Transmit and Receive Transmission Line Arrays for 7 Tesla Parallel Imaging

Transcription

1 Magnetic Resonance in Medicine 53: (2005) Transmit and Receive Transmission Line Arrays for 7 Tesla Parallel Imaging Gregor Adriany, 1 * Pierre-Francois Van de Moortele, 1 Florian Wiesinger, 2 Steen Moeller, 1 John P. Strupp, 1 Peter Andersen, 1 Carl Snyder, 1 Xiaoliang Zhang, 1 Wei Chen, 1 Klaas P. Pruessmann, 2 Peter Boesiger, 2 Tommy Vaughan, 1 and Ka mil Uğurbil 1 Transceive array coils, capable of RF transmission and independent signal reception, were developed for parallel, 1 H imaging applications in the human head at 7 T (300 MHz). The coils combine the advantages of high-frequency properties of transmission lines with classic MR coil design. Because of the short wavelength at the 1 H frequency at 300 MHz, these coils were straightforward to build and decouple. The sensitivity profiles of individual coils were highly asymmetric, as expected at this high frequency; however, the summed images from all coils were relatively uniform over the whole brain. Data were obtained with four- and eight-channel transceive arrays built using a loop configuration and compared to arrays built from straight stripline transmission lines. With both the four- and the eightchannel arrays, parallel imaging with sensitivity encoding with high reduction numbers was feasible at 7Tinthehuman head. A one-dimensional reduction factor of 4 was robustly achieved with an average g value of 1.25 with the eight-channel transmit/ receive coils. Magn Reson Med 53: , Wiley-Liss, Inc. Key words: RF coil; parallel imaging array; transmission line coil; high field; 7 T 1 Center for Magnetic Resonance Research, Department of Radiology, School of Medicine, University of Minnesota, Minneapolis, Minnesota. 2 Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland. Grant sponsor: NIH; Grant numbers: P41 RR08079, EB 00331, EB 00513, CA 76535; Grant sponsor: W. M. Keck Foundation; Grant sponsor: MIND Institute. The 7-T system acquisition was funded by the Keck Foundation; Grant numbers: NSF , NIH S10 RR1395 *Correspondence to: Gregor Adriany, Center for Magnetic Resonance Research, 2021 Sixth Street S.E., Minneapolis, MN gregor@cmrr.umn.edu Received 13 October 2003; revised 20 August 2004; accepted 20 August DOI /mrm Published online in Wiley InterScience ( Wiley-Liss, Inc. 434 High magnetic fields provide several advantages in human brain imaging. However, they also pose significant challenges in image acquisition, most notably with rapid imaging techniques, such as EPI or SPIRAL. Magnetic field inhomogeneities caused by the susceptibility differences between air-filled cavities and tissue increase with higher magnetic fields. Consequently, images obtained with EPI or SPIRAL suffer from increased distortions or blurring, respectively, as well as signal loss due to the shorter T 2 *. In many ultra-high-field studies to date, such as imaging of human brain function at 7 T, the difficulties encountered in rapid imaging have been alleviated by restricting the image field of view (FOV) to a small local region along the phase encode direction (1,2) or by extensively segmenting the data acquisition. Both of these approaches have significant drawbacks; the first is unable to provide coverage over the entire or large portions of the sample, and the latter leads to long acquisition times and image artifacts. Another solution to the problems encountered with fast acquisitions at high magnetic fields is the use of recently introduced parallel imaging strategies based on multicoil arrays (3,4). This approach is potentially an attractive alternative because high fields are expected to improve parallel imaging performance due to the more complex sensitivity profiles of each coil element and the increased signal-to-noise ratio (SNR) (e.g., (5 7)). However, the design of high-frequency RF coils for such applications poses challenges and has not been extensively explored. A number of recent papers comprehensively described the design of receive coils optimized for parallel imaging in heart and brain studies at clinically used magnetic fields (8 10). These works utilized a single volume transmit coil for spin excitation and a separate array of multiple local surface coils for signal reception. However, this approach is not necessarily appropriate for ultra high fields. While high-field body RF coils have been reported (11), imaging often relies on head gradient coils where space is not easily available for large volume transmit coils. In addition, with large transmit coils that generate a uniform RF field in the absence of a load, the B 1 profile becomes highly nonuniform with the human load, with some regions demonstrating impractically weak B 1 magnitude (12). Arrays with the ability to transmit and receive with the same coil structure can avoid or minimize these problems. One possible approach for such a transceive array is to build coils that have the ability to switch between a homogeneous transmit mode and a surface coil type receive mode (13,14). Another even more promising approach at ultra high fields is to build coils that allow additionally for independent phase and amplitude control of their elements and thus support RF shimming methods (11,15,16). Furthermore, such an array immediately can be used for transmit parallel imaging applications (17) and can also be combined with additional receive-only coil arrays. There are a number of challenges confronted when designing dedicated transceive coil arrays at ultra high fields. The strong coupling to the sample at high frequencies mediates interactions between the separate coils, making it more difficult to electromagnetically decouple them (18). There are image inhomogeneities, caused partly by interaction with conductive tissue and partly by the differences between reception and excitation profiles (19 21); these inhomogeneities may aid parallel imaging strategies but pose problems for constructing coil arrays that cover the

2 Transceive Arrays for Ultra-High-Field Parallel Imaging 435 FIG. 1. Sketches and photographs of the stripline loop arrays. (a) The four-channel coil with a loop size of cm each and a gap of 2.5 cm between coils. (b) The eight-channel coil with a loop size of 7 14 cm each and a gap of 1 cm between coils. (c) Schematic of the eight-loop coil indicating the position of the feed point tune and match capacitors and the distributed capacitor in the conductor loop. (d) Detail of the feed point for the loop coils. brain. It is also well established that the RF power requirements increase with field strength and frequency due to the frequency dependence of losses (12,18). Particularly prominent at ultra high frequencies are increased losses to the sample and radiation. Sample losses are for the most part unavoidable and thus can be contained to a degree only by targeting the coil dimensions to the desired FOV and by utilizing distributed resonance elements such as transmission lines to minimize E-field losses. Radiation losses are best addressed by coil designs that incorporate a ground plane or RF shield into the resonance structure. For volume resonators, this leads to the preference of designs such as a shielded birdcage, where the shield is typically electrically floating relative to the resonant structure to allow freely flowing mirror currents (22). An even more appropriate design for ultra high fields is the TEM resonator (15,23 25). Here, the radiating fields are well contained within a resonant cavity, and additionally the shield is an integral part of the primary resonance loop. Similarly, ultra-high-field surface coils or receive coil arrays can be built with an RF ground plane in close proximity (26,27) or a ground plane as an integral part of the resonant structure in the form of transmission line elements (15,28 31). The main challenge in implementing any array is to find sufficient means to electromagnetically decouple the individual coils. This becomes particularly difficult in transceive arrays, since combining low noise preamplifiers with transforming circuitry to significantly reduce coil coupling, often referred to as preamplifier decoupling (32), is not easily feasible. Alternative methods for decoupling, such as compensating for the next neighbor mutual coupling by introduction of a capacitor or inductor (33) or by decoupling networks (34,35), are, however, possible. Another decoupling possibility is to utilize the broadband decoupling characteristic of transmission line elements (29). This, combined with the radiation-related benefits that transmission lines posses at high frequencies, makes them an appealing building block for ultra-high-field transceive array coils. In this paper, we describe and compare two designs of transceive parallel coil arrays using strip transmission line (referred to as stripline for short in the rest of the paper) elements resonant at 300 MHz. The first design uses a stripline execution of the decoupled, multichannel loop array. The second design makes use of a linear stripline array, described previously as a multichannel TEM volume coil with decoupled elements (15). Imaging results obtained with these coils at 7 T (300 MHz) in phantoms and in the human head are shown. MATERIALS AND METHODS Coil Construction Four- and eight-channel transceive coil arrays were built according to stripline transmission line principles. Each loop in the four-channel loop coils was 13 cm wide and 12 cm long (along the main magnetic field direction), with 2.5-cm intercoil spacing and orientation in the 0, 90, 180, and 270 positions around an elliptical former. The dimensions of the elliptical former were 20 cm in the short axis

3 436 Adriany et al. FIG. 2. (a,b) Photographs and sketches of the stripline transmission line array with eight channels, an element length of 16 cm each, and a gap between resonance elements of 9 cm. and 24 cm in the long axis. To improve patient access and task presentation, the top coil of the four-coil array was built to be mounted flexibly (Fig. 1a). The eight-channel loop coil array was built from 7 cm wide and 14 cm long (along the z direction) stripline loops that were evenly spaced with 1.5-cm intercoil spacing (Fig. 1b). To compare the benefits of surface coil loop arrays versus linear strip arrays, eight-element straight line stripline arrays of 16 cm length were constructed from 5- and 12-mm Teflon substrate (Fig. 2). All substrates used for the coil arrays were polytetrafluoroethylene (PTFE) (Gapi, Clayton, OH) sheets, with a low loss tangent and a permittivity of 2.1. The width of the striplines was chosen according to standard microstrip design formulas to achieve close to 50- nominal line resistance (36). All 5-mm substrate coils were built using 12-mm-wide copper tape for the coil conductors and 20-mm-wide copper tape for the ground conductor. For the loop type arrays, all ground conductors were cut in at least one position and shorted by 330-pF capacitors to avoid eddy currents (Fig. 1d). This capacitor, however, represents low impedance for high frequencies and thus allows for RF currents to flow in the ground conductor. Bench measurements were performed using a calibrated Hewlett-Packard (Palo Alto, CA) HP 4396A network analyzer together with an 85046A S parameter test set. Imaging experiments were performed on a 7-T magnet (Magnex Scientific, UK) equipped with a Varian Inova console (Palo Alto, CA) and Siemens Symphony/Harmony gradient amplifier (Erlangen, Germany). We imaged healthy volunteers who had signed a written consent form approved by the Institutional Review Board of the University of Minnesota. Gradient echo scout images were obtained in sagittal, coronal, and axial orientation (TR/ TE,18 msec/5 msec). Inversion recovery turbo flash images were obtained with a nonselective adiabatic inversion pulse (TI 1.45 sec; recovery time 5 sec; TR/TE, 18 msec/ 5 msec; flip angle, 10 o ). We utilized a single 4-kW RF amplifier (CPC, Brentwood, NY) and then split the RF power with equal amplitude and phase with either a fouror eight-way power splitter (Werlatone, Brewster, NY). The four-way power splitter had 0.26-dB insertion loss per path and less than 1 phase variation; the eight-way splitter had 0.4-dB insertion loss and 1 phase variation. Transmit phase increments for each channel were then adjusted for optimal image homogeneity by altering the cable length in the transmit path. The resulting phase increments between neighboring coils for the four-element coil were The phase increments between neigh- FIG. 3. Schematics of the setup for the eight-channel experiments.

4 Transceive Arrays for Ultra-High-Field Parallel Imaging 437 boring coils in the eight-element coils was 45 but had a higher variation ( 10 ). For the data presented in this paper, this adjustment was done once for each coil as long as the coil was used for human measurements. Improvements can be achieved by performing small adjustments on every individual; this is planned for the future. Small differences in phase (not exceeding 10 ) were necessary between the phantoms and the human head for optimal performance. The decoupling among the different array elements was not affected by these phase adjustments, as expected. Transmit/receive switches (Varian, Inc., Palo Alto, CA) with low insertion loss of 0.2 db in each transmit path blocked transmitter noise during reception and enabled the use of low noise preamplifiers in close vicinity to the magnet bore. All four-channel experiments utilized the Varian Inova four-receiver channels. All eight-channel experiments were performed using a digital receiver system that was developed in-house; this receiver used an Echotek (Huntsville, AL) ECDR-814 board to oversample the 20-MHz IF at a rate of 64 MHz and 14-bit resolution, allowing for quadrature detection, band pass filtering, and down conversion to be done digitally (Fig. 3). B 1 Profile The B 1 profile generated by the coils in the human head during RF transmission was mapped at 7 T using magnetization preparation followed by ultra fast gradient recalled echo imaging ( turboflash ) with a small flip angle, as employed previously (12). Magnetization preparation was accomplished with a variable-duration hard (square) pulse ranging from 0 to 2100 sec (100- sec steps) as before (12) or a variable amplitude (1-dB steps), constant duration slice-selective sinc pulse followed by rapid gradient spoiling to eliminate transverse magnetization. The latter suppresses problems encountered due to magnetic field inhomogeneities if the shimming in the imaged slice is not good; otherwise the two techniques yield similar results. In order to minimize saturation effects due to rapid pulsing relative to T 1, the resolution was restricted along the phase encoding direction (thus reducing the total number of small flip angle excitations), the acquisition was segmented, and within each segment, the k-space was covered starting from the center to the periphery of the k-space, alternating positive and negative phase encoding steps. The last strategy avoids excessive saturation prior to acquisition of the central k-space data. After the magnetization preparation pulse, the resulting longitudinal magnetization is directly proportional to cos r B 1 d, 0 where and are the pulse duration and gyromagnetic ratio, respectively. The signal amplitude of an image acquired in this experiment will be affected by the transmit B 1 profile, the sensitivity profiles of each coil during reception, and the T 1 and T 2 * relaxation effects; however, the modulation of signal intensity in these experiments with changing RF power will be a function of excitation B 1 only. All B 1 calibration images were acquired in an axial orientation. SENSE Data Parallel imaging was performed using the SENSE approach as described in Ref. (4). Reduction in k-space acquisition was examined only in one dimension, namely the phase encoding direction, which was typically chosen along the X axis, as it corresponded to the shortest dimension of the head. Images were acquired in an axial plane, with a standard gradient recalled sequence (TR/TE, 16 msec/5 msec; flip angle, 10 at the center of the head; slice thickness 5 mm; one acquisition per phase encoding step (i.e., NT 1)). In each experiment, full field of view images were first obtained to assess the coil sensitivities. Then, separate reduced field of view images were acquired and reconstructed with the SENSE method. Full field of view data were acquired with a 256 (readout) 336 (phase) matrix. The 336 phase encoding steps were chosen in this study in order to attain reduction factors 2, 3, 4, and 8 on the same size data and to have an even number of phase encoding steps in the undersampled data. The field of view was 25.6 cm (readout) 20 cm (phase). Reduced field of view data were obtained by dividing both the field of view and the matrix size along the phase encoding direction by a factor R 2, 3, 4, 6, and 8. One full field of view image was acquired in about 5.4 sec, with the corresponding reduced field of view acquisition time being about 2.7, 1.8, 1.3, 0.9, and 0.7 sec. Additional acquisitions were obtained without RF pulsing in order to record noise data for all channels for the purpose of evaluating the noise correlation between channels. Signal to Noise To compare the signal to noise obtained with different coils, a gradient recalled axial image (TR/TE 5 sec/5 msec) was obtained with a matrix, a square field of view of 25.6 cm, 4 msec, 5 lobe [three positive (including the central lobe) and two negative] sinc excitation pulse, 70-KHz bandwidth, and a slice thickness of 5 mm. The choice for a TR of 5 sec was a tradeoff to allow for a large fraction of the longitudinal relaxation to occur, while limiting the total duration of the acquisition to reduce motion artifacts (21 min 20 sec with this setup). The flip angle was adjusted to reach 90 in the center of the slice (a B 1 calibration image series was obtained in each experiment prior to the signal intensity image). Noise data were also obtained without pulsing of the RF amplifier. Signal-tonoise ratio calculation was obtained following the method introduced in Ref. (37). Before this calculation, the signal of each receiver was rescaled with a normalization factor for each channel in order to compensate for gain imbalance (up to about 20 25% between channels in our system). The SNR measured in this way allows the evaluation of the relative performance of these coils with respect to each other. However, these SNR numbers do not take into account our specific receiver system noise figure or the impact of the pulse sequence. Especially since the dynamic range of the new multichannel receiver chain was not optimized, the SNR numbers obtained in the aforedescribed fashion cannot be taken as an absolute number and cannot be used for comparisons in an instrument-

5 438 Adriany et al. Table 1 Coil Decoupling, Quality Factors, B 1, and Signal-to-Noise Ratio Dependent on the Substrate Thickness H and Coil Type Coil type Substrate thickness H (mm) Gap between array elements (mm) Minimum isolation S 12 unloaded, without decoupling capacitors (db) Minimum isolation S 12 Unloaded, with decoupling capacitors (db) Minimum isolation S 12 with decoupling capacitors loaded with human head (db) Average Q 0 / Q 1 for all channels (mean SD) a (human head) Pulse width to achieve 90 in center of head with 1-kW square pulse (Mean SD) b SNR (human head, center) (mean SD) b Eight-stripline ( 0.1) sec (N 3) (N 3) Eight-stripline ( 0.1) sec (N 4) (N 4) Eight-Loop ( 0.15) sec (N 3) (N 3) Four-Loop ( 0.2) 350 sec (N 1) 467 (N 1) a The mean and SD are calculated from the data measured for each element separately. b Average and SD are given for the data obtained on N number of subjects. The value of N is given in parentheses. independent way. To overcome this problem, we performed additional measurements on one coil the eightelement stripline coil built on a 12-mm substrate to calculate an intrinsic signal-to-noise number (39,40). Since the SNR at higher frequencies is highly dependent on sample geometry and spatial location, we performed the measurement with a reproducible phantom and calculated the intrinsic SNR for a specific location. For this purpose, we utilized a 3000-mL spherical glass phantom (ACE Glass, Inc., Vineland, NJ) containing 100 mm NaCl. We choose such a simple phantom to ensure that our experimental setup can be easily replicated. The intrinsic signal to noise ratio was then derived from the reconstructed images as described in Ref. (40) using the formula SNR*F 2BW sin V N x N y NEX, [1] where we have added the term sin ( ) in the denominator since we used less than a 90 excitation in the ROI where the intrinsic SNR was calculated. In this expression, SNR is the observed signal-to-noise ratio in the final images, V is the sample or voxel volume in milliliters, N x and N y are the number of sample points, NEX is the number of excitations, F is the system noise figure, and 2BW is the full receiver bandwidth. The phantom was centered in the coil and at the isocenter of the magnet. We acquired one axial slice through the center of the sphere with a gradient echo sequence that minimized T 1 and T 2 effects on SNR. The sequence parameters were axial slice, 4 mm thickness; matrix size ; FOV cm 2 ; pixel volume ml; TE 6 msec (about 1% signal intensity loss only); TR 15 sec; flip angle 15 ; NEX 1; 2BW Hz. The RF power calibration was defined based on 90 flip angle at the center of the sphere. In order to avoid ADC overflow, while maintaining a receiver gain high enough to ensure a dynamic range that avoids noise truncation, it was necessary to reduce the RF power and use a flip angle of 15. To improve the accuracy of the pixel volume, we choose a seven-lobe sinc RF pulse, which, together with the relatively small flip angle, provided a better slice profile. The SNR was calculated from rooted sum-of-square images as described previously (37). A mean image SNR for one pixel (0.004 ml) was derived by averaging the SNR value within a pixel ROI located at the center of the phantom. RESULTS AND DISCUSSION Coil Decoupling Two means of minimizing the electromagnetic coupling of the individual coils to each other were pursued: broadband decoupling through the use of transmission lines and capacitive decoupling by introduction of decoupling capacitors between neighboring coils. First, the minimum substrate thickness and separation between adjacent coil elements were evaluated for avoiding resonance peak splitting in the different coil designs. In order to determine which substrate thickness represented the best compromise between sufficient B 1 penetration and high intercoil decoupling, coils with different substrate thickness were built, and a bench comparison was performed; some of the results from this comparison are summarized in Table 1. Prototype loop coils were initially built from 3- and 5-mm-thick substrate, consisting of only two resonance elements to simplify the experiment. The larger separation between resonance elements in the stripline linear arrays allowed for the use of a thicker, 12-mm substrate. The coils were tuned and matched, and then S 12 network analyzer measurements to evaluate coil decoupling were performed. When building the resonance structures using stripline technology with a 5-mm substrate, sufficient broadband coil decoupling to avoid peak splitting was feasible for the four-loop and the eight-linear strip coils without further decoupling means. The eight-loop coil, however, did require decoupling capacitors to avoid peak splitting and to attain sufficient decoupling for independent coil tuning and matching. For the loop coils, the separation between coils had to be chosen as a compromise between homogeneous coverage and broadband decoupling. A separation of 2.5 cm for the four-loop coil and 1.5 cm for the eight-loop coil was experimentally found to yield the best results. Since we were able to decouple all

6 Transceive Arrays for Ultra-High-Field Parallel Imaging 439 FIG. 4. The spatial distribution of the B 1 magnitude in the human head at 7 T. (a) The 20 images corresponding to an increase in (pulse length of a square pulse) starting from 0 (upper left) in 100- sec steps to 2 msec (lower right) as reflected in the modulation of image intensity according to cos( B 1 ) were acquired with the four-element loop array. (b) Analogous data, acquired with the eight-element 12-mm strip array where the pulse power was increased in 1-dB steps as described under Material and Methods. different coil types with 5 mm or thicker Teflon substrate, the 3-mm substrate was not further pursued due to the expected decreased B 1 penetration from coils built with this substrate. Additional introduction of lumped element decoupling capacitors between neighboring coils increased the coil decoupling in all coils. Decoupling in the four-channel coil was at least 25 db between neighboring coils and 35 db between opposing coils when loaded with a human head. We achieved similar values in the eight-channel loop coil, albeit using decoupling capacitors as an additional decoupling method. For the linear stripline type coils, we were able to achieve sufficient decoupling in the order of 18 to 20 db when loaded with the human head without resorting to decoupling capacitors. This is primarily due to the increased separation of 9 cm between the resonant elements. It was feasible to tune and match the array elements independently from one another for 50- match. The use of preamplifier decoupling strategies was not necessary. Since the physical length of the loop conductors in the four- and eight-loop coils was slightly longer than /2, we introduced lumped element serial capacitors in the coil conductor loop to achieve resonance and avoid coil current phase errors. This created a hybrid structure between a pure transmission line resonator and a surface coil, similar to that reported previously for planar strip arrays (e.g., (38)). We also found that the transition between the coaxial coil feed to the transmission line resonance structure did not require the typical extensive balun circuitry. This is, however, expected due to the continuous ground conductor between the coaxial feed line and the transmission line

7 440 Adriany et al. FIG. 5. The influence of phase adjustments on the B 1 transmit profile of the eight-channel strip coils is shown. The amplitudes for the eight transmit channels were kept equal. (a) A particular B 1 transmit map acquired as in Fig. 4 and together with the actual transmit phases for the indicated coil position. The phase measurement was performed with a calibrated network analyzer and includes all cable and T/R switch phases. (b) The same B 1 transmit map after the phase was altered in the three indicated positions. coil structure. We found this to be an important advantage in building arrays out of transmission line elements. B 1 Maps, Transmit Phase Adjustments, and Homogeneity The magnitude of the B 1 generated by the transceive array when used as a transmit coil was examined in the human head at 7 T as a function of spatial coordinates. The coil was used as a transmitter and a full k-space image was collected as described below. The data were received separately for each element; however, for calculation of the B 1 profile, the separate channels were combined simply as the square root of the sum of square of the magnitude images obtained from each channel. Bench measurements without a load, as well as imaging results at 7 T, showed that the B 1 penetration seemed not to be significantly altered if the Teflon substrate was at least 5 mm thick, which confirms a previously reported observation (30). The B 1 field maps of the four-loop (5-mm substrate), and eight-linear strip (12-mm substrate) transceive array coils were obtained and are illustrated in Fig. 4. The approach used for B 1 mapping relied on preparation of the longitudinal magnetization magnitude to reflect B 1 amplitude as described under Material and Methods and as used previously (12). The results for the center of the head are summarized in Table 1 in terms of pulse duration that would be required to achieve a 90 flip with a square pulse equivalent to 1 kw RF power. Figure 5 demonstrates the influence that transmit phase changes have on such B 1 field maps. In Figure 5, images were obtained with a power setting that resulted in a larger than 90 rotation by the magnetization preparation pulse in regions of the brain where the B 1 was highest (i.e., the center of the head); this central bright region was than surrounded by a dark band where the magnetization preparation pulse angle was 90, so that subsequent excitation pulse generated little or no detectable signal. Outside of these dark bands, toward the surface of the brain, the image signal intensity increased again as the magnetization preparation pulse became less than 90. Thus, in these images, the central dark band provides a convenient way to assess B 1 patterns in the head. Maladjustments of transmit phase in the order of 15 to 20 (or larger) noticeably altered the pattern of these dark regions (Fig. 5a); however, we found that it was possible to keep a standard phase setup (Fig. 5b) that yields acceptable homogeneity in all subjects. Small phase adjustments (approximately 10 or less) were necessary for phantom studies. While the coil design in principle allowed the use of different power input for the different elements, we limited all our experiments reported in this paper to an equal power distribution between the resonance elements. Figure 6 demonstrates the brain coverage that was achieved with the eight-element coils. As can be deduced directly from Fig. 4, the four-element coil attained relatively uniform B 1 field pattern throughout the head, with modest inhomogeneities becoming detectable only when the preparation pulse induced large rotation angles, approximately 180 or larger. For the eightchannel linear strip coil, B 1 generated in the transmit mode was stronger in the center than in the peripheral areas, similar to what is observed for the TEM coil with inductively coupled elements (12). However, there were differences as expected. With the four-channel loop array, peripheral areas were more efficiently excited relative to the center compared to the inductively coupled TEM. This can be observed in Fig. 4a where spin saturation was attained within 1- to 2-dB RF power differences for any location within this slice versus a similar data set acquired with a conventional TEM coil (12) that required more than FIG. 6. Demonstration of the brain coverage for the eight-element strip coils. (a) Sagittal scout image, FLASH. (b,c) Coronal and axial, IR Turbo FLASH images.

8 Transceive Arrays for Ultra-High-Field Parallel Imaging 441 FIG. 7. The spatial relationship between SNR and B 1 amplitude in the human head for an eightelement stripline coil (12-mm substrate thickness). (a) SNR values are measured as an average within a 2 2cm 2 ROI in five locations and are expressed as normalized to the ROI in the center. The absolute SNR in the center ROI averaged over several subjects is given in Table 1. The excitation pulse was 90 in the center. (b) The B 1 amplitude image was calculated from data obtained like Fig. 4b. The B 1 map was then normalized to the highest value, which was in the center of the brain. a 3-dB spread between the center and areas in the periphery. The eight-channel coils did require a 1- to2-db increase in RF transmit power compared to a TEM quadrature volume coil (12) when the flip angle was adjusted for the center of the brain. This difference in transmit efficiency is partly due to the additional splitter and RF front-end component losses in the array setup and partly due to the less effective coupling to the sample by the stripline array coils. Another indication for the weaker sample-coil coupling is the Q 0 /Q L ratio differences between these coils and a TEM coil where the elements are inductively coupled to each other; this ratio is between 2.8 and 4.4 for the stripline array coils (Table 1), compared to a ratio of 10 for an inductively coupled TEM volume coil (12). SNR SNR comparison of the coils among each other was obtained using essentially fully relaxed, gradient recalled echo (FLASH) images acquired from the human head. The SNR was measured as an average within a central pixel (2 2cm 2 ) region of interest, in an axial slice located typically superior to the ventricles. A similar, but not necessarily the exact slice, was aimed for in each different subject. Several subjects were measured except in the case of the four-loop coil. The results are presented in Table 1 as means standard deviation over all subjects. Within the ROI, the excitation pulse was adjusted to 90. For the different coils, the pulse-width for a fixed power was inversely correlated with SNR (Table 1) at the center of the head where these parameters were measured. The differences between the two eight-linear stripline coils were not highly significant, especially for SNR. However, the SNR of the eight-loop coil in the center was lower than that of the other two eight-element coils. This discrepancy is likely due to several fundamental differences inherent in the two designs. First, the loop design generates B 1 fields parallel to the static magnetic field (z-axis) because of current flow along the sections of the loops perpendicular to the z-axis; these B 1 components and the associated E-fields are a source of coupling to the sample without being useful for MR. Second, in the loop design, the cur- FIG. 8. (a h) Single channel sensitivities for the eight-channel strip (12-mm substrate). The individual sensitivity profiles were calculated from the individual full FOV images of each coil divided by the sum-of-squares image. Acquisition parameters were TR/TE, 16 msec/5 msec; flip angle, 10 at the center of the head; slice thickness 5 mm; 1 acquisition per phase encoding step (i.e., NT 1)). However, because of the normalization performed, the resulting data are independent of the spatial heterogeneities in flip angle, relaxation effects, and spin density.

9 442 Adriany et al. rents on the adjacent elements running along the z-direction oppose each other; therefore, the magnetic field from these elements tend to cancel each other except in close proximity to and in between the conductors, leading to longer pulses at fixed power to attain a 90 rotation. Third, in the eight-loop coil, elements necessarily come closer to each other, requiring capacitive decoupling. Finally, the length of the loops now gets to be longer than /2; therefore, unlike the strip coils, lumped capacitors must be employed in the coil conductor, generating a point of potentially increased E-field coupling. Thus, the conclusion is that the strip designs are superior to the eightelement loop configuration used in this study. The intrinsic SNR was also measured using a phantom for the eight-linear strip coil with 12-mm substrate, as described under Material and Methods. The average system noise figure F ( SD) for the eight-receiver channels was determined to be 1.59 ( 0.06) db. Using this system noise figure and Eq. [1], the intrinsic SNR for the eightlinear strip coil (12-mm substrate) in the center of the phantom was calculated to be ml -1 Hz 1/2 for a pixel ROI. Figure 7 illustrates the B 1 magnitude maps and SNR in different locations (both expressed relative to the center) acquired in one subject with the eight-linear stripline (12-mm-thick substrate) coil. The SNR data are from 2 2cm 2 ROIs, four in the periphery and one in the center. The central ROI was the same as the one used for the data given in Table 1. As described under Material and Methods, all SNR measurements, including the data illustrated in Figure 7, were obtained with the excitation pulse set to 90 in the center of the slice; the TR was 5 sec in order to allow for virtually full relaxation. The gray and white matter T 1 at7tis 1.9 and 1.4 sec, respectively. Thus, even in gray matter, magnetization will recover to 92% of its thermal equilibrium value 5 sec after a 90 pulse. This implies that the SNR measurements are essentially not confounded by relaxation. Therefore, in regions where the transmit B 1 was less than 90, the SNR reflects a lower value compared to what would be obtained if the excitation was adjusted to be 90 in that region as well. However, the SNR values are partially confounded by heterogeneity in the proton density of the different tissues and cerebral spinal fluid (CSF). Because of the proton density difference, white matter is a little darker even in fully relaxed images. Despite its higher proton density, CSF appears brighter than white matter and comparable to gray matter in intensity because it is partially saturated even at a TR of 5 sec. A correction for these effects was not made in the SNR data. In these coils, the transmit B 1 is strongest in the center and in regions very close to each element (Figs. 4 and 7). The SNR is, on the other hand, higher in the periphery. Unlike a single volume coil, such as a birdcage or an inductively coupled TEM coil operating in the same mode during transmission and reception, these arrays operate like a single volume coil during transmit and as independent surface coils during reception. Thus, the situation is not much different from transmitting with a single volume coil and receiving with a separate set of surface coils, except that in these designs, this is attained with the same physical structure. During transmission and spin excitation, the B 1 generated by the array results from the complex addition of the B 1 patterns of each element, adding constructively in some places or leading to partial cancellation in others (e.g., between the elements). Such a complex addition of signals is not performed in the receive mode, as it would be if the coil operated as a single volume coil during reception. Instead, images from each individual coil are recorded separately and combined subsequently in different ways, for example, as rooted sum of squares (as employed for SNR measurements in this study), addition of magnitudes, or SENSE. Sensitivity Profiles and Parallel Imaging The amplitudes of the sensitivity profiles of the different channels were calculated as s i x P i x / 2 P j x 1/2, [2] where P i refers to the measured complex signal in image space for coil i at pixel location x and j sums over all coils for that pixel. These sensitivity profiles are illustrated in Figure 8 for the eight-channel linear strip coil with the 12-mm substrate. The phase information from each channel is not shown even though it is utilized in the SENSE reconstruction based on these sensitivity maps (discussed later). Because of the normalization performed 1 in Eq. [2], these sensitivity maps are independent of the spatial heterogeneity in spin excitation (flip angle) as well as relaxation and proton density effects. It is apparent in each case that the sensitivity profile of individual elements is highly asymmetric at this field strength. This is in excellent agreement with our previous experimental data and simulations on surface coils (19). Figure 9 demonstrates the same data as in Figure 8 in a different format. In Figure 9, top row, all colored regions identify a location where s i x 0.7 (i.e., s i x 2 0.5) for one of the eight coils; the color code gives the actual value of s i x. Since s i x 2 1, the colored regions indicate areas dominated in signal amplitude by one coil over the combined sum of all the other coils (i.e., regions where s i 2 j i s 2 j x ). The bottom row is a similar plot except that it displays regions where s i x s j i x for any i and for all j; in other words, except for the central blue area, the color-coded regions depict territory where one coil dominates over any other coil (Fig. 9, bottom row). In the center, the signal amplitude of the array elements becomes comparable and no single coil dominates the sensitivity. The parallel imaging performance attained with these coils at 7Tinthehuman head is illustrated in Figure Note that P i x is directly proportional to x r x sin K B 1 x TR B 1i x where x denotes the spatial location, is the spin density, r accounts for relaxation effects, B 1 TR is a complex number describing the amplitude and phase of the excitation RF field generated by the entire array operating in the transmit mode (i.e. B 1 x TR ib 1i x ), K is a constant depending on pulse shape and duration, and B 1i is the reception profile of the ith element in the array. j

10 Transceive Arrays for Ultra-High-Field Parallel Imaging 443 FIG. 9. Combined sensitivity profiles for the four- and eightchannel transceiver array coils: (a and e) eight-linear strip (12-mm substrate) coil, (b and f) eightlinear strip (5-mm substrate), (c and g) eight-loop (5-mm substrate), (d and h) four-loop (5-mm substrate). The top rows (a d) display regions where individual coil sensitivities are larger then the combined sum of all others. White areas show where this statement cannot be achieved. The lower rows (e h) illustrates regions where individual coil sensitivity of one coil is larger than any other. The color coding is the normalized sensitivity where the square of the signal from all coils adds up to 1. FIG. 10. (a) SENSE reconstructed images and (b) corresponding g-factor maps for reduction factors of 1, 2, 4, and 6 (corresponding to maximum degree of aliasing, of 1, 2, 3, and 5, respectively) obtained with the eight-channel loop array. The grayscale was adjusted so that noise within the brain can be seen easily. This resulted in some areas of the image with high signal intensity becoming white. The intensity scales are identical for the different images and set so as to emphasize the noise amplification for increasing reduction factors. (c, d) The maximum and average mean geometry factors, respectively, for the four- and eight-channel loop arrays built with 5-mm substrate and for the eight-linear strip arrays built with 5- and 12-mm substrate, respectively. The geometry factors reported were obtained as an intermeasurement average from different subjects and different sessions with the same subject (two subjects and total of five independent experiments for the eight-loop coil; three subjects and a total of nine independent measurements for the eight-linear strip (5-mm substrate); three subjects and a total of four independent experiments for the eight-linear strip (12-mm substrate) coil.

11 444 Adriany et al. These are one-dimensional reductions in FOV along the phase encoding direction. Images obtained for reduction factors 1, 2, 4, and 6 with one of the coils (eight-loop) are illustrated in the top row (Fig. 10a). The second row of images (Fig. 10b) shows the corresponding g-factor maps. Note that the scale for g-factor maps is different for the different reduction factors and was adjusted to maximally display the g-factor values. The patterns seen in the g- factor map of R 2, where the maximum value on the g-factor scale is 1.06, would not be visible using the scale employed for R 6 where the scale is from 1 to 5. Figure 10c and d depicts average and maximal g-factor over the whole brain for the different coils as a function of reduction factor; the data plotted represent the mean and SD obtained from different subjects and the same subject studied for multiple times (the number of studies and subjects are given in the legend to Fig. 10). Using SENSE, we were able to achieve reduction factors of 3 in the human head, with the four-channel array with an average geometry factor of 1.41 when the FOV reduction was along the long axis (data not shown) and 1.65 when along the short axis. The eight-channel arrays achieved whole brain average g-factors of 1.25, 1.27, and 1.26 for a reduction factor of 4 (maximum aliasing of 3) for the eight-loop, eight-linear strip (5-mm substrate), and eight-linear strip (12-mm substrate), respectively. It should be emphasized that these reduction factors were not obtained on data that were extensively averaged to improve the SNR; rather they are for NT 1, low flip angle images that were acquired in 5.4 sec (for full k-space coverage) to sec (for R 4). Compared to reported 1.5- and 3-T data using the same SENSE methodology and similar number of coils (e.g., (9,10)), the results in the human head at 7 T appear to represent improvements, consistent with our studies on phantoms (5). However, a more rigorous comparison cannot be made from these data alone at this time due to different coil geometries and designs that are employed. CONCLUSIONS These studies demonstrate that multichannel parallel imaging coils that can act both as a single volume transmitter and multichannel receiver can be built easily for very high magnetic fields, such as 7 T. In fact, the short wavelength at this high-field magnitude allows the use of stripline approaches with closely coupled ground conductors, which in turn permit coil arrays where the individual coils can be decoupled from each other easily without having to resort to complex decoupling schemes involving low-impedance preamplifiers. This transceiver arrangement eliminates the need for the use of separate body or large head coil as a transmitter, thus simplifying the hardware requirements and demands for bore space. This arrangement also permits with ease the individual control of phase and amplitude of each channel for the transmit mode so that transmit SENSE or RF shimming can be implemented. In view of the strong interactions between the sample and the coil, the latter capability is a particularly useful feature at very high magnetic fields, as has been demonstrated previously in body imaging at 4 T (11). The resulting array coils can be used flexibly, combining the separate channels in different ways subsequent to acquisition. They can also be used in parallel imaging approaches such as SENSE with good reduction factors. REFERENCES 1. Pfeuffer J, Adriany G, Shmuel A, Yacoub E, Van De Moortele PF, Hu X, Ugurbil K. Perfusion-based high-resolution functional imaging in the human brain at 7 Tesla. Magn Reson Med 2002;47: Yacoub E, Duong TQ, Van De Moortele PF, Lindquist M, Adriany G, Kim SG, Ugurbil K, Hu X. Spin-echo fmri in humans using high spatial resolutions and high magnetic fields. Magn Reson Med 2003;49: Sodickson DK, Manning WJ. Simultaneous acquisition of spatial harmonics (SMASH): fast imaging with radiofrequency coil arrays. Magn Reson Med 1997;38: Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med 1999;42: Wiesinger F, Van de Moortele P-F, Adriany G, De Zanche N, Ugurbil K, Pruessmann KP. Parallel imaging performance as a function of field strength an experimental investigation using electrodynamic scaling. Magn Reson Med 2004;52: Ledden PJ, Duyn JH. Ultra-high frequency array performance: predicted effects of dielectric resonance. In: Proceedings of the 10th Annual Meeting of ISMRM, Honolulu, p Adriany G, Van de Moortele PF, Wiesinger F, Andersen P, Strupp J, Zhang X, Snyder CJ, Chen W, Pruessmann K, Boesiger P, Vaughan JT, Ugurbil K. Transceive stripline arrays for ultra high field parallel imaging applications. In: Proceedings of the 11th Annual Meeting of ISMRM, Toronto, Canada, p Griswold MA, Jakob PM, Edelman RR, Sodickson DK. A multicoil array designed for cardiac SMASH imaging. Magma 2000;10: Weiger M, Pruessmann KP, Leussler C, Röschmann P, Boesiger P. Specific coil design for SENSE: a six-element cardiac array. Magn Reson Med 2001;45: de Zwart JA, Ledden PJ, Kellman P, van Gelderen P, Duyn JH. Design of a SENSE-optimized high-sensitivity MRI receive coil for brain imaging. Magn Reson Med 2002;47: Vaughan JT, Adriany G, Snyder CJ, Bollinger L, Liu H, Tian J, Renz W, Ugurbil K. An efficient high frequency body coil for high field MRI. Magn Reson Med 2004;52: Vaughan JT, Garwood M, Collins CM, Liu W, DelaBarre L, Adriany G, Andersen P, Merkle H, Goebel R, Smith MB, Ugurbil K. 7T vs. 4T: RF power, homogeneity, and signal-to-noise comparison in head images. Magn Reson Med 2001;46: Leussler C, Stimma J, Röschmann P. The bandpass birdcage resonator modified as a coil array for simultaneous MR acquisition. In: Proceedings of the 5th Annual Meeting of ISMRM, Vancouver, BC, Canada, p King SB, Duensing R, Varosi S, Peterson D, Molyneaux DA. A four channel transceive phased array head coil for 3 T. In: Proceedings of the 9th Annual Meeting of ISMRM, Glasgow, Scotland, p Vaughan JT; RF Coil for imaging system. U.S. Patent 2003;6,633: Boskamp EB, Lee RF. Whole body LPSA transceive array with optimized transmit homogeneity. In: Proceedings of the 10th Annual Meeting of ISMRM, Honolulu, p Katscher U, Bornert P, Leussler C, van den Brink JS. Transmit SENSE. Magn Reson Med 2003;49: Keltner JR, Carlson JW, Roos MS, Wong ST, Wong TL, Budinger TF. Electromagnetic fields of surface coil in vivo NMR at high frequencies. Magn Reson Med 1991;22: Yang QX, Wang J, Zhang X, Collins CM, Smith MB, Liu H, Zhu XH, Vaughan JT, Ugurbil K, Chen W. Analysis of wave behavior in lossy dielectric samples at high field. Magn Reson Med 2002;47: Collins CM, Yang QX, Wang JH, Zhang X, Liu H, Michaeli S, Zhu XH, Adriany G, Vaughan JT, Anderson P, Merkle H, Ugurbil K, Smith MB, Chen W. Different excitation and reception distributions with a single-loop transmit-receive surface coil near a head-sized spherical phantom at 300 MHz. Magn Reson Med 2002;47: