|Home | About | Journals | Submit | Contact Us | Français|
Spectroscopic imaging of the human head at short TEs (≤15ms) typically requires suppression of signals from extracerebral tissues. However, at 7, decreasing efficiency in generation (Hz/Watt) and increasing spectral bandwidth result in dramatic increases in power deposition and increased chemical shift registration artifacts for conventional gradient based in-plane localization. In this work, we describe a novel method using RF shimming and an 8 element transceiver array to generate a field distribution that excites a ring about the periphery of the head and leaves central brain regions largely unaffected. We have used this novel distribution to provide in-plane outer volume suppression (>98% suppression of extracerebral lipids) without the use of gradients. This novel distribution is used in conjunction with a double inversion recovery method to provide suppression of extracerebral resonances with T1s greater than 400ms, while having negligible effect on metabolite ratios of cerebral resonances with T1s >1000ms. Despite the use of two adiabatic pulses, the high efficiency of the ring distribution allows RF power deposition to be limited to 3–4W for a TR of 1.5s. The short TE enabled the acquisition of images of the human brain displaying glutamate, glutamine, macromolecules and other major cerebral metabolites.
Ultra high field (≥7T) in vivo magnetic resonance spectroscopy and spectroscopic imaging confers the intrinsic advantages of increased SNR, spectral resolution and spectral simplification for J coupled resonances. Although these systems have been available since the late 1990s (1,2), there have been few reports of their use in spectroscopic imaging (SI) studies (3–5). This limitation is largely due to the inherent disadvantages of high field that include: 1) a roughly linear increase in required power for equivalent generation; 2) a linear increase in the required and bandwidth to minimize chemical shift dispersion errors and 3) decreasing T2s, requiring short TEs to minimize T2 losses and preserve intensity from J-modulating compounds such as glutamate and glutamine. Taken together, these effects result in a ~12 fold increase in power deposition for equivalent performance at 7T in comparison to 3T.
Spectroscopic imaging of the human head at short TEs (≤15ms) typically requires suppression of signals from extracerebral tissues. Conventionally, this is achieved in one of three ways: 1) in–plane selection to selectively excite or refocus regions of interest (6,7); 2) outer volume suppression using slice selective excitation pulses and gradient dephasing to suppress regions not of interest (8); and 3) non-spatially selective inversion recovery pre-pulses to suppress short T1 lipid resonances. When in–plane volume localization is used for spectroscopic imaging at 7T, the available and power deposition limit the achievable bandwidth. Due to the need to excite a large brain region (>10cm), the gradient selection strength is small, resulting in chemical shift mis-registration errors of up to 45% (5). Although the smaller spatial extent required for outer-volume suppression allows for higher gradient selection strengths, the large number of pulses required to conform to the exterior of the brain, 4–8 pulses, can result in substantial power deposition and lengthy application times. Alternatively, the short T1 resonances of extracerebral lipids can be suppressed using non-spatially selective inversion recovery methods (4,9). Although this last approach can eliminate chemical shift mis-registration errors, it typically results in relatively long TRs (2–3 seconds) and substantial signal intensity reduction (40–60%, depending upon TR and T1) for brain metabolite resonances.
Transceiver arrays (10,11), where the amplitude and phase of RF delivered by each coil can be adjusted, enable dramatically different spatial distributions to be generated (11) without using gradients. This creates the possibility for T1 and chemical shift independent outer volume suppression without gradient pulses, thereby reducing power deposition, required repetition times and preserving sensitivity for cerebral metabolites. Specifically, a novel distribution that excites a “ring” about the periphery of the head is generated using principles of RF shimming and an 8 element transceiver array. Unlike conventional outer volume suppression sequences that typically use 4–8 pulses to generate a band of suppression about the head (8,12), the elliptical symmetry of the ring distribution can be made to conform to the shape of the head, substantially reducing the number of pulses required. The ring distribution can be used to suppress resonances outside the brain while minimally perturbing resonances within the brain. We have used this sequence to acquire short TE (15ms) spectroscopic images of the human brain at 7T, with moderate TRs (1.5s), minimal power deposition (1–1.3W/kg) while retaining near full sensitivity for cerebral metabolites.
As described, this work focuses entirely on methods of RF shimming, so as to shape the spatial distribution of the relevant transmission field, . To simplify the notation throughout the remainder of this work we will refer to as simply B1 The B1 generated at any given point within the sample from n independent RF coils driven simultaneously can be expressed as
where the amplitude (B1j(r)) and phase (j(r)) are both functions of spatial coordinates determined by coil geometry and loading conditions, and aj and 0j are amplitude scaling and phase parameters set by the user. Since j(r) and B1j(r) vary throughout the sample for each coil and aj and 0j are spatially invariant, the phase of the RF generated by each coil cannot be perfectly coherent over all regions of the sample simultaneously (11). Thus, phase cancellation of the B1 will occur in some regions, reducing the amplitude of the B1 generated. The degree of phase coherence or relative efficiency (fraction of B1 generated divided by maximal B1 if all coils were “in phase”) is given by
such that “perfect” constructive interference (η(r) =1) occurs at a given location when all phase terms j(r)+0j are equal. Likewise, perfect suppression of the excitation field occurs when η(r)=0.
To achieve adequate lipid suppression with in-plane localization, the excitation and/or refocusing B1 distributions should be constructed to not perturb the extracerebral tissues (η≈0) while retaining efficient full excitation/refocusing from the brain interior (η≈1). Unfortunately, due to the B1 profiles of the surface coils and their phase properties, the condition under which optimal phase coherence in the center of the object is achieved, η~1, typically results in substantial intensity at the periphery of the object (11). Simultaneous cancellation of the B1 fields immediately adjacent to all coils is difficult, since the B1 field immediately adjacent to a coil is dominated by that coil. However, it is possible to perform the reverse, i.e. cancel the fields from the brain interior (η≈0) while retaining significant excitation from the periphery of the head (η≈1) by choice of the phase of the RF. This situation is ideal for outer volume suppression. Thus, outer volume suppression and efficient excitation/refocusing can be obtained if different B1 distributions are used for in-plane and slice selective localization, respectively.
The relative efficiency, η, in the center of the brain, can be maximized if the mean value of transmit phase over the desired ROI, ( , Fig 1B), is the same for all coils. Thus, if , then from Eqs.  and 
Typically is not a single value. However, if is relatively small and centrally located where the phase is changing slowly, will also be small such that if , then . Using this condition to set 0j, the B1 homogeneity can then be optimized over most of the brain, ( , Fig. 1C) using a least squares optimization of the amplitude coefficients, aj. In this case, aj are chosen to minimize the difference between the calculated B1 and the target B1.
For outer volume suppression, ideally the B1 should be constrained to a “ring” about the periphery of the head, (Fig 1D), with a central region, (Fig 1E) where . Since the B1 field from any individual coil at the center of the brain will be non-zero, the net B1 field from all coils over can only be cancelled if the phases of the applied RF are selected to result in destructive interference. For example, for locations , where the B1 amplitudes generated by two coils k and l are similar, akB1k(r′) ~ alB1l(r′), cancellation of the B1 field, η(r′)~0, is obtained for
Since transceiver arrays typically use more than 2 coils (n=8 in this application), as long as the individual B1 amplitudes) are similar and the phases are distributed over at least 2π radians, cancellation (i.e., ), will also generally occur. Thus, B1 distributions with low η over central brain regions can be generated when the RF is applied with
Note that ΔRing=0 generates the homogeneous mode. However, selection of ΔRing = 2mπ/n, leaves only the amplitude coefficients aj as free parameters to generate the desired B1 distribution over .
At locations immediately adjacent to each coil ( ), the net B1 field is largely dominated by the closest coil, such that B1closest(r″) B1j(r″) j ≠ closest and
Under this condition, the phase of the applied RF has minimal effect on the B1(r″). This allows the phases , used to generate the null over to be retained unmodified, while the coefficients aj are used to optimize the B1 over . In reality, these are idealized approximations and the coefficients aj and ΔRing do affect η(r) and B1(r) over both and . Notably, the cancellation of B1 from using incremental phase shifts ΔRing, is qualitatively similar to the central B1 nulls seen in some of the higher order modes of transverse electromagnetic and birdcage coils (13,14).
All data were acquired with a Varian Direct Drive system and a head only (68cm ID) actively shielded 7T magnet. The system uses a Magnex head gradient insert equipped with a complete set of 2nd and 3rd order shims, with each of the shims powered by two 10A supplies (Resonance Research Inc.). For the transceiver array, 8 fully independent transmit channels were used to drive eight 1kW RF amplifiers (Control Power Corporation). The T/R network consisted of 8 preamplifiers and 8 home built PIN diode single-pole double throw (SPDT) switches each capable of handling 1 kW of RF.
The transceiver phased array consists of eight rectangular surface coils circumscribing the head. All the adjacent coils are decoupled inductively to a level of at least -18–20 dB when loaded with a human head. The array is split in two sections with the posterior (bottom) section having five surface coils and the anterior (top) section having three coils. Due to inductive decoupling, no electrical connection between the sections of the array is required. To accommodate different head sizes, we constructed two different array tops (Fig. 2A). The two array tops consist of three coils of 9cm×6.5cm or 9cm×7.4cm. When combined with the posterior coils (9cm×7.4cm, Fig. 2B), the vertical inner dimensions of the coils were 21 and 23 cm, respectively. The smaller array was used for studies in 3 subjects (2 women and 1 man) while the larger array was required in the studies of the other two men. Each coil was individually tuned and matched for each subject. No adjustment of the inductive decoupling was necessary for the different subjects. For the phantom study, a 23cm diameter circular transceiver array of similar length (Fig. 2C) consisting of eight 9cm×8cm coils was used to better match the geometry of the phantom.
B1 maps (phase and amplitude) of the individual coils (single coil transmitting) or the combined array (all coils transmitting simultaneously) were acquired using a double excitation method (15). For these studies, the B1 maps used a target B1 of 1 kHz and a nominal excitation pulse of 60°. The images were acquired at 64×64 resolution over a FOV of 192mm using a 5mm slice thickness to maximize SNR. After selecting a target slice using default parameters (equal power with incremental 45° phase shifts), the B1 of each coil was mapped with each coil transmitting separately (8 coils × 0.5s × 64 = 256s). To optimize the performance, the phase and amplitude of each coil was mapped.
To maximize η in the center of the coil, a circular ROI of 60mm diameter ( ) was selected to determine the phase (Fig. 1B). The mean phase, was calculated and 0j determined, . Using these phases the amplitude scaling factors aj were then optimized over a second ROI, (Fig 1C), encompassing most of the brain within the slice. The optimization was performed using a least squares algorithm to minimize the difference between the target B1 and the achieved B1.
The target performance of the ring distribution was defined using two ROIs, (Fig 1D) and (Fig 1E). The outermost ROI, the suppression ring , used a 20mm thick elliptical ring. Using a scout image, the vertical and horizontal dimensions of the ellipse were adjusted by the user to include extracerebral tissues (Fig. 1D). The target B1 for this ROI was set to 1 kHz. The horizontal and vertical dimensions of the second ROI, , was automatically set to 0.8 times the inner dimension of the suppression ring (Fig. 1e). The target B1 for this region was 0Hz. The region between the ring and ellipse is analogous to the “transition” band of a selective pulse, where the target performance for the optimization is not specified.
For phantom studies, ΔRing was set to π/2, 3π/4 and π. Based on the phantom studies, a phase increment of 3π/4 was selected as a compromise between obtaining some homogeneity over and minimizing RF within . After selecting ΔRing, the amplitudes aj of the eight coils were optimized using a least squares minimization of the difference between the calculated B1 field and the target B1s for each subject using their acquired B1 map.
To calculate maps of the relative SAR for the homogeneous and ring distributions, we used finite difference time domain simulations (16–18) with a high fidelity human head mesh model (XFdtd BioPro v. 7.0 Remcom Inc.). The array was modeled as constructed, with the only difference being the replacement of the inductive decoupling loops between individual coils with their lumped element equivalents (Fig. 6A and B; note that for clarity the shield is not shown). The head mesh grid is set to 4mm. Each coil in the array was tuned and matched to 300MHz by adjusting the capacitance as determined from Smith charts calculated when each coil was driven individually. Maps of B1, Bx + iBy, and the SAR were then calculated by driving all coils of the array simultaneously using the homogeneous and ring distributions.
Spectroscopic imaging data were acquired using the sequences shown in Fig. 3. In all of the sequences, planar localization is provided by a slice selective excitation pulse (10 mm thickness). Refocusing and water suppression is provided by a broad band semi-selective refocusing pulse (9). Additional water suppression is provided by a frequency selective adiabatic inversion pulse (±0.33ppm, 100Hz peak B1) followed by a recovery delay (TIRW) to null the water magnetization. In Figs. 3A, C and D, TIRW/TR was 420/1500ms (6.4 min) while in sequence 3b, TIRW/TR was 1200/3000 (12.8 min). Use of a 1.5s TR with a non-spatially selective inversion pulse would reduce the metabolite SNR by ~80%. The excitation, refocusing and selective water inversion pulses were applied using the homogeneous distribution. In Fig. 3a, no outer volume suppression is applied. In Fig. 3B, a single non-spatially selective adiabatic inversion pulse (homogeneous B1 distribution, RF#1) is applied with a 350ms inversion recovery delay (TIRL) to minimize lipids. In Fig. 3C, to minimize the SNR loss from cerebral metabolites within , the inversion pulse is delivered using the ring distribution (RF#2). Finally, in Fig. 3D, two inversion pulses and delays (TIR1/TIR2, 600/180ms) are applied with the ring distribution (RF#2) to provide a broader band of T1 suppression. Two dimensions of rectangular phase encoding (FOV=192×192) were used with either 16×16, for the sequence comparison studies, or 32×32 encodes for the high resolution study (25.6 min).
To provide phase and amplitude scaling for the reconstruction of the metabolite SI data, an unsuppressed water SI was acquired using the same sequence with the exceptions that: 1) the initial IR pulses and their associated recovery delays were deleted; 2) the broadband semi-selective pulse was replaced by a non-selective pulse (500us) and 3) the TR was shortened to 0.5S. The data acquired from each receiver coil for the water SI was then reconstructed using a 3DFFT (1D spectral, 2D spatial) including both spatial (cosine filter) and spectral (3Hz Gaussian) filtering. To minimize broad baseline components a convolution difference was applied using a line broadening of 100Hz. No other baseline correction method was applied. The water resonance from each pixel was then automatically phased and integrated to determine the phase and amplitude factors to reconstruct the metabolite SI. The relative noise from each coil was determined by automatically measuring the standard deviation of the noise from a region outside the brain. Spectroscopic images from each receive coil were reconstructed identically to that of the water reference image. The spectrum from each pixel from each coil was phased using the values determined from the water reference image. A single metabolite image was then generated by summing the phased spectra from each coil weighted by the SNR of the coil at that location (integral of the water signal divided by the noise of the coil).
Spectroscopic images of individual metabolites or rations of metabolites were then generated. To assess the distribution of NAA intensity throughout the brain and lipid contamination we integrated the resonance area between 1.92 and 2.12 ppm without any baseline correction or spectral fitting. As described, the NAA images are not corrected for differences in receiver sensitivity profile or performance of the pulse sequence. To demonstrate the ability to detect gray and white matter differences (19,20) we generated ratio images of creatine/NAA and glutamate/NAA. These images were calculated by fitting all image pixels in the spectral domain (21). To eliminate pixels outside of the brain where the calculated ratio reflects noise, those pixels having an integrated NAA resonance area below 5% of the maximum NAA resonance area within the brain were excluded. The spectroscopic images were interpolated to 256×256 resolution for display purposes.
To assess the extent that the ring distribution could be generated, we acquired data from a spherical phantom at 7T using fixed phase increments per coil. For this example, the phases and amplitudes of the applied RF were optimized as described for the homogeneous distribution (Fig 4A) and the ring distribution (Fig. 4B). Figure 4C displays the result of optimizing the homogeneous distribution over the phantom. In Figs 4D–4F the ΔRing was set to π/2, 3π/4, π and the amplitudes to the 8 coils were optimized using the target B1s. Displayed in Fig. 4G are line plots from a column through the center of the phantom. Table 1 lists characteristics of the B1 distributions and power requirements for the homogeneous distribution and ΔRing = π/2, 3π/4, π. For the ring distributions, the disparities between the target B1 of 1kHz and the achieved B1 is a result of simultaneous optimization of the B1 for both (1kHz) and (0kHz) with a narrow transition band between the two ROIs. Specifically, the solution is a mathematical compromise between too much B1 in and too little B1 in . As a whole, the effect of increasing ΔRing results in greater cancellation of the B1 from and restriction of significant B1 amplitude to . This comes at the cost of decreased B1 homogeneity over and increased applied power. The decreased homogeneity is due to B1 cancellation from regions approximately equidistant from two RF coils. Scaling the mean B1 values achieved across to 1kHz results in applied power levels of 736, 995 and 1912 W for ΔRing = π/2, 3π/4, and π, respectively. Notably the applied power used to drive the ring distribution with ΔRing =3π/4 at 1 kHz is ~60% lower than that required to drive the homogeneous distribution.
Fig. 5 displays the target ROIs (Fig 5A, B), the acquired B1 maps (Fig. 5C, D) and calculated relative efficiency maps η (Fig 5F, G) from an adult female volunteer. Fig 5E displays line plots of the B1 from Figs 5C, D. Table 2 lists the characteristics of the homogeneous and ring distributions averaged over five volunteers. In comparison to the phantom, the homogeneous distribution showed better homogeneity, versus 160Hz, and decreased peak power requirements, 1.9kW versus 2.4kW. For the ring distribution with ΔRing = 3π/4, the in vivo performance was similar in mean B1 in comparison to the phantom 706Hz versus 710Hz, but was achieved at a lower power level, 306±30W versus 502W. For a B1 value of 1 kHz, this equates to a power level of 613W. This remains more than a factor of 3 lower than the homogeneous distribution. As seen from Figs 5E and F, the efficiency for the ring distribution is much higher peripherally than the homogeneous distribution. Unlike the homogeneous distribution however, there is significant inhomogeneity across . To compensate for this effect, we used adiabatic inversion pulses that achieve >97.5% inversion for B1 values above 650Hz.
The ring distribution uses a dramatically different phase relationship between the coils, resulting in cancellation of the B1 in the interior of the brain. However, the extent to which the electric field is also cancelled or modulated requires numerical simulation. To determine the relative SAR maps for the homogeneous and ring distributions, we calculated B1 and SAR using a human head model (Fig 6A, B). Although the RF shield surrounding the array has been removed to better visualize the relationship between the RF coils and the head, in Fig 6A, it was in place for all simulations. The B1 and SAR maps for the homogenous mode are calculated using summed inputs to all coils of 1V for B1 maps and 1W for SAR maps. To mimic the conditions used in the human brain, the summed input power to all coils for the ring distribution was set to 16% of the homogeneous distribution (i.e. 0.4V for B1 maps and 0.16W for SAR maps). Figs. 6C and E show the B1 and SAR maps of the homogeneous distribution. The SAR map is scaled to 1W of applied power. Consistent with the in vivo data, the B1 map shows slightly increased B1 at the center and the periphery of the brain. Consistent with our in vivo results, the ring distribution shows B1 values at the surface of the head which are similar to the homogeneous distribution despite the much lower power applied for the ring distribution (Fig 6C, D). The SAR maps demonstrate that the ring distribution also generates similar levels of SAR at the periphery of the head as the homogeneous distribution and much lower SAR in central brain regions with (Fig 6E, F).
To compare the efficiency of lipid suppression over the peripheral ring and signal retention in the target ROI we acquired four consecutive spectroscopic images from each of five volunteers without removing them from the magnet. The data (Fig. 7) were acquired using the sequences depicted in Fig 3. In the first acquisition, we acquired data without any outer volume suppression (Fig. 3A) to provide a reference for calculating the percent suppression of extracerebral components. These data are presented at 1/10th vertical scale. At TE=15ms, the extracerebral lipid signals are large and contaminate spectra from interior brain locations due to the point spread function of the SI acquisition. In the second and third studies, a single adiabatic inversion pulse was applied with a TIRL of 350ms. The adiabatic inversion pulse was applied using either the homogeneous (sequence Fig. 3B, Fig. 7B) or the ring distributions (sequence Fig. 3C, Fig. 7C) with TRs of 3.0 and 1.5s, respectively. Use of the single IR dramatically decreases the extracerebral lipid signal, minimizing contamination of spectra from within the brain. Despite the reduction in TR from 3.0 (Fig 7B) to 1.5s (Fig 7C), there is a 50–100% increase in signal amplitude of the spectral data acquired with the ring distribution. This increase in signal amplitude (despite a halving of the TR) is due to the spatial specificity of the ring distribution. Theoretically, the single non-spatially selective IR and delay reduces the signal amplitude by ~60%.
In both Figs 7B and 7C (single IR) substantial lipid intensity outside the brain remains. This residual intensity arises from the fact that extracerebral lipids have multiple resonances with a range in T1 values (22) and the inversion is incomplete (B1<650Hz) in some regions with the ring distribution. Use of two inversions pulses and delays can substantially extend the T1 range of suppression (23) and reduce the effects of incomplete inversion. Displayed in Fig 8 are simulations of the predicted signal using single (TIRL/TR=350/3000ms, broken line) and double IR (TIR1/TIR2/TR=600/180/1500ms, solid line) sequences. The double IR provides a much broader range of suppression with respect to different T1 values. Use of the double IR reduces the extracerebral components (Fig 7D) by an additional factor of 5–6, albeit at a loss of some signal intensity in brain regions that are in the transition regions or at the edge of . This is due to residual B1 in these regions.
To provide a quantitative estimate of the degree of suppression over the peripheral ROI, eight locations about the periphery of the brain (at approximately 45° increments) were selected and the resonance intensity between 0.7 and 1.7ppm was integrated (Ss) for the three outer volume suppression methods (sequences Figs. 3B–D). These values were normalized by the integrated intensity, Sns, in the non-suppressed acquisition (sequence Fig 3A), thereby generating a mean suppression factor (Sns/Ss) for each volunteer. The average values from the volunteers were pooled to provide group statistics. In comparison to the homogeneous distribution, the ring distribution results in a decrease in suppression factors from 18±3 to 9±1. However, since the sensitivity for the metabolites increases by 1.5–2.0 fold when the ring distribution is used, this results in minimal increase in relative lipid contamination. Use of the double inversion recovery substantially improves the suppression for the ring distribution by ~6 fold, to 58±20 consistent with suppression of a much larger range of T1 values (solid line Fig 8). Although the use of two inversion pulses does increase the average deposited power, the average power deposition for a 1.5S TR was 3–4W. Assuming the head represents a 3kg load this equates to ~1–1.3W/kg, well below the FDA guidelines of 3W/kg.
As described, the amount of signal retained from within will be a function of both T1 and B1. Figure 9 displays a contour plot of the calculated signal retention as a function of T1 and B1 due to application of the double IR sequence. The signal value has been normalized by the calculated intensity for the same TR, 1.5s, omitting the inversion pulses. For B1 values below 50 and 100Hz, greater than 81 and 95% respectively of the signal is retained for all T1 values less than 5 seconds. Perhaps even more significantly, the contour lines for the region where T1>1s and B1<300Hz show remarkably little T1 dependence. Thus, although the intensity of cerebral metabolites will be altered by the spatial dependence of the residual B1 within , the ratios of those metabolites to each other is relatively unchanged.
To determine the in vivo signal enhancement, we integrated the signal from 1.87–2.17ppm (predominantly NAA) from all pixels contained within . In each case, only pixels whose nominal volume was entirely within the ROI were included in that group (Fig. 5). The values from each pixel were normalized using the data acquired with a single inversion pulse delivered with the homogeneous B1 distribution (Fig. 3B), as a reference. Despite halving the acquisition time, the use of the ring distributions resulted in mean increases of 68±12% (single inversion, Fig. 3C) and 49±15% (double inversion, Fig. 3D) respectively over . As expected, the more interior the location, the lower the B1 value, the greater the enhancement in signal.
Based on the substantial increase in SNR and reduction in required acquisition time we acquired a high resolution data set using 32×32 encodes (FOV=192mm×192mm) with a slice thickness of 10mm (nominal voxel of 0.36ml). Displayed in Figs 10A–D are a scout image, an NAA intensity image and ratio images of creatine/NAA and glutamate/NAA. Increased creatine/NAA and glutamate/NAA from gray matter is seen from gray matter regions along the intrahemispheric fissure and the cortical periphery. Displayed in 10E are spectra from 5 representative locations within and a scout image. Despite the small voxel size, excellent SNR is obtained throughout the volume, allowing visualization of glutamate, glutamine in addition to the major resonances of NAA, creatine and choline. Additionally, resonances from macromolecules (24) are now present in the range from 1.0 to 2.0ppm in Fig. 6C. These resonances have similar T1s to that of the methylene lipid resonance and are suppressed when non-spatially selective IRs are used (9).
Independent parallel transmission and the use of transceiver arrays has opened new avenues for achieving volume localization. Recently, methods for multi-dimensional in-plane localization using simultaneous transmission from multiple independent coils have been described (25–27). These methods achieve spatially selective excitation/refocusing by modulating the spatial distribution of the RF temporally in conjunction with gradient modulation. Although recent work has demonstrated that large tip angles can be created (28,29), their use for spectroscopy is limited by the strong off-resonance dependence of the pulses (28). In our approach, we build on the existing intrinsic geometry of the transceiver array to develop specific B1 distributions for homogeneous excitation/refocusing and outer-volume suppression. Since the instantaneous spatial distribution of B1 during any individual pulse is constant for refocusing and outer-volume suppression and no gradients are utilized; off-resonance effects are eliminated. Although the excitation pulse is slice selective (3kHz bandwidth, 10mm thickness), this results in only a 1mm shift between the volumes of NAA and creatine.
In the human head, differences in size and shape can result in variability in the relationship between the distance from the coil and the region of desired suppression. This will tend to reduce efficiency and increase power deposition. To reduce this effect we utilized an elliptically shaped split transceiver array with two different tops, which allowed the anterior-posterior size of the coil to be varied between 21 and 23cm. To compensate for the remaining residual differences, we also optimized the amplitude of the B1 distributions using a least squares method based on B1 maps acquired from each individual. Using this method we were able to obtain relatively consistent performance across five subjects studied (3 studies in the 21cm coil and 2 studies in the 23cm coil) for both the homogeneous and ring distributions.
As described, the combination of coil size and geometry along with amplitude optimization enables a relatively homogeneous distribution over , to be attained with relatively high efficiency. However, optimization of the ring distribution reflects a compromise between minimizing B1 spillage into and attaining a mean value of 1 kHz over . Use of an adiabatic pulse with a threshold for B1 compensation of 650Hz, largely compensates for the failure to reach 1 kHz over ROIouter. In this work we have chosen to be fixed relative to , having dimensions 0.8 times that of the inner aspect of . It is likely that significant improvements in the ring B1 distribution could be gained if the geometry, size and placement of the transceiver coils generating the ring distribution were optimized to generate a more rapid profile of B1 decay into the sample. To maintain efficiency in generation of the homogeneous distribution, it will likely require that separate arrays be used for each distribution. Thus, multiple layers of actively detuneable coils providing the homogeneous or ring distributions at different points within the sequence could provide substantial improvement.
In this report, we have utilized both single and double inversion recovery methods to suppress extracerebral signals. The single ring IR method has the advantage of preserving the greatest amount of metabolite signal throughout the target ROI, but it also achieves the lowest suppression factor. This decreased efficiency is due to a combination of some residual B1 homogeneity and the presence of lipid groups (methyl, methylene, etc.) with very different T1s. Use of a double ring IR provides a large bandwidth of T1suppression (greater than 96% for T1s between 400 and 5000 ms), thereby improving suppression of extracerebral lipids. Additionally, by virtue of its improved T1 bandwidth, the pulse also reduces tissue water from regions within . At 7T the T1s for gray and white matter have been reported to be 1220 and 2132ms respectively (30). This is likely to be advantageous for suppression of water from poorly shimmed regions such as adjacent to the sinuses or ear canals that are more peripheral in location. Since the adiabatic pulses are broad banded in frequency (2 kHz band width), water suppression based on outer-volume suppression will be preserved in these poorly shimmed locations.
As described, the ring distribution minimizes the signal intensity losses incurred with the use of non-selective inversion recovery sequences. The mean increase in signal for for the single and double IR sequences in comparison to the sequence using a non-selective inversion pulse was 68 and 49% respectively across the 5 volunteers. Due to the B1 profile across , the signal intensity varies, with greater signal retention from more central locations and lower values from regions closer to . This is analogous to loss of signal due to the transition bands on slice selective excitation using conventional gradient selection. However, unlike slice selective localization methods, which can distort metabolite ratios by convoluting spatial and spectral frequencies, the absence of gradients eliminates this effect. Analysis of the T1/B1 dependence of the sequence (Fig 8) indicates that although spillage of B1 into will reduce the measured signal intensity, there will be minimal changes in metabolite ratios. Thus, a simple correction for B1 amplitude applied to the entire spectrum provides a reasonably accurate representation.
For these studies of a rostral brain location, a uniform thickness (20mm) for the suppression ring provided excellent suppression of extracerebral resonances. For a given coil geometry, the homogeneity across the ring and its spatial extent is largely determined by the phase increment per coil. However, for brain regions where variable thickness suppression widths are desirable (such as in the temporal lobes or frontal lobes) optimizing the phase increment per coil simultaneously with its amplitude should confer the ability to asymmetrically adjust B1 penetration and thus ring thickness.
In conclusion, we have demonstrated that outer volume suppression of unwanted resonances can be achieved without the use of gradients using principles of RF shimming with transceiver arrays. When used with two inversions and suitable delays, the extracerebral signal is reduced by a factor of 58. Despite some spillage of the RF into and halved TR, substantial SNR gains were obtained with the ring distribution, in comparison to the non-spatially selective IR equivalent. Notably, the ring distribution, due to its higher η in peripheral brain regions, also reduces the peak power by a factor of 6 for the inversion pulses, allowing the total average applied power to be reduced to 3–4W. Maps of the SAR indicate that the ring (0.16W input) and homogeneous (1.0W input) distributions result in similar local SAR levels at the periphery of the head (scalp and muscle), while in the interior of the brain the ring distribution results in dramatically lower SAR levels. Thus, in this particular case, the calculated SAR per kHz of B1 generated in the regions of greatest exposure, the periphery of the brain, are similar for the two distributions, despite dramatic differences in overall applied power levels and B1 distributions. Finally, it is anticipated that significant improvements in performance of the ring distribution can be achieved by using coil configurations optimized for the ring distribution.
Funding provided by R01-EB-009871.