|Home | About | Journals | Submit | Contact Us | Français|
A 20-channel phased-array coil for Magnetic Resonance Imaging (MRI) of mice has been designed, constructed and validated with bench measurements and high resolution accelerated imaging. The technical challenges of designing a small, high density array have been overcome using individual small-diameter coil elements arranged on a cylinder in a hexagonal overlapping design with adjacent low impedance preamplifiers to further decouple the array elements. Signal-to-noise ratio (SNR) and noise amplification in accelerated imaging were simulated and quantitatively evaluated in phantoms and in vivo mouse images. Comparison between the 20-channel mouse array and a length-matched quadrature driven small animal birdcage coil showed an SNR increase at the periphery and in the center of the phantom of 3-fold and 1.3-fold, respectively. Comparison to a shorter but SNR-optimized birdcage coil (aspect ratio 1:1 and only half mouse coverage) showed an SNR gain of 2-fold at the edge of the phantom and similar SNR in the center. G-factor measurements indicate that the coil is well suited to acquire highly accelerated images.
Combining mouse models with modern imaging modalities facilitates in vivo monitoring for morphologic and functional changes in biology properties (e.g. tumor tissue), for example, the longitudinal analysis of the tumor response to therapy, resulting in an increased efficiency of animal experiments. To visualize small murine structures, a sufficient spatial resolution with good soft tissue contrast as well as good signal-to-noise-ratio (SNR) in the resulting image is required. On this account, many groups use dedicated small animal, high-field MRI systems, which are available up to 16.4 Tesla (1) or even upright MR scanner equipped with gradient inserts can be found up to 20.1 Tesla (2). Unfortunately, such MR imagers are very expensive and only usable for limited purposes. Furthermore, physical contrast characteristics and imaging performance differ from clinical MRI machines. Therefore, the results can not always be matched with clinical studies in one-to-one basis. Additionally, state-of-the-art small animal MR-scanners are typically only equipped with up to four or sometimes eight receiver channels, and thus do not have the capability for highly accelerated parallel imaging applications. Subsequently, an alternative to these dedicated devices is the use of already existing clinical MR scanners, which although themselves more expensive than typical small-bore scanners, are already available to the most medical science institutions conducting animal experiments and can thus be utilized for animal imaging in off-hours with minimal incremental cost.
In the last decade, human MRI parallel imaging has gained great popularity in nearly all clinical MR applications. Roemer et al. (3) laid the foundation using a revolutionary approach in hardware and image combining methodology, he referred to as the NMR phased-array in order to increase the SNR in MRI. He implemented a system using four reception coils and four receiver channels, which allowed simultaneous and independent data acquisition. Through the use of such multiple receiver coils, it is possible to cover a larger volume of interest with additionally increased SNR. Furthermore, it allows for the application of accelerated parallel imaging techniques, e.g. sensitivity encoding (SENSE) (4) or generalized autocalibration partially parallel acquisitions (GRAPPA) (5). These techniques acquire the images with a reduced number of phase-encoding steps, resulting in under-sampling in k-space and thus reduced scan time. Different channel counts and array layouts (e.g. gapped, shared conductor, or overlapped) vary in spatial modulation of signal intensity and thus providing different abilities to unalias folded images (SENSE method) or synthesize spatial harmonics (GRAPPA methods) (6, 7).
MRI systems equipped with 32 independent receive channels are now becoming standard in clinical practice. Furthermore, as many as 128 radiofrequency (RF) channels may become available as the next generation for clinical MRI applications (8, 9). However, the application of parallel imaging techniques with high density coils for mice has not yet been fully developed. Mouse dedicated phased-arrays using up to four receive channels have demonstrated the capability of accelerated parallel imaging (10–13). Eight channel mouse arrays have been reported for 3T clinical MRI systems (14) including a dedicated 8-channel 13C array (15). Recently, an 8-channel body array for a 9.4 T animal scanner has been described (16). Similar increase in number of coil elements can be observed in rat MRI applications using 8- and 16-channel body arrays (17–19). Extending the capabilities of parallel imaging to higher accelerations requires the implementation of a miniature design and solving the challenges of a high density coil with on-coil preamplifiers.
In addition, we used electromagnetic simulations to specify information regarding parallel imaging performance for different array designs, which included gapped, shared conductor, and overlapped designs for 8- and 20-channel whole-body mouse arrays. The constructed phased-array was evaluated with bench and imaging metrics and its characteristics were compared to circularly polarized (CP) birdcage volume coils. The developed coil array was used to perform in vivo parallel imaging to speed up the data acquisition for whole-body mouse imaging.
Twenty circular elements were arranged in a 5×4 hexagonal matrix on a 34 mm diameter fiberglass tube. The coil array has a length of 95 mm, which covers the whole-body of a mouse. Each coil element has a diameter of 32 mm and was constructed from tubular conductors (Goodfellow Corp., Oakdale, PA, USA) with an inner and outer diameter of 0.8 mm and 1.2 mm, respectively. This design was chosen after the evaluation of the unloaded-to-loaded Q-ratio of three different loop materials/geometry configurations. Namely, beside the tubular conductors, flexible circuit board material (Pyralux®, Du Pont, Wilmington, DE, USA) and a 18awg solid copper wire coated with silver (Fig. 1a–c) were tested.
The geometrical structure of the whole design was engraved on fiberglass tube (Krülit 750, Krueger & Sohn, Landshut, Germany) before the loops were mounted (Fig. 1d). Conductor crossings of nearest neighboring elements are realized by bridges. The loops and bridges were also made out of the tubular conductors with gaps for placement of capacitors (Fig. 1e). We divided each loop symmetrically by two gaps where we place the capacitors (see schematic circuit in Fig. 2). Thus, each half of the loop is joined at the top by the tuning circuit and at the bottom by the output-circuit. The tuning circuit contains one chip capacitor C2 (Series 11, Voltronics, Danville, NJ, USA) and an adjustable capacitor C1 (BFC5 808 11339, Vishay Intertechnology, Inc, Malvern, PA, USA) to fine-tune the loop resonance to 123.25 MHz. The output circuit-board contains a capacitive voltage divider (C3, C4) and a series capacitor C5 to match the elements output to 50 Ohm under loaded conditions. The drive-point circuit board also incorporates an active detuning circuit across one of the voltage divider capacitor C4. The active detuning is achieved using a PIN diode D in series with a hand wound inductor L, which together with capacitor C4 resonates at the Larmor frequency. Thus, when the PIN diode is forward biased (transmit mode), the resonant parallel LC circuit inserts a high impedance in series with the coil loop, blocking current flow at the Larmor frequency during transmit. A self shielded resonant toroidal coaxial cable-trap was placed between the coil element and preamplifier to suppress common modes on the cable. Each cable-trap was made out of semi-rigid coaxial cable (UT-0.047, Micro-Coax, Pottstown, PA, USA) and bridged with a high voltage capacitor C6 (Series 25, Voltronics, Danville, NJ, USA). The trap was placed close to coil elements, directly behind the matching network.
Adjacent loops were overlapped to minimize the mutual inductions between the elements. At the locations where the coil array has curvature due to the form of the tube, a coil overlap ratio of 0.73 of their diameter was found to provide an optimal geometrical decoupling. The critical overlap of adjacent loops along the tube (z-direction) was found to be 0.76 (Fig. 1d). The crossing bridges were bent to empirically improve the decoupling between the neighboring loops. This procedure was controlled via an S21 measurement.
Next nearest neighboring and other coil elements were decoupled using low impedance preamplifier decoupling (3), where the series matching capacitor and the phase shift of the cable transform the preamp impedance to a parallel inductance across the matching capacitor in the coil loop. This parallel LC circuit is set to resonate at the Larmor frequency and introduces a high serial impedance in the coil loop (3). In this mode, minimal current flows in the loop and inductive coupling to other coils is minimized. The transformation of the impedance was achieved by carefully controlling the cable length.
Preamp outputs were connected to a second common mode cable trap to reduce common mode currents by the body coil on the output coax of the preamplifier. Three standard coil plugs (Odu© Mueldorf, Germany) were used to connect the coil to the MR-Scanner. Each plug cable has an additional common mode cable trap to avoid interaction with the RF transmission system. The whole geometrical design of the coil is enclosed in an acrylic box, which is shown in Figure 3.
We constructed two eight-rung low pass quadrature transmit-receive birdcage volume coils (20, 21) for SNR comparison. One birdcage coil has a length of is 95 mm and a diameter of 40 mm, which is slightly bigger than the 34 mm diameter of the 20-channel phased-array. This avoids infiltration of the transmit B1 inhomogeneity into the imaging subject, which occurs close to the proximity of the conductor structure. Each leg used two series 47 pF capacitors, one at each end of the rung. The second birdcage coil is shorter (both diameter and length of 40 mm). This aspect ratio reflects a more optimum geometry for SNR performance (22), while the longer birdcage provides flank to nose coverage of the animal. Each leg used two series 81 pF capacitors, one at each end of the rung. The birdcages resonators were connected to a custom-made hybrid combiner network with integrated transmit-receive switch using a matching network and common mode traps. Additionally, both birdcage coils are shielded with a 65mm diameter cylindrical shield (10 cm length) made with 35 µm thick copper foil. The shield is separated into eight strips bridged with 2.2 nF capacitors to reduce eddy currents.
Bench testing has been carried out by a custom made test rig that provides each analog control signal from an MRI scanner during image acquisition for each separate channel. That includes the power supply of 10 V for preamplifiers, biasing of the PIN diodes with 100 mA to get detuning during the transmit mode, as well as reverse biasing of the diodes (−30 V) during image acquisition. All signals for each channel can be individually controlled. Furthermore, the test rig has BNC connections to allow the evaluation of each preamplified signal. Four coil characteristics, which are important for the performance in an MRI experiment, were optimized for each single coil element at the bench: detuning, decoupling, preamplifier decoupling, and element tuning. Additionally, the ratio of unloaded-to-loaded quality factor (QU-to-QL) was measured to specify the resistive losses of sample and coil circuit (rSample/rCoil=QU/QL−1). Decoupling measurements were done on the nearest neighboring elements using a direct S21 measurement with cables directly connected to the preamplifier sockets of the two elements under test. When measuring the decoupling between an adjacent pair, all other unused elements of the phased-array were detuned. Additional measurements (final tuning, Q-ratio, active detuning, and preamplifier decoupling) used the S12 measurement between two decoupled (~80dB) inductive probes lightly coupled to the array element under test. The probe consists of two overlapped, geometrically decoupled, 12 mm diameter semi-rigid coaxial cables with a gap in the shield placed symmetrically in the middle of each loop. This double probe is small enough to fit into the coil former. All the double-probe S12 measures are described below:
Tuning: The fine-tuning of the resonant frequency was set by adjusting the variable capacitor C1, while monitoring the S12 resonance with the pair of decoupled pickup probes.
Ratio of QU-to-QL: All the determinations of QU/QL-ratio were measured for both a single loop and when each loop under test was surrounded by its six non-resonant neighboring elements. Q-measurements were performed on an isolated loop, without a coaxial cable, or preamplifier. We always ensured the same distance between all the tested loops and the double-probe using a spacer. The coil quality factor is simply given by Q=f0/Δf, where f0 is the center frequency and Δf is −3 dB bandwidth. All loaded Q measurements were performed with a 27 ml volume cylindrical phantom (diam. 28 mm) filled with 1.25g NiSO4 × 6 H2O, 5g NaCl per 1000g H2O.
Active Detuning: The S12 change between decoupled probes was recorded for a given array element with and without forward bias current, applied to the PIN diode trap while the preamplifier was replaced by a 50Ω load.
Preamplifier decoupling: Preamplifier decoupling was determined as the difference in S21 measurement in case of power match, which has a conventional 50 Ohm termination (without presence of preamp) and the case of noise match, which has termination by low impedance preamplifier. The difference between the two states determines the preamp decoupling (23). Furthermore, we measured the stability of the preamp decoupling inside the scanner and compensated changes on the bench.
We also estimated the overall mutual coupling between the elements to verify the efficiency of both combined decoupling methods (geometrical overlap and preamp decoupling) under phantom loaded conditions. This was done using a 3-port network analyzer measurement, where a small pick-up loop was placed near each loop of the coil pair under test. One array coil element was powered directly (0 dBm) from port 1 of the network analyzer using a power-match and with no preamp in place. Port 2 (receive port) of the network analyzer was connected to a single pick-up probe placed near the driven coil. Port 3 of the network analyzer (also receive port) was connected to a second pick-up probe placed near the other coil element of the pair under test (preamp was present). The difference (S31-S21) between the power coupled indirectly over the second coil element into the pick-up probe and the power coupled directly into probe 1 gives an estimation of the overall coupling. Pick-up probe 2 was then successively moved to all the other 19 elements, and after this procedure, the next element (out of 20) was powered over port 1, to measure the mutual coupling between all its neighbors. When the pair under test was measured, all the other elements of the array were actively detuned.
All experiments were performed on a state-of-the-art 3T whole-body MRI scanner (TIM Trio, Siemens Healthcare, Erlangen, Germany), equipped with 32 RF-receivers. The system has 40 mT/m maximum amplitude gradients and a maximum gradient slew rate of 200mT/m/ms. When images were acquired with the 20-channel array, the whole-body transmit coil was used for excitation. SNR calculations were performed using the 27 ml cylindrical phantom. To evaluate the non-accelerated SNR and the noise amplification during parallel imaging, phantom images were acquired using a PD weighted 2D FLASH sequence (TR/TE/Flip = 300 ms / 4.2 ms / 25°; matrix transverse: 64 × 64; matrix coronal: 192 × 64, voxel size: 0.52×0.52×2 mm3).
Initial measurements were performed to evaluate the influence of the surrounding coil elements on the SNR achieved with one single isolated coil element. A single coil element, constructed with the same diameter and design as used in the 20-channel phased-array, was mounted alone on the coil former. The SNR from a single element alone was compared to the case where surrounding elements were present but actively detuned in order to assess losses associated from eddy currents in the conductive components of the surrounding elements. To assess the SNR decrease due to inter-element coil coupling during array detection, SNR measurements were made with a single activated coil element in the center of the array, while all its neighbors were actively detuned. The same measurements were repeated when all the neighboring elements were activated (similarly to the real MR-Experiment). In this case the corresponding channel from the uncombined image was assessed. By comparing the SNR in these two images, we can measure the degradation in image SNR from coupling to the other coil elements.
The noise correlation matrix was used to assess coil coupling and also noise covariance-weighted sum-of-squares array element combination (3) as well as the G-factor calculations using custom software written in Matlab© (MathWorks Inc., Natick, MA, USA). The noise covariance matrix Ψi,j=ni·nj* was calculated from the complex noise variances, ni and nj of the i-th and j-th coil elements obtained from a noise-only image acquired without RF excitation. The noise correlation matrix between coil receive elements was computed using: and is shown in Figure 4.
We compared the measured SNR of the 20-channel mouse coil to the SNR obtained from the custom made quadrature transmit/receive birdcage coils. All SNR maps were pixel-wise calculated using estimation of second order noise statistics from complex noise-only raw data and raw signal data to give SNR maps in absolute SNR units as described in detail elsewhere (24). The G-factor was calculated to assess noise amplification in SENSE reconstructions caused by the ill-conditioned unaliasing of the accelerated images and is estimated from the complex coil sensitivities and noise correlation matrix (4). To determine the maximum G-factor, while avoiding bias due to noise singularities, a 5×5 sliding window operator was used. The maximum mean value detected from the operator was defined as the highest G-factor. The FOV was set as tight as possible around the phantom to avoid underestimating the G-factors.
Initial in vivo measurements were performed using a male adult wild type mouse, that was anesthetized with an intraperitoneal injection of xylazin and ketamine (0.1 ml/l0 g of solution containing 0.8 ml Rompun™, 1.2 ml Ketavet™, and 8 ml NaCl). Whole-body in vivo anatomical mouse images in sagittal and coronal orientations were obtained using a 3D double echo-at-steady-state (DESS) sequence (TR/TE/Flip=120 ms / 6.52 ms / 30°; matrix: 256×128; Slices: 36, voxel size: 0.31×0.31×1 mm3, R=3×2, TA=3.46 min) and a 2D FLASH (TR/TE/Flip = 300 ms / 5.2 ms / 70°, matrix: 256×128, Slices: 22, voxel size: 0.31×0.31×2 mm3, R=3, TA=1.31 min). In vivo image reconstructions were performed online using standard GRAPPA reconstruction method (5) provided by the MR scanner software.
In order to compare our array geometry to alternative choices, we simulated the accelerated and non-accelerated image SNR for six different arrays (including the constructed array geometry). We model the relative SNR for a given loop using the principle of reciprocity (25) as the ratio of transverse B1 field and the square root of its equivalent series resistance (26). To determine the later, we construct one loop from each simulated layout determine the total series resistance by measuring the loop inductance and QL using: QL=ωL/rtot. The B1 field map was numerically simulated using a quasi-static electromagnetic model by numerically integrating the Biot-Savart equation along the current path of each coil The SNR for the combined array was then determined using the sum-of-squares SNR calculation and the G-factors using the formulation for SENSE reconstruction (4). This simplified estimation neglects inter-element coupling (we assume good preamplifier decoupling) and wavelength effects.
The simulations were performed for the six geometries shown in Figure 6 and and77 overlaid on their simulated G-factor and SNR maps. These included 8-channel and 20-channel cylindrical arrays laid out with i) a gap design, ii) a shared conductor design (no gap, no overlap), and iii) a critical overlap design (Fig 6). All six simulated array coils have the same over-all dimensions (95mm × 32mm). The 20-channel overlap simulation matches the geometry of the constructed coil. The G-factors were directly compared between the simulated and measured 20-channel overlapped array (Fig. 8).
Figure 1a–c shows test loops, from which we measured the ratio of QU-to-QL. The loop constructed from thin tubing conductors provided the best Q-ratio (QU/QL=235/115=2.0), when its six non-resonant neighbors were present. Thus, noise is expected to be equally divided between sample and circuit. The solid wire loop and flat circuit board loop showed QU/QL=190/110=1.7 and 142/100=1.4, respectively. QU/QL-ratios for isolated elements (no neighbors present) show overall slightly higher values of QU/QL=245/115=2.1, QU/QL=225/116=1.9, and QU/QL=226/103=2.2 for the tubular inductor, the sold wire and the flat circuit board, respectively. Also in the isolated mode, we observed a resonance frequency down shifts on loading of 0.3 MHz for the flat inductor and 0.2 MHz for the tubular and wire conductor. In comparison to when the neighboring elements were brought into place, the flat inductor conductor increased its loaded resonance frequency by 1.0 MHz. We did not observe a shift in resonance frequency during loading for the wire and tubular conductors.
The long and the short birdcage coils showed QU/QL=265/80=3.3 and QU/QL=210/75=1.8, respectively. The S21 measure between the two ports showed a 28 dB and 26 dB isolation.
Figure 4 shows a representative noise correlation matrix obtained from noise-only phantom images. The noise correlation ranged from 0.013% to 36% with an average of 11%. The diagonal elements of the matrix represent the noise level of each coil elements, which ranges between 0.86 and 1.18 (average noise level normalized to 1.0).
Bench tests showed a range of decoupling between nearest neighbor elements from −15 dB to −21 dB with an average of −18 dB, which is improved by additional reduction of 25 dB via preamp decoupling. The coupling between next-nearest neighbors (non-overlapped pairs) ranged from −5 dB to −17 dB with a mean value of −9 dB (measured without the presence of preamplifier). The overall mutual coupling in presence of the entire system ranged from −19 dB to −34 dB (mean= 27 dB). Highest coupling could be measured from coils, which directly facing each other across the coil former. The average mutual coupling from next-nearest neighbors was 29 dB. Furthermore, active PIN diode detuning resulted in 46 dB isolation between tuned (PIN diode forward biased) and detuned states (PIN diode reverse biased) of the array elements. Inside the magnet, a shift of 0.8 MHz was observed in the position of the preamp decoupling minimum in the S21 measurement, due to the perpendicular orientation of the preamps relative to the main static magnetic field. The correct tuning of preamp decoupling was achieved by compensating the phase shift arising from the Hall effect in the preamplifier Field Effect Transistor (FET) (27).
Image SNR measurements obtained from a single isolated coil (no other elements present) was 5% higher than the averaged SNR for the case where the other elements were present but detuned. Comparison of the average SNR of a single activated coil element (other channels detuned) to the SNR from that channel extracted from the uncombined data (other channels tuned), showed a drop of 11% in SNR.
Figure 5 shows the measured optimum SNR maps in a transverse slice acquired through the center of each coil. In comparison with the length-matched CP-birdcage coil, the 20-channel array showed an SNR gain by a factor of 1.3-fold in the phantom center and an increase of 3-fold at the surface of the phantom. Comparison with the shorter, SNR-optimized birdcage coil showed an SNR benefit of only 3% in the phantom center and a 2-fold SNR gain at the edge of the phantom.
Figure 6 shows the inverse local G-factor maps in the transverse planes for 1D and 2D acceleration for both phantom measurements and from the simulated 20-channel overlapped array using a matched FOV and voxel size. The simulated G-factors are in close agreement with the measured data; the simulated data shows slightly lower mean G-factors (5.4%). Table 1 and and22 summarize the measured G-factor for the constructed coil (mean ±SD, as well as minimum and maximum values).
Figure 7 shows the simulated inverse G-factor maps for six different array layouts (gap, shared conductor, and overlap with 8 channel or 20 channels for R=3). Table 3 shows the corresponding average and peak G-factors of the simulated coronal slices. In this comparison, the increased channel count improved the G-factor as did the gapped design. While G-factor is informative, the SNR of the accelerated scan is the more meaningful metric. Figure 8 shows a comparison of the SNR for both R=1 and R=3 images for the six simulated coil geometries. For R=3 imaging, the 20 channel overlap design was improved (6%) compared to the gapped design at the image center. In contrast, at the edge of the image, the gapped design outperformed the overlapped design by 10%.
These simulations and measurements show that the array has potential for highly accelerated imaging, as verified by the R=3×2 (R=3 in-plane phase encoding (PE) direction and R=2 through-plane PE direction in vivo whole-mouse imaging in Figure 9.
In this study we present the design and characterization of a high-density whole-body mouse phased-array coil with 20 receive-only elements on a cylindrical former. The coil was developed and used on a clinical 3T MRI system and characterized with bench tests as well as phantom and in vivo rodent imaging. Bench tests included the evaluation of unloaded-to-load Q-ratio, tuned-detuned isolation, and nearest neighbor coupling. Imaging validations included pixel-wise SNR maps, G-factor maps, and noise correlation as well as highly accelerated whole-body mouse images.
Special attention was given to both mechanical and electrical construction, to provide a reliable RF-system for long term use in MRI murine application. However, a number of technical issues arise in the implementation of a large channel-count array with small element size. In particular, the inter-element decoupling, QU/QL and SNR performance became more challenging. Prior to the coil construction, we investigated different conductor types to maximize QU/QL for the case when the surrounding elements were present. The flat circuit board loop showed the largest QU/QL-ratio when examined as a single isolated loop (no neighbors). This is attributed to the stronger capacitive coupling (measured down-shift of loaded resonance) of the flat inductor into the sample and therefore, its increased dielectric losses. Arranged in an array configuration, the flat conductor showed the highest QU loss compared to the wire and tubular conductors. Thus, the flat loop is significantly negative impacted due to eddy currents induced by its neighboring conductive material. This yields in an increased resistance and a decreased inductive reactance (measured up-shift of unloaded resonance) of the flat conductor loop. This is consistent with recent studies, which show that eddy current losses in the conductors of neighboring elements can be significant and the optimum configuration for an array uses spatially sparser conductors than would be optimum for a single element alone (9, 28). Additionally, the circuit board elements utilize twelve additional solder joints for the overpasses required to bridge the conductors of neighboring coils, each joint adding additional losses (28). The other test-loops were made out of solid wire or a tubular conductor and showed better QU/QL in the test, where neighboring elements were present. Here, the bridges over the conductors of adjacent coils were formed by bending the wire rather than soldering an additional bridge, thus minimizing the number of solder joints. Lower losses in the tubular conductor configuration are likely due to the slightly bigger cross-sectional diameter of the tubular conductor. Since the current flows on the surface of the conductor and is largest toward the inner edge of the loop coil (29), a larger diameter conductor is expected to have lower losses. Finally, since capacitive coupling between elements is largely determined by the area and spacing of the cross-over regions, the wire and tubular design reduce this source of capacitive coupling by minimizing the area and increasing the spacing between the crossing conductors. We also found that the ability to mechanically optimize the overlap between two loops by bending these wire bridges facilitated the element decoupling procedure.
Despite optimization, our QU/QL = 2, shows that the sample and component losses contribute almost equally to the image noise, for these small diameter loops with limited tissue volume within under each element. This suggests that significant improvement could be gained by reducing component losses (e.g. through coil cooling), or increasing the relative tissue load (e.g. using multi-turn coils) but significant changes in design strategy are likely needed to significantly improve the unloaded-to-loaded Q-ratio.
One of the primary aims of the current study was to investigate, if a 20-channel MRI phased-array with small receiving elements would result in SNR gains compared to a commonly used circularly polarized birdcage resonator with similar geometry. Our findings in SNR comparison between the array and the birdcage coil suggest that a significant improvement in SNR is possible even at the coil center, when the birdcage has similar geometry. In contrast, when compared to a shorter birdcage coil (aspect ratio of 1:1), the SNR was almost equal at the coil center, however, the SNR benefit at the edge of phantom was still apparent (2-fold gain). Thus, the highly parallel array can be thought as added extended coverage and gains in peripheral SNR compared to an optimized birdcage, as well as the ability to accelerate image encoding. A competing approach is to form an array of extended birdcages to extend coverage while optimizing central SNR (30).
Comparison with a standard root sum-of-squares reconstruction showed less benefit in SNR performance. This was likely due to the coupling between the coil elements, is not accounted for in the sum-of-squares reconstruction. For example, a maximum noise correlation of 36% was seen for elements facing each other across the FOV. Part of the SNR gain over the birdcage coil likely arises from the smaller over-all size of the 20-channel phased-array (conductive elements on a 95 mm/34 mm length/diameter cylinder) compared to the birdcage (95 mm/40 mm). The larger birdcage is typically used to avoid transmit B1-field inhomogeneity in the animal. In contrast, receive-only arrays, which utilize a homogeneous transmit coil (the RF body coil in our case) are not constrained by this consideration and can be made with a tighter geometry. Secondly, our requirement for whole-mouse body imaging requires a length to diameter ratio of nearly 2:1, which is larger than the optimum ratio (0.7:1) for a birdcage (22).
The measured SNR from an element in the 20-channel array with other elements detuned was approximately 11% higher compared to that element’s SNR in the uncombined image data, obtained with the other elements active. This sensitivity loss might result from residual coupling between the elements not eliminated using geometrical and preamp decoupling. Furthermore, the measured SNR from the single active element (all others detuned) was 94% of the SNR obtained from a single isolated receive element (no other loops present). This SNR drop of 6% might be attributed to eddy current losses from the surrounding coils and preamplifiers.
In most small animal MRI, whole-body coverage is not required. Thus, literature about coil development for mice is focusing mainly on dedicated local applications (e.g. brain, cardiac, or abdominal imaging). These localized MRI examinations have no benefit at the coil center for non-accelerated imaging compared to an optimized birdcage coil. Likely they would only have moderate peripheral sensitivity benefits compared to an optimized single channel surface coil or small array. But we note that increasing the channel count of array of small surface coils is that adding coverage does not detract from sensitivity. If the channels are present (as they increasingly are on clinical scanners), then adding coverage down the body by adding more coils comes only at the cost of increased construction complexity. However, several technical coil studies addressed already whole-body mouse imaging using long birdcages (31) or solenoid coils (32–34), capacitive-decoupled overlapped array (35) or array coils (14, 15, 18). Not only in research, but also several commercial coil vendors included 8-channel whole-body coils for small animals in their portfolio. Whole-body mouse imaging applications are widely used for body composition examinations in rodents (e.g. fat quantification) (36, 37). Furthermore, some multi modal mouse imaging uses MRI for anatomical information to match with nuclear-medicine imaging. In particular, when bio-distributions of radio-labeled tracers are addressed, whole-body information is necessary (38–40). Multimodal small animal imaging (e.g. PET-MRI, SPECT-MRI,) are very time consuming imaging procedures and could take up to several hours. Therefore, accelerated MR-imaging is desirable for in-vivo mouse experiments since it reduces scan time significantly. Several tumor mouse models are very sensitive to anesthesia, and so MR-imaging has to be performed as fast as possible. Additionally, high thru-put mouse imaging likely requires accelerated image encoding. High density array coils might solve these problems and provide sufficient image quality with short scan times through the use of high acceleration rates.
The accelerated imaging performance of the 20-channel phased-array is expressed through the G-factor, which reflects the SNR penalty of ill-conditioned image reconstruction in parallel imaging. Distributing the coils in all spatial directions makes the 20-channel coil well-suited to the use of highly accelerated parallel imaging for 3D whole-body mouse acquisitions. Even six-fold (R=3 in-plane phase encoding (PE) direction, R=2 through-plane PE direction) 2D accelerated data shows sufficient image quality in in-vivo mice imaging (Fig. 7). Note, for G-factor calculations the FOV was cropped as tight as possible to the phantom. In typical in vivo MR-experiments the FOV is usually set slightly bigger around the sample, which will result in a better G-factor. The calculation of average G-factor values listed in Tables 1 and and22 also includes background area of the images.
Alteration to the present coil design in order to achieve higher performance in accelerated parallel image acquisition could be possible. First, our simulations suggest that potentially higher performance can be achieved by using twenty channels rather than eight. Our G-factor simulation showed the lowest noise amplification, when the coils are arranged with an inter-element spacing, rather than overlapped. Those designs provide better distinguishable sensitivity profiles in magnitude and phase from the coil elements in order to encode spatial information (6). But acceleration in superior-inferior direction does not show any benefit in non-overlapped array design (41, 42). Furthermore, non-overlapped arrays provide less baseline SNR at the image center (but usually higher SNR close to surface, assuming sample noise domination). However, the gapped vs. overlapped choice becomes less critical for large coil arrays. Wiesinger et al. showed a break-even number of 16 channels arranged around a spherical phantom, when gap design becomes less favorable for accelerated imaging (7). Although, the 20-channel overlapped array provide 17% and 12% less average G-factor at acceleration R=3, compared with gap design and shared conductor layout, respectively, after image reconstruction the SNR was slightly higher in the center as in the non-overlapped arrays (Fig. 7). Thus, the larger and overlapped elements overcome the improved G-factors afforded by the gap and sheared conductor in the center regarding SNR, which is the metric we ultimate care about. In our simulations, while the 20-channel gapped array had an improved G-factor compared to the overlapped design, it had slightly reduced central SNR for R=3 accelerated imaging.
Commercially available small bore high field animal scanners provide significantly better gradient performance for small animal experiments compared to human clinical MR systems. Only a few small animal systems can accommodate 20-channel coils, nevertheless, the spatial resolution obtained using the 20-channel coil in combination with a clinical 3T system is sufficient for many studies. Furthermore, the MR signal is influenced by a number of parameters, including proton density, water diffusion, blood flow, T1 relaxation time, T2 relaxation time and magnetic susceptibility. The latter cases revealed considerable dependence in the magnetic field strength. The use of small animal MRI scanners with high field strength can therefore make it hard to directly address clinical questions in mouse models. Whereas, imaging results can potentially be transferred directly to the clinical setting, if high quality animal images can be obtained from clinical scanners. In particular, contrast agent evaluations benefit from field strength matched image acquisitions. Thus, the presented 20-channel mouse coil gives an effective solution for those who work on clinical MR scanners but also require accelerated small animal MR imaging.
In this study we presented a 20-channel mouse coil for MRI and experimentally compared its performance to two different birdcage coils. We evaluated the constructed array design to gapped and shared-conductor designs in simulation studies assess its performance within this parameter space. Finally we compared the 20-channel simulations to 8-channel simulations to determine the benefit derived from the increased channel count. The challenges of a miniature high density receive-only phased-array coil have been overcome and its performance validated with bench and imaging metrics. The small coil geometry showed good decoupling between elements and high SNR. Special attention was given to the preamp decoupling, due to the high near neighbors density (each element has 12 or 15 next-nearest neighbors), which do not obtain reduction of mutual inductance from critical overlap. Simulations and phantom measurements showed that the constructed array has potential for highly accelerated imaging. Finally, in vivo tests demonstrated that the 20-channel coil is well-suited to applications requiring high sensitivity and highly accelerated imaging.
This research was supported by Von-Behring-Roentgen foundation and NIH grants P41RR14075 and R01EB006847.