Human magnetic resonance imaging (MRI) and spectroscopy (MRS) studies conducted at high magnetic fields operate in the regime where the ratio of object dimension L to the wavelength of the radio frequency (RF) field used (i.e. L/λ, the relative object size) is comparable to or greater than 1. In this regime, B1 phase and amplitude in the object imaged (e.g., human head and body) become highly complex and spatially nonuniform (see Refs.1–12 and references therein). These deleterious RF inhomogeneities can be mitigated using multielement transmit array systems that use a high number of independent channels for RF transmission (13–18) and support methods such as RF shimming (19, 20) and transmit sensitivity encoding (SENSE) applications (21–23). Furthermore, such multielement transmit arrays can be used for multielement signal reception and can even be combined with additional receive-only arrays by using preamplifier decoupling for all coils during signal reception (24–26). Previously, it has been demonstrated that it is possible to build up to 32-element, cylindrically arranged transmit array head coils based on transmission line elements at ultra high frequencies (9, 27, 28). Such arrays achieved sufficient electromagnetic decoupling, avoided resonance peak splitting, and maintained transmit efficiency (7, 27, 29). However, given the strong coupling between the sample and the coil at high RF frequencies, it was difficult to accomplish equalized individual resonance element performance for different subjects and/or varying head positions for a given subject in the RF coil array. One potential way to improve RF transmission efficiency and signal-to-noise ratio (SNR) is to design a coil that allows for adjustments of the coil geometry depending on the head size. This way, the filling factor can be optimized, the higher B1 field closer to the coil conductor can be consistently better utilized, and a more balanced spacing between the subject and the individual resonance elements can be achieved, leading to a more even unloaded to loaded Q ratio between the different resonance elements.
The biggest challenge when allowing for such adjustable coil geometries however, is to find an equally flexible solution for the element decoupling network (29–32). For linear transmission line elements, the most sensitive points for lumped element decoupling options are capacitors between neighboring strips at the feed ends of the conductors. In this fashion, a fraction of the feed current with the proper phase can be diverted into the neighboring resonance element to compensate for the mutual inductance. It has been demonstrated previously that decoupling capacitors placed between immediate neighboring transmission lines provide an effective means to achieve array element decoupling below approximately −12 dB between any two array elements (33–35). It has furthermore been shown that for fixed array element geometries, a one-time decoupling network adjustment was sufficient to achieve array element decoupling for most standard load situations. This is an important point since adjustments from subject to subject are not feasible for practical purposes.
Flexible coil geometries, however, do not immediately lend themselves to fixed decoupling networks. Here, we present a 16-element decoupled transceiver array that allows for flexibly adjustable coil geometry and additional design changes compared to previously described eight- and 16-channel transceiver array coils based on the same transmission line principles (7, 9). In addition to these design novelties, the present study reports for the first time parallel imaging performance comparisons on the human head at these ultrahigh magnetic fields with radially distributed 16-element transmission line arrays (TLAs). This new coil provides excellent parallel imaging capability and RF efficiency at 7T, both attributed to its adjustability and to the design of the individual elements. The coil array incorporates one possible solution that addresses the need for decoupling capacitor value adjustment depending on geometry; self adjusting capacitors, where the capacitor value is a function of the relative distance between the immediate neighbor elements. Results are presented on SNR and parallel imaging performance of the coil at 7T as a function of the proximity of the coil elements to the human head.
MATERIALS AND METHODS
Two versions of a 16-channel geometrically adjustable TLA were built as shown in Fig. 1. The resonance elements were manufactured from Cu-tape (3M, Minneapolis, MN, USA) and Teflon bars of 4 cm × 1.2 cm × 18 cm. All capacitors, including the variable tune and match capacitors (NMNT 12-6; Voltronic, NJ, USA) and high voltage ceramic chip capacitors (100E series; American Technical Ceramics, NY, USA) were embedded into the Teflon substrate and shielded to minimize E-field exposure. Shielding of the chip capacitor from the load was accomplished by soldering one lead of the ceramic chip capacitor right onto the ground plane and by extending the ground plane around the front of the Teflon bar. The ground conductor for each array element was 4-cm wide and electrically separated from neighboring elements. To further improve next neighbor decoupling between elements the ground plane was extended to cover both sides of the Teflon bars as shown in Fig. 2 (34). The length of this side extension was 15 cm and the width was 9 mm. These additional side ground planes are a major design difference as compared to previous eight- and 16-element transceiver coils we have described using transmission line principles (7, 9).
An opening that allows task presentation and reduces patient anxiety was created in front of the face by shortening either one or two elements to 8 cm in length as shown in Fig. 1a and b. The effective electrical length of the remaining resonance element conductor strips was 15 cm. A “flexible” adjustable patch capacitor network was built between neighboring elements to change decoupling capacitance according to the geometrical distance. The needed capacitive range for decoupling capacitors was determined in a preliminary bench study to be 2.2 pf ± 1 pf. Capacitive patches built from 2-mm-thick and 15-mm-wide Teflon substrate were attached to one side of each Teflon bar as shown in Fig. 2b. The neighboring Teflon bar element had an appropriate groove of similar dimensions to guide the 2-mm Teflon patch and allow for variability depending on distance between the elements. A 12-mm wide copper tape in the bottom of the groove was soldered to the output circuitry and to the conductor strip for each element as shown. The copper tape could be shaped in appropriate ways to generate capacitive functions that match the experimentally predetermined decoupling capacitor needed for various coil sizes. While the patch capacitor design allows the implementation of nonlinear decoupling capacitor functions, such functions were not utilized in this work. Simple linear decrease of the decoupling capacitor value scaled with the increase in physical element distance was found to be sufficient in praxis to achieve better than −14-dB coil decoupling.
Adjustable Holder Mechanics
The holder mechanics of the coil allow for rapid individual spatial readjustment of the single transmission line element by 2.5 cm radially (Figs. 1 and 2). Besides individual adjustment, all elements can also be moved together. This way, the “global” coil geometry can be changed between a minimal setting of 22 cm in the long axis (anterior-posterior) and 16 cm in the short axis (left-right) to a maximal geometry of 26 cm × 21 cm. The elements were evenly spaced within the holder for the smallest possible coil geometry (16 × 22 cm2). The coil can slide on the patient bed along the z-direction and is used in conjunction with a customized head holder to allow for patient positioning on the bed before bringing the coil over the head. The head holder is firmly attached to the table bed but allows for adjustments of the holder height along the y-axis by ± 2 cm. This way the subject can be centered in the coil depending on individual head size. The head holder has a firm part consisting of a 10-cm-wide and 18-cm-long curved section (r = 10 cm) built from ¼-inch Delrin to solidly support the subject's head independent of the coil holder. To allow for flexibility depending on individual head/coil size, this firm holder section was merged with a flexible part built from 1/16-inch Teflon. Foam cushion around the inside of the entire head holder ensure patient comfort and a minimal distance of 1.5 cm to 2 cm from the resonance elements.
Initial imaging experiments were performed on a 7T magnet (Magnex Scientific, Oxford, UK) equipped with a Varian Inova console (Palo Alto, CA, USA) and Siemens Symphony/Harmony gradient amplifier (Erlangen, Germany). These experiments were performed using a 32-channel digital receiver system that was developed in-house. Later, our 7T instrument was altered to a 7T Siemens console with 32 receive channels based on total imaging matrix (TIM) and Siemens Avanto body gradient technologies. With both the Varian and subsequent Siemens based consoles, an 8-kW CPC RF amplifier (Communication Power Corporation, Hauppauge, NY) was utilized and the RF transmit power was divided on the high power side utilizing a 16-channel equal amplitude splitter (Werlatone, Brewster, NY). All data presented here were acquired with equal RF transmit power per channel. The transmit phase increments for each channel were adjusted for optimal image homogeneity. T/R switches (Varian Inc., Palo Alto, CA, USA and Stark Contrast, Erlangen, Germany) in each of the 16 transmit paths blocked transmitter noise during reception and enabled the use of low noise preamplifiers in close proximity of the coil.
Bench measurements were performed using a calibrated Hewlett Packard HP 4396A network analyzer (Palo Alto, CA, USA) together with an 85046A “S” parameter test set.
In order to compare the parallel imaging performance of the above-described adjustable coil to that of fixed geometry designs two additional coils with fixed geometries were built. These additional comparison coils had the same substrate thickness, conductor width, and number of resonance elements. The smaller of the coils was an elliptical coil with the dimensions of 19.5 × 25 cm2, the larger a circular coil with an inner diameter of 32 cm.
Healthy volunteers who had signed a written consent form approved by the Institutional Review Board of the University of Minnesota were imaged. For all measurements, the subject's position in the head holder and inside the coil was kept consistent while the coil geometry was changed. However, it was necessary to retune the array elements after each such change in coil size to maintain high transmit efficiency. Gradient echo scout images were obtained in sagittal, coronal, and axial orientation (TR/TE = 18 ms/5 ms). T1-weighted images were obtained with inversion-recovery turbo fast low-angle shot (FLASH) using a nonselective adiabatic inversion pulse (TI = 1.45 s, recovery time = 5 s, TR/TE = 18 ms/5 ms, flip angle = ∼10°). With the Siemens 7T console, 3D magnetization prepared rapid gradient echo (MPRAGE) data were acquired (TR/TE = 3.5 s/2.88 ms, TI = 1.5 s, matrix size = 256 × 256, 1-mm3 isotropic voxels).
Parallel imaging performance was ascertained in terms of the g-factor as described in Ref.36. Images were acquired in an axial plane, with a standard gradient recalled sequence (TR/TE = 16 ms/5 ms, flip angle = 10° at the center of the head, slice thickness = 5 mm, 1 acquisition per phase encoding step). Full FOV data were acquired with a 256 (readout) times 336 (phase) matrix, with the readout in the anterior-posterior (y) direction. The FOV was 25.6 cm (readout) × 20 cm (phase) and was kept relatively tight to the heads. The 336 phase encoding steps were chosen in this study to attain 2D reduction factors 2 × 2, 3 × 3, and 4 × 4 on the same size data and to have an even number of phase encoding steps in corresponding undersampled data. The g-factor was estimated solely based on the sensitivity profiles, which were calculated from the FOV as the ratio of images from individual channels to the root-sum-of-squares image of all channels (36). An excessive FOV underestimates the g-factor relative to a tight FOV. For the estimation of comparably defined g-factor the FOV was therefore selected as tight as possible to the head, and 2D reduction factors of 2 × 2, 3 × 3, and 4 × 4 were evaluated in terms of maximal aliasing and mean and maximal g-factors (which are FOV-dependent). Additional acquisitions were performed without RF pulsing in order to record noise data for all channels for the purpose of evaluating the noise correlation between channels.
To be able to compare the SNR between the different coil sizes, for each coil size with a given volunteer the following datasets were acquired: transmit B1 mapping in a central axial slice, signal intensity measurement, and noise measurements. The transmit B1 maps were obtained as previously described (7). Signal intensity measurements were obtained with gradient echo images of 128 × 54 matrixes, TE = 3.8 ms, TR = 15 s, slice thickness = 4 mm, and FOV = 25.6 × 16.2 cm2. Noise measurements were obtained with no RF pulsing and identical parameters except for a TR of 50 ms. Based on the flip angle map derived from the B1 calibration, it was possible to establish normalized SNR maps, as described (7). The SNR was calculated using a central axial slice in four locations and for four coil sizes.
The introduction of a decoupling capacitor patches between neighboring coils close to the capacitive feed-points (Fig. 2b) was essential to avoid RF peak splitting while allowing for coil size changes (7, 29, 30, 33). The most difficult decoupling adjustment was found to be for the unloaded coil since a load either in the form of a spherical phantom (3 liters of 90 mM saline) or a human head consistently dampened coupling between neighboring coils. The value of the variable capacitive patches was thus first assessed on the bench in an S12 transmission experiment for the unloaded coil. For this, in an initial step, optimal decoupling capacitor values minimizing next neighbor coupling for four different coil geometries (small = 17 cm × 22 cm, medium = 18 cm × 23 cm, large = 19 cm × 24 cm, and extra large = 20 cm × 25 cm) were experimentally found. The resulting actual capacitor values of all capacitors in the decoupling network were then measured with an LCR meter (Fluke 6303A, Eindhoven, the Netherlands). For this measurement, the decoupling capacitors were unsoldered from the resonance circuitry. Last, the experimentally found decoupling capacitor values for four positions were realized with the proper copper width and overlap for the patch capacitors between the resonance elements. With various subject head sizes, it was consistently feasible to tune and match the array elements independently from one another for a 50-Ω match without changing the once-adjusted decoupling capacitor network. Furthermore, the use of preamplifier decoupling strategies was not necessary. Extending the shield around the edges of the Teflon substrate reduced the coupling between next nearest neighbors by an additional −3 dB for the largest coil setting and by −5 dB for the smallest coil geometry.
The complete noise correlation matrix was experimentally determined for each subject in a noise measurement inside the 7T magnet. Figure 3 shows a typical noise correlation matrix for four different coil sizes acquired in a 7T measurement with the same subject. The highest coupling typically occurs between next nearest neighbors and was on the order of −14 dB or better. It was observed that the location of the highest coupling does change depending on head geometry, coil size, and head position inside the coil, but that the preadjustments were sufficient to reduce this coupling to values better than −14 dB. For the geometrical change of the coil size, we did not have to change the head position inside the head holder. However, a retune and rematch after each change in coil geometry was found to be essential to maintain homogeneous excitation and good transmit efficiency. While such readjustments are acceptable within a research environment, they represent a limitation for clinical routine use of the coil.
Transmit Phase Adjustments
The transmit phases for all experiments were initially set to 22.5-degree phase increments between channels. We limited all our experiments reported in this work to an equal power distribution between the resonance elements and were still able to achieve good brain coverage and acceptable transmit homogeneity. For the smallest coil sizes, however, B1 shimming using phase adjustments (8) in one or two locations were consistently necessary to correct for local signal voids due to transmit phase cancellation in close proximity to the resonance elements. Since the noise correlation matrices did not typically indicate a breakdown in coil decoupling for such locations we attributed the presence of the signal voids to the unique loading conditions and the relative closeness to the resonance elements. The effect that the distance between the coil and the sample has on such destructive B1 interferences is part of further in-depth investigation (37). Such effects can be corrected in a single location through a phase adjustment, as demonstrated in Fig. 4.
Figure 5 summarizes the power requirements for three coil sizes and the same subject when the phases of the RF input are simply set to differ by 360/16 degrees between adjacent elements. As perhaps expected, the 16-channel adjustable coil in the larger size setting required an increase in RF transmit power. The coil size had a clear influence on the power requirements to achieve a 90° spin flip in the periphery, as measured in a peripheral region of interest (ROI) located in the occipital lobe (see indicated area in Fig. 5) but significantly less in the center of the head. Consequently, we consistently found that the center vs. periphery ratio of the transmit power pulse length to attain a 90° spin flip angle was strongly dependent on the coil size. Depending on the head size, an optimal coil geometry could be found that improved the transmit power requirements between periphery and center. Note for example the similar pulse length of 487 μs and 464 μs for the center and periphery to achieve a 90° flip angle with a 1-kW square pulse when using a 19 × 23.8 cm2 coil size (Fig. 5) and furthermore the reversal of the ratio for the larger setting in which the pulse length for the periphery is longer than for the center. This is encouraging and points to the possibility to positively affect the transmit homogeneity when utilizing an experimental setup that allows for flexibility in RF pulse amplitude and shape. It is important to note, however, that the adjustable TLA driven with 360/16 phase increments and equal amplitude does not solve known significant B1 efficiency variations at 7T, such as in the lower temporal lobe. This is perhaps not entirely unexpected, since the B1 fields that can be generated with an adjustable transmission line coil are similar to transverse electromagnetic (TEM) volume coils in Mode 1 and birdcage volume coils where the problem were reported. One elegant solution for imaging the temporal and occipital lobe was offered by Wiggins et al. (38), when he was able to address the problem by utilizing two modes of a birdcage coil. The issue of homogeneous whole-head excitation at 7T, on the other hand, still remains to be addressed and deserves further evaluation, and is a part of an ongoing research effort utilizing transmit arrays with a higher number of resonance elements and gains in B1 field control (39).
The SNR comparison for the adjustable coil, for various sizes was obtained from the subjects, and one representative example of such a data set is discussed and presented here. For each of the subjects the coil size was altered up to four times per session to determine the relationship between coil size, transmit efficiency, and SNR. Figure 6a and b summarizes, for one of the subjects and four different holder sizes, the experimentally measured SNR attained under the data acquisition conditions used. The SNR was normalized to the value measured in the top (i.e., anterior) ROI in the smallest coil size setting. We evaluated the influence that the head size has on the average SNR. Our group of volunteers had relative similar head shapes and varied in size by 1.2 cm in the anterior-posterior direction and 0.5 cm in the left-right direction as assessed using the central sagittal and axial slice from the scout images. Comparing average SNR for the central axial slice for the largest head with that of the smallest head for the respective smallest possible coil setting, we found an increase in average SNR for the smaller head to be 10.7%. Consistently, it was found that the SNR increased for smaller coil sizes in all indicated regions of the brain (anterior, middle, lateral, and posterior). On average the SNR changed by 14% for a coil geometry change of 2 cm between the larger coil setting and the smaller setting (Fig. 6b). The largest SNR as well as the largest change in SNR with size (28% going from size 1 to 4) was typically observed for all the subjects in an area of the frontal cortex that is closest to the shorter resonance elements. This can perhaps be explained by the relative greater influence of an increased distance between sample and coil, and a lower B1 penetration of the shorter coil elements.
Parallel Imaging Performance Comparison
The parallel imaging performance, evaluated in terms of the geometry factor, was compared between the 16-channel geometrically adjustable decoupled TLA and fixed geometry decoupled TLAs of similar length and substrate thickness but with different inner coil diameters. Table 1 shows the significant improvements in parallel imaging performance as indicated by maximum and average achievable g-factors when using close fitting adjustable arrays compared to either a fixed geometry array of similar size or a fixed geometry array of significantly larger size. The reduction factor R in this case corresponded to the maximal aliasing. Using SENSE (36) and the adjustable geometry array, for example, we were able to achieve a 2D reduction factors of 3 × 4 in the human head, with an excellent maximum g-factor of 1.73 compared to 3.11 and 9.74 for larger fixed geometry coils. The corresponding improvements for the mean g-factor were 1.31 compared with 1.80 and 3.10, respectively. Since the coil sizes for the adjustable and the smaller of the two fixed size coils were comparable, the difference must be ascribed to other changes in the coil design rather than size alone. In this case, the only substantial difference was the utilization of a three-sided ground-plane for each element. However, coil size was found to have an effect as well since a clear parallel imaging performance increase was observed when comparing the two smaller coils (∼19 cm × 24 cm) with the larger TLA (32 cm inner diameter). This latter point has potential implications for transmit/receive coil configurations primarily intended for transmit parallel imaging applications; namely that, for a given number of array elements, if a larger transmit coil is used to accommodate a receive coil it will not be as efficient in parallel transmit applications. Our volunteers reported significantly improved patient comfort because of the individual resonance elements separated with open space between them as opposed to using conductive elements attached to solid formers as in the other coils used in this work for comparison, and the “open face design” with shorter elements covering the frontal lobes. The resulting tailored 16-channel coil allowed us to cover the entire brain including the cerebellum with good sensitivity, as can be observed in Fig. 7. Images in Fig. 7a were deliberately not corrected for intensity variations in order to give a fair picture of the field inhomogeneities that one can expect at 7T for such close fitting coils. The 3D MPRAGE imaging data shown in Fig. 7b demonstrates how these noticeable intensity variations in 7T whole-head images can be corrected with appropriate intensity correction algorithms (40). The resulting intensity corrected images demonstrate that excellent gray/white matter contrast can be achieved throughout the brain.
Table 1. 2D Parallel Imaging Performance Comparison Between Three 16-Channel Transmission Line Arrays Built With Similar 12-mm Teflon Substrate Thickness and Resonance Element Length But Different Coil Sizes*
Large circular coil 16-channel TLA continuous Teflon cylinder with one-sided ground plane (circular: 32 cm i.d.)
In all cases the FOV is set tight in both the phase-encoding and readout direction.
i.d. = inner diameter.
2 × 2
3 × 3
3 × 4
4 × 4
A new geometrically adjustable transceiver array for brain imaging based on 16 decoupled linear transmission line elements is presented in this work. This design follows our earlier work with eight- (7) and 16-element (9) transceiver arrays. A novel mechanical holder was developed that allowed for defined adjustments of all the elements together as well as individual elements. Furthermore, we introduced means to achieve self-adjusting decoupling capacitor values depending on geometry. This is essential to operate the coil under realistic conditions, which do not allow for time consuming subject to subject changes of the decoupling capacitor values. Individual match and tune for the elements, however, are essential for good coil efficiency and SNR. It was found that array element decoupling on the order of −14 dB or higher could consistently be achieved while allowing for flexible array coil geometries. When comparing the novel adjustable transceiver coil to a fixed geometry transceiver array, clear gains in peripheral SNR, transmit efficiency, and, in particular, parallel imaging performance were observed for the new closer-fitting coil. One important aspect of the adjustable design for future evaluation is if the gains in transmit efficiency can offset the potentially higher local specific absorption rate (SAR) values that are expected for such closer fitting designs. The most noticeable improvement compared to previous 16-channel TLAs was the clear gain in parallel imaging performance. This gain can be foremost attributed to a three-sided RF ground plane and to the coil size. Since the presented coil allows for flexibility in transmit phase and amplitude as well as excitation with up to 16 independent RF waveforms, we see the largest benefit of such close-fitting design for future parallel transmit applications. Phase adjustments were often found to be essential to correct for potential destructive transmit phase interferences for very tight coil sizes and thus we consider the smallest coil sizes only a realistic option for experimental setup that allow for on the fly RF shimming. While this work focused primarily on the more difficult transceiver situation, similar receive-only arrays built from a higher number of short transmission line elements should prove to be a very valuable design option at ultra high fields. Such designs immediately address the need to be close to the subject and at the same time allow for geometrical flexibility, circumventing the restrains of tight-fitting fixed-geometry holders.
We thank Bill Voje for his invaluable contributions to the mechanical design and construction of the holder, Peter Andersen and John Strupp for the development of the 32-channel digital receiver used in the initial experiments, and Udo Adriany for help with the AUTOCAD drawing. The 7T system purchase was funded by the W.M. Keck Foundation, National Science Foundation (NSF) grant 9907842, and National Institutes of Health (NIH) grant S10 RR1395.