at 7.0 than at 11.7 T.
A, B, Histogram of the group analysis of BOLD percentage change (A) and functional CNR (B) for all voxels. Images within one study were averaged and BOLD and CNR were calculated afterward and filled into histograms. The 11.7 T data (orange) are located in the upper half, and 7.0 T data (blue) are in the lower half of the diagram. The filled curve represents the analysis contralateral to the activated hemisphere showing a Gaussian curve around zero. The bars represent the analysis at the activated hemisphere. For BOLD, the distribution of bars is smaller for 11.7 T because of better temporal SNR (A), while for fCNR both distributions are equal (B). For BOLD and fCNR, distribution of bars is almost identical.
Comparison study of spin-echo and gradient-echo EPI
SE-EPI and GE-EPI scans were recorded alternately within a single session performed at either 7.0 or 11.7 T (Figs. 2B,C, 4B). All animals showed BOLD activation in all scans. However, the size and strength of activation showed great variation between sessions and animals. Therefore, the ratio of GE over SE was calculated for each animal individually to reach greater stability in the BOLD response.
Table 3 shows the GE/SE ratio averaged over all sessions. For both field strengths the same trend was found: MIN ≈ 0.7, AVG ≈ 0.9, MAX ≈ 1.4, and CS ≈ 3.3. No significant differences for MIN, AVG, MAX, or CS were found between field strengths. These results indicate that using SE, MIN is higher, AVG the same, MAX is ∼2/3, and CS is ∼1/3 compared with using GE.
BOLD cluster parameter for SE-EPI and GE-EPI at 7.0 and 11.7 T averaged over four sessions
Statistical analysis
The power analysis delivered an effect size d = 0.92 (large effect) for a measured BOLD signal MAX = 2.76 ± 1.34% at 7.0 T and a theoretical MAX = 4.61 ± 1.49% (mean calculated, deviation taken from measured data), based on an assumed field-dependent BOLD increase by 67% for 11.7 T with a power = 1 − β = 0.98. The maximum increase in light of our results was an increase of 52% with field strength (p = 0.9, d = 0.72, MAX at 11.7 T = 4.2%). To reduce the detectable threshold increase, the number of sessions needed would be >40 for d = 0.72 (52% increase) and >60 for d = 0.5 (36% increase), with a probability of p = 0.99.
Discussion
We demonstrated for the first time that the BOLD signal increases no further with magnetic fields >7.0 T under equal experimental conditions at compared field strengths. A maximum increase of 52% is theoretically possible (but not detected!) with the data measured (n = 18, p = 0.90), which is far below expectations of 67% or higher. BOLD signal of 1.5–1.8% with SE-EPI at 7.0 and 11.7 T is in agreement with our previous results of 1.5 to 2.1% at 7.0 T (Weber et al., 2006; Weber et al., 2008), and slightly >1.25 ± 0.17% in medetomidine-sedated rats reported at 9.4 T (Zhao et al., 2008). Results of BOLD are still rare for 11.7 T and were performed in α-chloralose-anesthetized rats: 1.6–8.4% BOLD signal depending on resolution and cortical depth using a FLASH sequence (Silva and Koretsky, 2002), an average signal intensity increase of 7.5 ± 1.8% when using a different MRI protocol than the one in our study (Keilholz et al., 2004), a variation between 3.5 and 12% BOLD (depending on stimulation frequency) using the same MRI protocol with α-chloralose and different stimulation parameters (Sanganahalli et al., 2008, 2009), and 8.1 ± 2.4%, again using a different MRI protocol on a vertical bore magnet (Chen and Shen, 2006).
These results at 11.7 T, all under α-chloralose, show a wide variation in the BOLD intensity. Differences in animal strains, stimulation, and analysis strategies, particularly anesthesia and MRI protocols, make a direct comparison between these studies impossible. Furthermore, none of these groups reported an increase of the BOLD effect with a higher field.
While the nominal SNR increased by 78% from 7.0 to 11.7 T, the gain of temporal SNR in the EPI scans was only 33%. This apparently disappointing performance is explained by the presence of physiological noise, which is the dominating source of noise (Kalthoff et al., 2009) where a theoretical tSNR gain of only 40% is expected. Therefore, the widths of BOLD signal distribution of control and activation side are decreasing. However, neither fCNR nor BOLD signal could benefit from higher tSNR at 11.7 T.
Additionally, many of the previous studies use high spatial resolution, while in the present investigation the resolution was kept lower to allow direct comparison to results from the 7.0 T system with limited resolution capability. We ourselves were able to record a high BOLD signal up to 10% using array coils and high-resolution imaging at 11.7 T by exploiting increased SNR and smaller partial volume effects (Seehafer et al., 2009). However, using the same low-resolution parameters and more conventional detection systems at both field strengths no increase was observed at 11.7 T having higher SNR and higher field strength. Low-resolution images at 7.0 T show already the complete activated area with high SNR and therefore no increase with 11.7 T is expected.
Four BOLD components: model of extravascular, intravascular, dynamic, and static effects
The four-component model (Hoogenraad et al., 2001) is described by extravascular (EV) and intravascular (IV) and dynamic and static contributions to the observed BOLD effect (Norris, 2006). EV and IV spins interact with the susceptibility gradient of the magnetic dipole of the hemoglobin iron complex. In the static case, the dominant process is the frequency offset. This is detectable with GE sequences sensitive to local frequency offsets, while SE sequences refocus this offset. In the dynamic case, the frequency offset becomes dependent on spin motion and diffusion: spins experience different frequencies during dephasing and rephasing and are thus not refocused by SE. Therefore, these dynamic effects are detectable with both GE and SE sequences. When the diffusion becomes stronger, there can be local motional averaging that can decrease the field inhomogeneities experienced by the spins and lead to longer relaxation times (Kennan et al., 1994); under such conditions, GE-EPI would not detect dynamic effects. In summary, the BOLD signal behavior is characterized by four components up to a moderate level of diffusion, according to Norris (2006) as described below.
EV static (GE only)
Brain tissue around vessels with diameter >20 μm experiences a constant susceptibility gradient and therefore different frequencies.
EV dynamic (SE and potentially GE)
Diffusion in brain tissue around vessels smaller <20 μm leads to different frequencies with time. Spins cannot be refocused.
IV static (GE only)
Spins with different distances to vessel border and hemoglobin in the direct neighborhood experience different frequencies.
IV dynamic (SE and potentially GE)
Diffusion, movement, and interaction lead to different frequencies with time. Spins cannot be refocused anymore.
These four BOLD components are field dependent. In consequence, their fractional contribution to the total signal intensity changes with field strength. At low field, IV effects are dominant due to long transversal relaxation times of blood, while EV effects play a minor role due to small susceptibility gradients. EV components increase with field and become dominant, while IV effects become negligible because of the short transversal relaxation times of blood (Lee et al., 1999; Duong et al., 2003; Zhao et al., 2006). Consequently, at high fields only the static and dynamic EV effects contribute significantly to the BOLD signal. Since SE-EPI visualizes exclusively the EV dynamic effect, we conclude from our results that EV dynamic effects are no longer increasing between 7.0 and 11.7 T. Since GE visualizes EV static contribution (dynamic effects were already found to be constant based on SE-EPI results), the difference in BOLD signal when comparing SE- and GE-EPI provides information about the EV static effects at high-field strengths.
Comparison of SE-EPI and GE-EPI
We have shown that the SE-EPI BOLD signal is not increased from 11.7 to 7.0 T and that EV dynamic effects do not therefore increase further beyond 7.0 T. To analyze the EV static effects, we also performed GE-EPI at 7.0 and 11.7 T. Here, we compared the ratio of the BOLD signal for both imaging sequences, GE and SE, separately for both field strengths but within the same session to avoid intersession variability in BOLD activation. We found a constant ratio of GE/SE for both field strengths and therefore conclude that GE-EPI-based BOLD is also not increased.
At both field strengths, GE-EPI shows a lower detection limit (MIN) and an increased MAX BOLD signal; AVG is increased by a factor of 1.4. Bandettini et al. (1994) found a GE/SE ratio of 1.87 ± 0.40 of average signal difference at 1.5 T in the primary motor cortex and concluded that the ratio may be closer to 1 at higher field strengths. At 4.0 T, a factor of 1.4 in the number of activated pixels was reported (Cohen et al., 2004).
Some authors argue that the dynamic averaging will no longer pertain to very high-field strength, even for very small vessels, meaning that SE-EPI will show only minor clusters compared with GE-EPI (Di Salle et al., 2003; Norris, 2003). However, SE-EPI, although responsible for only one-third of the activation in our study, showed a similar BOLD increase in the activation core at both of the field strengths studied.
No general increase of BOLD signal per se
Explanation for the lack of continuing BOLD increase at very high-field strength comes from a theoretical BOLD model (Uludag et al., 2009) covering a field regime up to 16.4 T. Here, IV and EV BOLD components were analyzed according to contributions from arteries, arterioles, capillaries, venules, and veins: the field dependence of EV effects around capillaries and venules is increasing, but it is decreasing around venules for magnetic field strength values >4.0 T. Depending on tissue mixture, similar BOLD signals are expected for 7.0 and 11.7 T in their model, in agreement with the present experimental finding. Using a numerical vascular network model calculation, Fujita (Fujita, 2001) predicted, in full agreement with our present results, no pronounced BOLD increase between 7.0 and 11.7 T for SE-EPI, independent of vessel diameter. For GE-EPI, the model expects a field-dependent increase only for very small vessel diameters of 3 μm, while larger vessels show no field dependence for GE-EPI in the relevant field range. Therefore, in accordance with those model calculations, a GE-EPI BOLD increase across the relevant field range can only be expected for high spatial resolution to emphasize the contribution of very small vessels. However, with our spatial resolution of 400 × 400 × 2000 μm3, the rather large voxels will have a strong weighting toward vessel diameters >3 μm, in correspondence with vessel size determinations by Troprès et al. (2001, 2004), who reported diameters of 5 μm or more in rat brain based on both their experimental data and Monte Carlo calculations. These spatial conditions therefore support also the lack of BOLD field dependence for GE-EPI of our data.
Conclusions
Based on our results, BOLD signal per se does not increase with field strengths above 7.0 to 11.7 tesla. Furthermore, sensitivity to the different vascular compartments changes at even higher fields, and the underlying biophysical modeling has to be refined. We conclude from our findings that fMRI studies can only profit from these ultra-high fields if protocols are accordingly optimized and SNR gain from higher field is efficiently invested, e.g., into higher spatial resolution.
Footnotes
-
The work was supported by the German Federal Ministry of Education and Research Network “Stem cell-based regeneration after stroke” (01GN0509) and the European Union-funded Networks of Excellence of European Molecular Imaging Laboratories (LSHC-CT-2004-503569) and Diagnostic Molecular Imaging (LSHB-CT-2005-512146). We thank Drs. David Norris (F. C. Donders Center for Cognitive Neuroimaging, Nijmegen, The Netherlands), Kâmil Uludağ (Max Planck Institute for Biological Cybernetics, Tübingen, Germany), Pedro Ramos-Cabrer (Clinical Neurosciences Research Laboratory, Santiago de Compostela, Spain), and Prof. Oliver Speck (Otto-von-Guericke University, Magdeburg, Germany) for scientific discussion, Anne K. Rehme and Dr. Christian Grefkes for critical discussions concerning the statistical analysis, and Andreas Beyrau for help with animal preparation.
- Correspondence should be addressed to Prof. Dr. Mathias Hoehn, In-Vivo NMR Laboratory, Max Planck Institute for Neurological Research, Gleueler Straße 50, Cologne D-50931, Germany. mathias{at}nf.mpg.de
