7T Human Spine Imaging Arrays With Adjustable Inductive Decoupling

Abstract

Ultrahigh-field human spine RF transceiver coil arrays face daunting technical challenges in achieving large imaging coverage with sufficient B₁ penetration and sensitivity, and in attaining robust decoupling among coil elements. In this paper, human spine coil arrays for ultrahigh field were built and studied. Transceiver arrays with loop-shaped microstrip transmission line were designed, fabricated, and tested for 7-tesla (7T)MRI. With the proposed adjustable inductive decoupling technique, the isolation between adjacent coil elements is easily addressed. Preliminary results of human spine images acquired using the transceiver arrays demonstrate the feasibility of the design for ultrahigh-field MR applications and its robust performance for parallel imaging.

I. INTRODUCTION

High magnetic fields and parallel imaging are synergistic and complementary, providing both high spatial resolution and high temporal resolution in MR imaging and spectroscopy. Parallel imaging with multicoil arrays provides a solution to many of the problems encountered with fast acquisitions at high magnetic fields [1]–[5]. In turn, high fields are expected to improve parallel imaging performance due to the more complex sensitivity profile of each coil element and increased SNR [3], [6]–[8]. A challenge at ultrahigh fields, however, is the increased difficulties encountered when designing arrays, which are optimized for parallel imaging, especially for samples requiring a large field of view (FOV) such as the human spine. The conventional method utilizing a large body volume transmit coil for spin excitation and a separate array of multiple local surface coils for signal reception at lower field strengths is not readily extendable to ultrahigh fields, because the required large transmit volume coil is usually unavailable. In addition, the transmit B₁ (i.e., ${\mathrm{B}}_1^{+}$) profile generated by the large volume coil at ultrahigh fields is highly nonuniform in the human body owing to the body’s high permittivity and large size. Alternatively, a promising approach when working with ultrahigh fields is to build transceiver coil arrays that allow for independent phase and amplitude control for each coil element. This method would enable B₁ shimming commonly required at high fields and parallel excitation. These transceiver arrays are usually built with the structure of a microstrip transmission line (MTL) [3], [9]–[16] which, in its simplest form, consists of a thin strip conductor and a ground plane separated by a low-loss dielectric substrate [9]. Due to its unique semiopen structure, substantial electromagnetic energy is stored in the dielectric substrate. As a consequence, this structure combines the advantages of phase-adjustment on each coil element and intrinsic RF shielding of the transmission-line, enabling low radiation losses, reduced mutual inductive coupling and improved sensitivity distribution. In contrast to conventional lumped-element RF coils, the MTL has demonstrated superior advantages in ultrahigh-field transceiver coil array designs for human parallel MR imaging applications.

A number of technical challenges and design considerations in the design of transceiver arrays for the imaging human spine at the ultrahigh field of 7T need to be addressed. Sufficient imaging coverage and B₁ penetration with efficient MR signal excitation and reception are required in order to detect the signals from the entire length of the spine. The strong coupling of coil elements to the sample mediates interactions among the coil elements, making it more difficult to electromagnetically decouple them [17]. Although the interconnecting L/C components [18], [19]would be employed for decoupling coil elements at 7T or higher fields [3], [13], [15], and [16], the method is not always efficient for coil arrays with heavier loading. In addition, the required decoupling capacitance is often impractically small and extremely sensitive to different loads, resulting in the difficulties of on-site tuning and matching [15], [16]. Inductive decoupling is an improved scheme for high fields [15], but in practice it is not convenient to achieve optimized decoupling by tuning a multiturn inductor.

In this work, we explore the feasibility of designing transceiver arrays for human spine parallel MR imaging at 7T, using loop-type microstrip elements with proposed adjustable inductive decoupling. Preliminary 7T spine images acquired from healthy volunteers are presented. Accelerated images with Generalized autocalibrating partial parallel acquisition (GRAPPA) are also obtained with MATLAB to demonstrate its parallel imaging performance at the ultrahigh field of 7T.

II. MATERIALS AND METHODS

A. Size of Coil Elements and B₁ Penetration

Loop-shaped MTL coils were selected to build the spine array. These loops were linearly placed along superior–inferior direction without coil overlap to optimize its parallel imaging performance. Sufficient B₁ penetration is critical for such planar spine arrays to study spinal cords and discs. Unlike the conventional low-field spine arrays in which the homogeneous ${\mathrm{B}}_1^{+}$ is easily achieved with a body volume coil, this spine array may produce nonuniform ${\mathrm{B}}_1^{+}$ field, which is highly confined around coil elements. For this reason, the size of each MTL loop and number of loops were investigated for a sufficient imaging coverage.

Three coil geometries were considered with different coil size and coil number. The element dimensions are 12 cm × 12 cm with four channels, 12 cm × 8 cm with six channels, and 12 cm × 6 cm with eight channels. Gaps between nearest neighbors for all the versions were 1 cm and low loss substrate with 1.2 cm thickness was selected. The total lengths of those arrays were all approximately 50 cm to fit the human spine scan. Coil elements with a size larger than 12 cm × 12 cm are difficult to build at 7T (~ 300 MHz), and those, which have a size smaller than 12 cm×6 cm may not have enough B₁ depth. To investigate the B₁ penetration of those three arrays, three MTL coils with the three sizes were built with a resonant frequency of 300 MHz and tested with a shielded 1.5-cm single-loop inductive pickup probe. S₂₁ was taken on an Agilent Model E5070B network analyzer by connecting the measured coil and the sniffer into two ports of the network analyzer. The penetration measurements for each coil were taken along the main axis of the loop and based on the depth that the picked voltage decayed to 5% of the maximum voltage. The comparison results are shown in Table I.

TABLE I.

B₁ Penetration of Three Coil Sizes

Coil size (cm)	Channel number	B1 field penetration (cm)
12 × 12	4	11.6
12 × 8	6	10.2
12 × 6	8	8.5

Based on our measured results, the MTL loop with the size of 12 cm × 6 cm has limited penetration less than 9 cm, which is insufficient for spine detection. Other two versions with larger coil elements were finally selected to build the 7T transceiver spin array.

B. Decoupling Scheme for Transceiver Coil Array

Although the RF shield of MTL reduces the mutual coupling between coil elements, in most cases when the substrate thickness is increased to pursue deeper B₁ field penetration, or the coil array is built with narrow gaps between coil elements, the mutual coupling among coils may be strong enough to cause the resonance peak to split. Hence, additional decoupling circuitry is necessary to isolate individual coil elements. This becomes particularly important in transceiver arrays where low-input impedance preamplifier decoupling is not easily implemented. Decoupling methods such as placing capacitors across nearest coil elements are currently used for microstrip arrays [3], [14], [16]. As shown in Fig. 1(a), the required decoupling capacitance can be expressed as follows [13]:

$$\begin{array}{c}{C}_d=\frac{1}{L{\omega }^2}\frac{2{(1+C_t/C_1)}^{2}k}{{(C_t/C_1)}^2(1-k^2)-2(C_t/C_1)k(1+k)},\\ \text{\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}for }\frac{C_t}{C_1}>\frac{2k}{(1-k)}\end{array}$$ (1)

where L is the selfinductance of each coil element; k is the coupling coefficient defined as $k=M/\sqrt{L_1{L}_2}$, in which M is mutual inductance; L₁ and L₂ are the selfinductances of each loop. For the special case when L₁ = L₂ = L, k can be defined as k = M/L. ω is the desired frequency. C_t and C₁ are tuning capacitors on the MTL loop. If there is no mutual inductance involved between coil loops, the relationship between C_t and C₁ and resonance frequency ω can be simply expressed as

$$\frac{1}{\omega }\left(\frac{1}{C_T}+\frac{1}{C_1}\right)=\omega L.$$

Fig. 1.

(a) Schematic of decoupling circuits with interconnecting capacitors, (b) interconnecting inductors, and (c) adjustable inductive decoupling.

Based on (1), the decoupling capacitance is inversely proportional to square of the resonance frequency. At a resonance frequency higher than 300 MHz, the required decoupling capacitance is usually very tiny and quite sensitive to different loadings. As a result, the coil adjustments for each patient are complicated and time consuming. The alternative method is applying inductors across the nearest neighbors for decoupling, which is shown in Fig. 1(b). The required decoupling inductance is expressed in (2) [13]

$$\begin{array}{c}\hfill L_d=L\frac{{(C_t/C_1)}^2(1-k^2)-2(C_t/C_1)k(1+k)}{2k{(1+C_t/C_1)}^2},\hfill \\ \text{\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}for }\frac{C_t}{C_1}>\frac{2k}{(1-k)}.\end{array}$$ (2)

It is revealed that implementing inductors is particularly advantageous at ultrahigh fields since the decoupling inductance is independent of resonance frequencies. After choosing the appropriate decoupling inductance, the coil array will be more stable to different loads. In this case, the loss from inductors is negligible since the sample loss is dominated at ultrahigh fields. A disadvantage to this method, however, is that it requires inconvenient decoupling adjustments.

C. Adjustable Inductive Decoupling

If the tuning capacitors C₁ and interconnecting inductors L_d in Fig. 1(b) are treated as a decoupling network, the decoupling condition can be achieved by varyingC₁ instead of L_d. This new method is depicted in Fig. 1(c), in which C₁ is marked as C_d for decoupling adjustment. For achieving the best decoupling performance among the elements, the overall inductance of the decoupling circuitry is adjusted by tuning the decoupling capacitor while the interconnecting inductance keeps unchanged. Although a capacitor is introduced in the decoupling circuitry, this scheme is still inductive decoupling, thus it is particularly useful and convenient in practice at ultrahigh fields. The relationship between inductor L_d and the ratio C_t/C_d was plotted in Fig. 2 according to (2), assuming the selfinductance of each coil is 1 µH. It is clear that C_t/C_d is proportional to interconnecting inductance. As indicated in Fig. 2, coils with weak mutual coupling (k < 0.2) should use small interconnecting inductance values (less than 1 µH), otherwise, adjusting C_t and C_d will be difficult both for decoupling and frequency tunings. Very small inductance values (less than 0.4 µH) may be sensitive to different loadings and should also be avoided.

Fig. 2.

Explanation of adjustable inductive decoupling. Decoupling condition is achieved by varying C_t/C_d, when interconnecting inductors have constant value.

D. Design of 7T Spine Array

Two spine arrays with different coil element sizes were designed and built for proton MR imaging at 7T, corresponding to the resonance frequency of 298.3 MHz. The structures of two spine arrays are depicted in Fig. 3. One array had six coil elements [Fig. 3(a)]. The size of the coil elements in the square shape was 8 cm × 12 cm. The second array had four elements [Fig. 3(b)], each element with a bigger loop size of 12 cm × 12 cm. MTL loops in the two designs were all linearly placed with only 1 cm innergaps. All the substrates (dielectricmaterial) used were 1.2-cm-thick polytetrafluoroethylene (PTFE), which has a low loss tangent (tanδ < 0.000 15) and a permittivity of 2.1. Other kinds of low-loss dielectric materials with different permittivities can certainly be used. Because corners of the coil tend to radiate surface waves, and potentially creating hot spots in images and degrading the Q values of the coils, the corners had been chamfered to reduce the radiation loss and improve B₁ distribution. The strip conductors and ground planes of MTL loop arrays were made from 36-µm-thick adhesive-backed copper tape (3M, St. Paul, MN). The width of the strip conductor for all coils was 1.2 cm. Three trimmer capacitors (Voltronics, Denville, NJ) ranged within 19 pF were mounted on each loop, for frequency tuning, matching, and decoupling. Two homemade nonmagnetic inductors were connected between the nearest coils as indicated in Fig. 1(c). These inductors were made of 15-turns copper wires with 5 mm diameter. Inductances were approximately 0.8 µH at 100 kHz. By slightly changing the decoupling capacitance and the tuning capacitance on each loop, the decoupling condition was finally achieved. On-site tuning and matching of each coil should be conducted with the array inside the magnet bore and loaded with the human spine. It is unnecessary to adjust the decoupling capacitors again at the loading case, which indicates an advantage of the inductive decoupling scheme.

Fig. 3.

(a) Photos of the six-channel and (b) four-channel human spine array at 7T.

Each coil element was connected to the T/R switch of the MR scanner via a coaxial cable. For each coil array the resonance frequency and decoupling were confirmed using the reflection coefficient S₁₁ and the transmission coefficient S₂₁ measurements taken on a network analyzer. The reflection coefficient S₁₁ measurement was also used to measure the coils’ Q values for the loaded (with the human spine) case.

E. MRI Experiments

The MR imaging experiments with these coils were performed on a 7T/90 cm magnet (GE healthcare, Milwaukee, WI). This scanner is equipped with two quadrature transmit channels and two T/R switches. To test the transceiver spine arrays on this system, scans were conducted by connecting two coil elements into the transmit channels each time and combined offline. To avoid the signal cancellation from the phase difference of the channels, each scan was performed on two nonadjacent coil elements. All other coil elements without scanning were terminated with 50 Ω terminators. Ten healthy volunteers were tested with these arrays by using gradient echo (GRE) images with different echo time (TE) and repetition time (TR) at 20° flip angle. Other MRI parameters were matrix size: 256 × 256, FOV: 35 cm × 35 cm (the maximum achievable FOV) at sagittal plane and 20 cm × 20 cm at axial plane.

The individual channel images are shown to demonstrate the isolation of coil elements. Combined images at axial and sagittal planes are also shown. To test its parallel imaging performance, GRAPPA [20] with the subset selection strategy [21] was applied at reduction factors of 2, 3, and 4 (autocalibration lines, ACS line = 60).

III. RESULTS

Each element of the transceiver array was tuned to 298.2 MHz (the proton Larmor frequency of the 7T magnet that we used in this paper) and matched to system’s 50 Ω. At ultrahigh fields where the sample loss is dominated, the losses from the coil and decoupling circuits are negligible. Decoupling between the nearest neighbors was approximately −23 dB without loading and approximately −17 dB after loading of the human spine. The loaded isolation decreased by several decibels due to the coil’s strong coupling to the sample, but it is still in the acceptable range for such transceivers. Isolations among all nonadjacent elements were better than −25 dB both with and without loading. Fig. 4 shows the measurement results on scattering parameters S₁₁, S₂₂, and S₂₁ of the prototype transceiver array loaded with the human spine.

Fig. 4.

Scattering parameters S₁₁, S₂₂, and S₂₁ measurements between the nearest neighbors of the four-element array after loading with the human spine. The isolation is −17 dB. Frequency span is 30 MHz.

Spine images acquired from the same healthy volunteer by using the four-element array and six-element array are shown in Fig. 5. The thoracic-spine disc images were acquired with a GRE sequence within one breathhold (flip angle: 20°, matrix size: 256 × 256, slice thickness: 5 mm, TE/TR: 4.2ms/117 ms). Power used for transmissionwas approximately 250 W per channel. The measurements showed that the four-element array had a deeper penetration of ~12 cm while the six-element array had ~9 cm, which may show limitations in intervertebral disc imaging unless higher transmit power (>250 W for each channel) is applied. This conclusion is in accordance with our results from bench tests. Thus, the four-element array was finally selected for the following spine imaging.

Fig. 5.

Images acquired with (a) six-channel array and (b) four-channel array. Spinal disc images indicate that four-channel array with 12 cm ×12 cm coil size has deeper B1 penetration and better SNR for discs. The same transmit power ~250 W per channel was applied. Images were acquired with GRE sequences (flip angle: 20°,matrix size: 256×256, slice thickness: 5 mm, TE/TR: 4.2 ms/117 ms).

Individual channel images in Fig. 6(a)–(d) illustrate the good coil isolation between the four elements. After combination with the sum-of-squares (SoS) method, the whole spine image [see Fig. 6(e)] has an acceptable homogeneity at the spinal disc area. Almost all the discs are clearly visible except the ones between the coil element gaps. Fig. 7(a) shows the sagittal image of lumbar spine (GRE flip angle: 30°, matrix size: 256 × 256, FOV: 35 cm × 26 cm, slice thickness: 4 mm, TE/TR: 3.7 ms/150 ms, NEX 10). An axial image is shown in Fig. 7(b) (GRE flip angle: 20°, matrix size: 256 × 256, FOV: 20 cm × 20 cm, slice thickness: 5 mm, TE/TR: 4.2 ms/117 ms), in which the spinal cord is clearly visible.

Fig. 6.

(a)–(d): Images acquired from each of the loops of the spine array. (e): Composite image produced using the root of the SoS method. Images were acquired with a GRE pulse sequence (flip angle: 20°, matrix size: 256 × 256, slice thickness: 5 mm, TE/TR: 4.2 ms/117 ms, FOV: 35 cm × 35 cm). A portion of the images covering the human spine is shown.

Fig. 7.

GRE images (flip angle: 30°, matrix size: 256 × 256, FOV: 35 cm × 26 cm, slice thickness 4 mm, TE/TR: 3.7 ms/150 ms, NEX: 10) were acquired from a healthy volunteer in (a) sagittal and (b) axial orientations (flip angle: 20°, matrix size: 256 × 256, FOV: 20 cm × 20 cm, slice thickness: 5 mm, TE/TR: 4.2 ms/117 ms).

The parallel imaging capability of this array was also tested. Datasets were reconstructed offline with a customized GRAPPA-based parallel reconstruction algorithm developed in our laboratory in MATLAB(The Math Works, Natick, MA). Parallel imaging with reduction factors 2, 3, and 4 was successfully employed with 60 autocalibration signals lines and therefore the real acceleration rates are 1.62 [see Fig. 8(b)], 2.04 [see Fig. 8(c)], and 2.35 [see Fig. 8(d)]. The curvature distortion of the images was caused by gradient nonlinearity and imperfect B₀ shimming. This effect was not corrected in the images, since our research goal at this point was to demonstrate parallel imaging performance.

Fig. 8.

Spine images reconstructed using the root of the SoS method (a), and using GRAPPA of reduction factors 2, 3, and 4 (ACS line = 60), so the real reduction factors were (b) 1.62, (c) 2.04, and (d) 2.35. The distortion of the images is mainly caused by gradient nonlinearity and imperfect B₀ shimming.

IV. DISCUSSION AND CONCLUSION

The feasibility of 7T human spine arrays was studied. Four-channel spine transceiver arrays with MTL structure were designed and constructed. Preliminary in vivo images of the human spine were also acquired and demonstrated with the coil array. The microstrip loop design associated with the extra adjustable inductive decoupling is suitable for human spine imaging, providing large image coverage, appropriate B₁ penetration, and decent parallel imaging performance.

As preliminary work for ultrahigh-field human spine imaging, several issues need to be addressed. First of all, multiple transmit channels are necessary to be utilized for such coil arrays. Due to the lack of transmit channels, we tested this spine array through multiple scans and combined the individual channel images offline. Thus, the independent phase control of each channel was not investigated. With the capability of phase control, better homogeneous and B₁ depth may be achieved. Second, the pulse sequences applied to the spine arrays were all GRE sequences with small flip angles for subject safety and MR system reliability considerations. To avoid a large specific absorption rate, we have not tested pulse sequences that require high-RF power for imaging such as fast spin echo. We believe that higher RF power (bigger flip angles) and better B₀ shimming may improve the image quality with deeper B₁ penetration for imaging deeper structures such as the intervertebral discs. Third, the motion artifacts caused by B₀ distortion from respiration and heartbeat may significantly degenerate the image quality especially for thoracic spine imaging. As a result, the thoracic spine images were acquired within one breathhold to minimize the motion artifacts. The spinal imaging SNR, therefore, was not optimized. The future work will be focused on parallel excitation and investigation of the transmit efficiency of the proposed design.

Biographies

Bing Wu received the B.E. degree in electric engineering and automation from Tianjin University, Tianjin, China, in 2002, and the Ph.D. degree in electrical and electronic engineering from the University of Hong Kong, Hong Kong, in 2006.

He was a Radio Frequency (RF) Engineer at the General Electric Healthcare Global Technology Organization (GTO)-MR, China. He is currently a Postdoctoral Research Scholar of Radiology and Biomedical Imaging at the University of California, San Francisco. His research interests include RF coil development and image reconstruction for magnetic resonance imaging.

Chunsheng Wang received the B.Sc. and M.Sc. degrees from the University of Science and Technology of China, Hefei, China, in 1999 and 2002, respectively, and the Ph.D. degree from the University of Hong Kong, Hong Kong, in 2006.

He is currently a Postdoctoral Scholar at the University of California, San Francisco. His research interests include the magnetic resonance coil design, electromagnetic calculation, and parallel imaging algorithm.

Roland Krug received the Ph.D. degree from the University of Heidelberg, Heidelberg, Germany, in 2003.

He is currently with the University of California, San Francisco. His research interests include ultrahigh-field magnetic resonance imaging, pulse sequence development, and digital image processing with a main focus toward the musculoskeletal system.

Douglas A. Kelley, photograph and biography not available at the time of publication.

Duan Xu graduated with B.A. degree in integrated science program and B.S. degree in biomedical engineering from Northwestern University, Evanston, IL, and received the Ph.D. degree in developing advanced techniques in magnetic resonance imaging from the University of California, San Francisco (UCSF)/UC Berkeley Joint Graduate Group in Bioengineering, San Francisco and Berkeley.

He was a System Analyst at the Federal Reserve Board of Governors, Washington, DC. He is currently an Assistant Professor at the Department of Radiology and Biomedical Imaging, UCSF.

Dr. Xu is a member of the International Society of Magnetic Resonance in Medicine and a Research Scientist Member for the American Society of Neuroradiology.

Yong Pang received the B.E. degree from Tianjin University, Tianjin, China, and the Ph.D. degree from the University of Hong Kong, Hong Kong, in 2008.

He is currently with the Department of Radiology and Biomedical Imaging, University of California, San Francisco, as a Postdoctoral Fellow. His research interests include radio frequency (RF) coils for animal magnetic resonance imaging/magnetic resonance spectroscopic study, multitransmit channel, multidimensional RF pulses, and sequence design for in vivo applications at ultrahigh field.

Suchandrima Banerjee received the Bachelor’s degree in electrical engineering from Jadavpur University, Kolkata, India, and the Ph.D. degree from the University of California, San Francisco (UCSF)/UC Berkeley Joint Graduate Group in Bioengineering, San Francisco and Berkeley.

She is currently a Magnetic Resonance Scientist at the Applied Science Laboratory, General Electric Healthcare, Menlo Park, California. She was with the Department of Radiology and Biomedical Imaging, UCSF. Her research interests include parallel imaging methods in magnetic resonance imaging.

Daniel B. Vigneron received the B.A. degree in chemistry from Wesleyan University, Middletown, CT, in 1983, and the Ph.D. degree in pharmaceutical chemistry from the University of California, San Francisco, in 1988.

He is currently a Professor of Radiology and Biomedical Imaging at the University of California, San Francisco (UCSF), where he is engaged in research on the development of magnetic resonance imaging methods. He is also with the UCSF/UC Berkeley Joint Graduate Group in Bioengineering, San Francisco and Berkeley.

Sarah J. Nelson, photograph and biography not available at the time of publication.

Sharmila Majumdar, photograph and biography not available at the time of publication.

Xiaoliang Zhang received the B.E. degree in radio engineering from the South China University of Technology,Guangzhou, China, and the Ph.D. degree from the University of Minnesota, Minneapolis.

He was with The Ohio State University, Columbus, developing the 340 MHz transverse electromagnetic resonator for the world’s first 8-tesla whole body magnetic resonance (MR) system. He was with Picker International (currently Philips Medical Systems) working on the 3T project. In 1999, he joined the Center for Magnetic Resonance Research, University of Minnesota. He was an Assistant Professor at the Department of Radiology, University of Minnesota Medical School. In 2006, he joined University of California, San Francisco (UCSF), where he is currently an Associate Professor of Radiology and Bioengineering, and a core faculty member of the UCSF/UC Berkeley Joint Graduate Group in Bioengineering, and California Institute for Quantitative Biosciences (QB3). His research interests include parallel imaging and high-field MR technology, and their in vivo applications.