A 16-Channel ¹³C Array Coil for Magnetic Resonance Spectroscopy of the Breast at 7T

Abstract

Objective:

Considering the reported elevation of ω-6/ω-3 fatty acid ratios in breast neoplasms, one particularly important application of ¹³C MRS could be in more fully understanding the breast lipidome’s relationship to breast cancer incidence. However, the low natural abundance and gyromagnetic ratio of the ¹³C isotope lead to detection sensitivity challenges. Previous ¹³C MRS studies have relied on the use of small surface coils with limited field-of-view and shallow penetration depths to achieve adequate signal-to-noise ratio (SNR), and the use of receive array coils is still mostly unexplored.

Methods:

This work presents a unilateral breast 16-channel ¹³C array coil and interfacing hardware designed to retain the surface sensitivity of a single small loop coil while improving penetration depth and extending the field-of-view over the entire breast at 7T. The coil was characterized through bench measurements and phantom ¹³C spectroscopy experiments.

Results:

Bench measurements showed receive coil matching better than −17 dB and average preamplifier decoupling of 16.2 dB with no evident peak splitting. Phantom MRS studies show better than a three-fold increase in average SNR over the entirety of the breast region compared to volume coil reception alone as well as an ability for individual array elements to be used for coarse metabolite localization without the use of single-voxel or spectroscopic imaging methods.

Conclusion:

Our current study has shown the benefits of the array. Future in vivolipidomics studies can be pursued.

Significance:

Development of the 16-channel breast array coil opens possibilities of in vivolipidomics studies to elucidate the link between breast cancer incidence and lipid metabolics.

I. Introduction

Cancer is among the most common causes of death worldwide with breast cancer being the most prevalent form among women [1] with lifetime incidence and mortality rates of 1:8 and 1:36 respectively [2]. However, worldwide incidence rates vary, with North America and Europe having highly-elevated incidence rates compared to Asian countries [1, 3, 4]. This has been partially attributed to differing sources of fat consumption between these populations [5, 6] and has led researchers to look more closely at the role of trans-fats [7] and the ω-6/ω-3 polyunsaturated fatty acids (PUFA) ratio [8–21],which is shown elevated in both western diets and in neoplasms [9, 12, 16, 19]. To further assess the link between diet and breast cancer, researchers are interested in longitudinal dietary modulation studies [8, 10, 18, 20] but these require repeatable methods of lipid assessment, making biopsies unsuitable.

One mostly unexplored alternative to biopsies is the use of ¹³C magnetic resonance spectroscopy (MRS). ¹³C MRS offers a completely non-invasive, repeatable method of quantification of ω-6 and ω-3 PUFAs as well as trans-fats [22]. However, the low natural abundance and gyromagnetic ratioof the ¹³C isotope, 1.1% and 10.57 MHz/T respectively, compared to those of ¹H (99.9% and 42.57 MHz/T)[23] lead to a relative detection sensitivity loss of four orders of magnitude compared to the ¹H nucleus. Combatting this limitation can be partially accomplished by moving to higher field strengths [24–26], but the low sensitivity of the ¹³C MRS experiment has still necessitated the use of small RF surface coils for their enhanced surface sensitivity [27–30] which has mostly limited studies to superficial anatomies.

Use of RF array coils is a common approach to retaining the surface sensitivity of a small coil while increasing the penetration depth and field-of-view [31–34] for ¹H studies, but arrays are still under-utilized for non-¹H (X-nuclei) studies largely due to lack of available receiver hardware. However, workaround techniques [35, 36]are being explored, including a frequency translation system used by our group [37]and vendor support is increasing with vendors now offering multi-channel X-nuclei receive capabilities on many systems. Recently, higher channel-count array coils [38–46] for X-nuclei have been reported and are showing significant promise for in vivo spectroscopy. Specifically for ¹³C, some studies have made use of phased array coils[47–52]; however, these studies have focused primarily on hyperpolarized applicationsat clinical field strengths.

This work describes the design and testing of a 16-channel ¹³C array coil and interface box intended for interrogation of the breast lipidome at 7T. Additionally, a quadrature ¹³C transmit coil and linear forced-current-excitation (FCE)[53–55]¹H transmit/receive coil were designed for RF excitation, shimming, and scout imaging. The overall coil system was evaluated using bench measurements and phantom MRS studies to demonstrate its improved SNR and localization ability compared to a volume coil.

II. Methods

A. Coil Design and Construction

A custom interface box containing preamplifiers, low-pass filters, and active detuning routing circuitry was first constructed. To provide protection for the ¹³C preamplifiers from the high-power ¹H transmit pulses, low-pass filters were designed as fifth-order Chebyshev filters in a pi network topology with a cutoff frequency of 125 MHz using the online LC Filters Design Tool calculator available through RF Tools [56]. Inclusions of these filters is especially important for studies involving proton decoupling which require very high isolation of the preamplifier input at the proton frequency [57]. Although this technique was not pursued in this work, the interface box was designed to accommodate the method for future experiments. Filter attenuation was checked on the network analyzer through S₂₁ measurement at the ¹³C and ¹H frequencies to ensure minimal noise insertion due to inclusion of the filters.

Low noise preamplifiers (WanTcom WMA74D) with 28 dB gain, 1.5Ω input impedance, and a noise figure of 0.4 dB are included in the interface box and were mounted in separate aluminum shielded enclosures (Hammond) after the low-pass filtering stage. DC blocking capacitors (1200 pF Passive Plus 1111C) were included at the RF input and output of each low noise amplifier (LNA).

The interface box additionally supplied the −5V/12V active detuning signal to each coil element which was routed to individual receive elements at the interface box input using bias-T networks and current limiting resistors (50Ω Vishay) to ensure the appropriate ~100 mA forward bias current during detuning. All interface circuitry was housed in an aluminum enclosure with input BNC coil connectors and output SMA connectors which interfaced with a novel frequency-translation system [37].

The array coil was designed around a hemispherical 3D-printed breast former designed in Solidworks with an outer diameter of 14.6 cm and wall thickness of 1.3 cm. The former included etchings on its outer surface for appropriate coil element positioning. An identical former was printed to support the other breast but left unpopulated. The array consisted of 16 individual loop elements arranged in a “soccer-ball” geometry [55, 58] with 10 larger loops and 6 smaller loops. The total number of elements was chosen to make use of all 16 available receive channels of the frequency-translation unit with which the system interfaced. Element positioning and the large/small element diameter ratio were chosen to ensure proper geometrical decoupling between all adjacent elements as is typical with the “soccer-ball” geometry [58]. Overall array volume was chosen large enough to accommodate 80% of breast volumes of American women [59] to allow use of the array with most breasts while maintaining a high filling factor for most cases. Dimensions of the receive array and transmit coils are shown in Figure 1.

Fig. 1.

Dimensioned rendering of coil system. The ¹³C transmit coil system consisted of Helmholtz (green) and saddle pair (red) coils. The ¹H Helmholtz coil (blue) used slightly wider loop spacing to encompass the ¹³C Helmholtz coil. Orientation of receive array elements (yellow) relative to the main magnetic field direction are shown with large elements labeled 1–10 and small elements labeled 11–16 (top right).

Each receive element was constructed of insulated 12 AWG wire, and elements were labeled 1–16 (large elements: 1–10; small elements 11–16). Components for the receive array included fixed capacitors (Passive Plus 1111C), variable capacitors (Sprague Goodman SGC3), variable inductors (Coilcraft 165), and PIN diodes (Macom MA4P7470).

Each receive element contained three breaks in the loop conductor for printed circuit boards with various circuit networks: an LCC trap to provide isolation from the ¹H transmit/receive coil [60, 61], an active detuning trap, and an SMA connection to removable match/tune and preamplifier decoupling circuitry. Each of these boards was milled on standard 1.5 mm, 1 oz copper-clad FR4 using an in-house circuit-milling system (LPKF Protomat S63). Before adding these additional networks, an element of each size was first resonated at the 7T, ¹³C frequency of 75 MHz using equal-valued capacitors soldered across each break to determine appropriate break reactances. Resonance was checked using a VNA (Agilent Technologies E5071C) dual-loop S₂₁ measurement.

Active detuning traps, ¹H decoupling traps, andmatch/tune/decoupling networks based on the design presented by Reykowski [62] were constructed for each element. Resonant cable traps tuned to the ¹³C frequency were also included on for each element with additional floating cable traps tuned to the ¹H frequency included on receive coil cable bundles. The SMA connector between the coil loop and match/tune/decoupling network allowed elements to be connected in isolation for initial tuning. The element circuit diagram and various component values for each loop size are shown in Figure 2 and Table 1 respectively.

Fig. 2.

Circuit diagrams and images for single elements of the ¹³C receive (left) and transmit (right) coils.Insets shown in each image show detailed views of the labeled networks of each circuit diagram. All receive elements used similar layouts. The saddle pair transmit element used a similar layout to the ¹³C Helmholtz element shown pictured but with the LCC decoupling traps replaced by additional segmenting capacitors.

TABLE I.

Coil Component Values

Component	Large Rx (pF/nH)	Small Rx (pF/nH)	Tx (pF/nH)
CD	82	150	100
LD	(43–60)	(25–32)	(33–48)
CT1	3.3	4.3	10
CT2	33	43	12
LT	(68–120)	(54–78)	(43–64)
RFC	560	560	810
CM1	68 + (8–30)	150 + (8–30)	18 + (2–10)
CM2	(5.5–20)	(5.5–20)	(2–20)
CM3	(8–30)	(5.5–20)	18
LM	(33–48)	(79–105)	N/A
RFC	560	560	810
CS	N/A	N/A	25

All capacitances are show in pF while inductances are shown in nH.

Note that variable components are shown with their tuning range indicated in parenthesis.

Tuning of the detuning and LCC traps to 75 MHz and 298 MHz respectively was accomplished by adjustment of the variable inductors while observing the S₁₁ measurement from a small, lightly-coupled pickup probe. For the detuning traps, the PIN diode was forward biased by 100 mA during this measurement. Cable traps were constructed using ~1 cm diameter, three-turn solenoidal windings of semi-rigid cabling (Pasternack) shielded in 0.5” diameter PVC tubing covered in copper tape. The windings were resonated at the ¹³C frequency using fixed capacitors (~60 pF, Passive Plus 1111C) while observing the common-mode attenuation through a network analyzer S₂₁ measurement.

With individual networks and cable traps tuned, each coil element was assembled and mounted on the former. RG174 cables (Pasternack) were included after each element’s match/tune network to allow connection to the network analyzer or interface box. An S₁₁ measurement of each coilelement was taken while adjusting coil trimmer capacitors on the match/tune/decoupling network to ensure impedance matching. During the process of these adjustments, preamplifier decoupling was repeatedly checked through dual-probe S₂₁ measurements taken with each element connected to the powered interface box, and trimmer capacitors were adjusted to center the S₂₁ measurement peak split at the ¹³C frequency. Though all trimmers affected both matching and preamplifier decoupling to some extent, adjustment of C_T and C_M were primarily responsible for controlling the coil match while C_X adjustment was used to control preamplifier decoupling. After adjustment of all trimmer capacitors, active tuning and preamplifier decoupling values were taken as the difference between the S₂₁ measurements with the elements terminated in 50Ω compared to detuned or decoupled. Final matching values for each element were taken with the remaining elements preamplifier decoupled. Coil Q factors were taken in this configuration using the S₁₁ 7 dB bandwidth method [63].

In addition to the 16 receive elements, additional test elements were fabricated to study the effects on loss of the various circuit networks – LCC trap, detuning trap, and match/tune/decoupling network. These test elements were identical to the receive elements but with networks other than the one under test replaced by appropriate segmenting capacitors (82 pF, Passive Plus 1111C). Four test elements were constructed for each element size: one element with all networks replaced by segmenting capacitors, one element containing an LCC network only, one element containing a detuning network only, and one element containing a match/tune/decoupling network only. The element containing the match/tune/decoupling network also contained a resonant cable trap identical to those used on array elements. Elements in all configurations were resonant but unmatched except for the case of the element containing the match/tune/decoupling network. Element unloaded Q factors were calculated for each based on the 3 dB bandwidth of a dual-loop probe S₂₁ measurement for each test element.

The transmit coil system consisted of a ¹³C quadrature Helmholtz [64] and saddle coil pair [65] and linear ¹H Helmholtz coil with geometry as shown in Figure 1. Components for the transmit system included fixed capacitors (Passive Plus 2225C), variable capacitors (¹³C - Voltronics NMNT20; ¹H – Voltronics NMAT40HVE), variable inductors (Coilcraft 165), and PIN diodes (Macom MA4P7470). Support structures for all coils were milled on 1.5 mm thick, 1 oz copper-clad FR4 with an in-house circuit milling system (LPKF Protomat S63). Each loop of the ¹³C Helmholtz coil was made with 14 AWG wire mounted on the support structure and included four equally-spaced breaks for two segmenting capacitors (25 pF), a ¹H decoupling LCC trap, and connection to a match/tune board through a 10 cm RG58 cable (Pasternack). A semi-balanced matching network was used, and a cable trap double-tuned for ¹³C and ¹H was included after the matching network. As with the receive elements, chokes were included to allow a path for the DC detuning signal. The ¹³C Helmholtz coil circuit diagram and various component values are shown in Figure 2 and Table 1 respectively.

After the matching network and cable trap, the coil was connected through coaxial cable (RG58 Pasternack) to a bias-T box to separate the RF pathway from the DC PIN diode biasing pathway. Current-limiting resistors (50Ω Vishay) were also included in the bias-T box to produce the desired 100 mA current when the −5V detuning signal was provided. ¹H trap tuning as well as active detuning and matching was accomplished via various S parameter measurements as was done with the receive elements.

The saddle pair coil used a similar design to the Helmholtz coil but with the ¹H decoupling trap replaced by an additional segmenting capacitor (36.8 pF). Like the Helmholtz coil, the saddle pair was made using 14 AWG wire, though hollow copper tubing was used to connect the top and bottom portions of each saddle element. The saddle coil match/tune board, cable-trap, and bias-T designs were identical to those of the Helmholtz coil, and coil matching/detuning used the same approach.

A circuit diagram of the ¹H coil is shown in Figure 3. Helmholtz loops for the ¹H transmit/receive coil were aligned on the same plane as the top and bottom portions of the ¹³C saddle coil. Loops were milled on 1 oz copper with a trace width of 2.5 mm, and 10 breaks were included around each loop for 9 segmenting capacitors (4.7 pF Passive Plus 1111C) and connection to a match/tune board. 298 MHz quarter-wave twinaxial cable (RG108 Pasternack) connected each loop to a common-voltage point (CVP) on the match/tune board to enforce the forced-current-excitation (FCE) condition [53, 54].

Figure 3:

A Helmholtz coil utilizing the FCE condition and a typical L matching network was employed for transmission/reception of the ¹H signal.

The FCE condition was needed in this case to ensure equal current distributions on the two Helmholtz loops despite their unequal loading conditions caused by the proximity of the upper loop to the chest wall and array elements. These loading conditions have been found to negatively affect field homogeneity at the 7T proton frequency using more standard free excitation methods but can be counteracted using the FCE method [53, 54].

A 4 cm length of semi-rigid cable (Pasternack) was used to transform the coil impedance presented at the CVP to allow straightforward impedance matching using a typical capacitive L-network with variable capacitors (1.5 – 40 pF). A dual-tuned cable trap was included after the matching network to suppress common-mode currents. Additionally, all transmit cables were bundled with floating cable traps [66] for further common-mode suppression. The receive array and transmit coils were mounted on a laser-cut acrylic former, and coil matching, active detuning, and preamplifier decoupling were rechecked in the final configuration while loaded with a 500 mL round-bottom flask of canola oil.

B. Benchtop Characterization

With the array detuned, isolation between the ¹³C transmit elements, and between each transmit element and each receive element, was characterized using an S₂₁ measurement at the ¹³C frequency. Isolation between the ¹H coil and each ¹³C transmit element was additionally characterized at both the ¹³C and ¹H frequencies.

Relative transmit field intensities were checked through S₂₁ measurements using a small ~1.5 cm diameter pickup probe situated in the middle of the transmit coil while transmitting with each transmit element. ¹³C transmit fields were further quantified through bench-measured B₁ field maps. To obtain these, a small quadrature pickup probe was mounted to a Cartesian positioning system [67] and scanned across five coronal slices of the coil volume with 1.9 cm between each slice as the ¹³C coil transmitted in quadrature mode. Single slices slightly above and below the coil volume were also measured for seven total slices. To achieve quadrature transmission on the bench, the Helmholtz and saddle coil pair transmit ports were connected to the 0° and 90° ports respectively of an external quadrature combiner (Anaren 10011–3) with the input port connected to a network analyzer and the isolation port terminated. Quadrature reception through the probe was achieved with a similar setup with the two probe channels connected to the 0° and 90° ports of a quadrature combiner (Anaren 10011–3), the isolation port connected to the network analyzer, and the transmission port terminated.

Relative B₁⁺ was taken as the normalized S₂₁ between the coil and probe. For a central slice, the measurement was repeated while transmitting in anti-quad mode by switching the transmit coil connections such that the saddle coil pair was now connected to the 0° quadrature combiner port and the Helmholtz coil was connected to the 90° port. Measurements in quadrature mode were also repeated after inserting the array coil with/without active detuning enabled though only four slices could be obtained in this configuration due to the decreased available volume.

C. Spectroscopy Experiments

A series of spectroscopy experiments were run to characterize the coil in terms of SNR improvements and localization ability. For these experiments, two different phantom configurations were used: a 500 mL round-bottom flask of canola oil termed the bulk phantom and a ping-pong ball filled with 99%¹³C-enriched, 20 mmol/L,bicarbonate solution (Sigma Aldrich) termed the localization phantom.The canola oil phantom was chosen as an analog for fatty breast tissue. Though the abundance of ¹³C within the localization phantom was much higher than would be expected in vivo, the phantom was chosen to avoid the need for signal averaging when characterizing the array element’s coarse localization ability. Neither phantom was intended to fully mimic the complete chemical composition of the breast, and these phantoms were intended only as a means of coil characterization.

The coil loaded by the bulk phantom, interface box, and frequency-translation unit [37] were set up on a Philips 7T Achieva scanner at UT Southwestern Medical Center. Transmit coil matching was verified and adjusted slightly in-bore to account for the scanner environment. Transmit coils were plugged into the standard Philips Multix interface box.

Using the ¹H coil, a survey scan, a flip angle series, and a B₀ auto-shim routine were first run, though the flip angle series was stopped before reaching a 90° tip angle due to high power requirements with a 140V peak, 200 μs sinc pulse being the highest power pulse used. The drive scale calculated from the Philips scanner to achieve a 90° tip angle was noted, and the linewidth was measured after shimming as the FWHM of the main lipid peak in the fingerprint region.

A ¹³C spectrum was next acquired from the quadrature volume coil in T/R mode using a pulse-and-acquire sequence with a hard pulse RF excitation (nominal flip angle – 18°; NSA – 8; SW – 8000 Hz; T_acq – 4096 ms). Prior to data acquisition, drive scale of the ¹³C transmit coil was increased sequentially using the stated acquisition parameters until spectral SNR appeared to decrease based on visual inspection. The drive scale was then set to the value which gave maximum SNR. This was considered to be a 90° pulse despite the system reporting nominal flip angle of 18°. For later calibration, the transmit pulse amplitude and duration out of the Philips Multix interface box (where the transmit coil would be connected in this setup) were measured by connecting the interface box transmit output to an oscilloscope (LeCroyWavejet 332A) through a 70 dB attenuator.

The system was next set up to acquire equivalent datasets from the 16-channel array. This required the scanner to be operated as if performing a ¹H spectroscopy experiment but with the frequency-translation system shifting the transmit and receive frequencies appropriately to the ¹³C resonances. To ensure comparability between the two setups, the amplitude and duration of the transmit pulse produced by the frequency translation system (where the transmit coil would be connected in this setup) were measured as was done previously while drive scale and pulse duration were adjusted to match previous measurements. Using the same scanner parameters as the T/R mode dataset, 16-channel spectra were acquired for comparison.

The bulk phantom was next replaced by the localized phantom which was placed at the apex of the receive array directly over receive element 1 and outside the sensitive region of the more distal elements. The same sequence used for previous experiments was rerun with reduced averaging (NSA – 1) to account for the ¹³C enrichment of the phantom.

After separation of the individual channel data all raw spectroscopy data was imported into Matlab for further processing. From the bulk phantom datasets, a noise-correlation matrix for the array was calculated using the last tenth (3200 points) of each channel’s FID datapointsand the Matlab correlation function. Contributions of the NMR signal were negligible within this region of the FID, and the datapoints were assumed to consist only of noise data.

For processing spectra from each phantom experiment, individual channels were first averaged in the time domain. After manually applying first order phase-correction and 3 Hz Lorentzian line-broadening on each channel, real-valued spectra were obtained through Fourier-transform of the timedomain data. The frequency axis was manually shifted to place the canola oil main fingerprinting peak at 30 ppm.

For the bulk phantom experiments, the SNR-weighted combined array spectrum was compared to that acquired by the volume coil in T/R mode. SNR-weighting was accomplished using the highest peak height of each spectrum as a measure of signal intensity [32] after normalizing noise levels and assuming zero mutual resistance. SNR calculation of the combined spectra, as well as for all other spectra obtained, was calculated by dividing the highest intensity peak height by the standard deviation of the first tenth of the spectra which contained no spectral peaks and was considered a noise region.

For the localized phantom experiment, average SNR was calculated for rows of receive elements with each row corresponding to a set of coils situated at a particular height on the former. All coils within a row were approximately equidistant from the localized phantom. Row 1 consisted only of the coil at the apex of the array former where the phantom was placed. Row 3 consisted of the 9 elements most distal to the apex, and Row 2 consisted of the 6 intermediate elements. The relationship between spectra SNR versus phantom proximity was used as the indication of the ability of individual receive elements to receive coarsely localized information.

III. Results

A. Benchtop Characterization

The completed coil system and interface box is shown in Figure 4. Interface box low-pass filters showed the expected frequency response with insertion loss less than 0.11 dB at the ¹³C frequency and stopband attenuation of greater than 43 dB at the ¹H frequency. The expected LNA gain of ~28 dB was observed through the fully-connected interface box.

Fig. 4.

Completed hardware for 16-channel ¹³C data acquisition including (from left to right) previously-constructed frequency-translation unit, coil system, and interface box. The coil is shown with acrylic paneling removed to allow better visibility of the array coil.

Relevant benchtop measurements for the ¹³C receive and transmit elements are summarized in Table II. Average loaded coil Q factors were 60.6 for the large receive elements and 49.8 for the small receive elements and were approximately equal to the unloaded Q factors. We attributed the unfavorable Q unloaded/loaded ratio to both the negligible conductivity of the phantom as well as the generally low unloaded Q factors of the elements.

TABLE II.

Summary of Coil Benchtop Measurments

Coil	S11 (dB)	Active Detuning (dB)	Preamp Decoupling (dB)	Transmit Coupling (dB)	Q
Helmholtz	−47.7	32.4	N/A	N/A	146.9
Saddle	−46.0	36.5	N/A	N/A	156.1
Rx 1	−25.5	45.2	17.6	−34.3	62.5
Rx 2	−24.2	47.9	16.6	−36.2	62.5
Rx 3	−26.5	51.4	16.1	−35.4	53.5
Rx 4	−23.0	48.1	16.0	−37.4	57.7
Rx 5	−28.0	49.6	19.1	−42.3	62.5
Rx 6	−23.3	46.0	17.8	−34.9	57.7
Rx 7	−26.3	50.2	17.2	−49.3	65.2
Rx 8	−23.1	50.1	17.1	−35.0	65.2
Rx 9	−21.3	49.6	17.3	−36.8	68.1
Rx 10	−29.5	49.3	16.4	−35.9	51.7
Rx 11	−26.0	44.5	14.8	−42.5	46.8
Rx 12	−29.0	51.8	15.4	−43.3	50.0
Rx 13	−25.6	48.3	15.1	−46.5	53.5
Rx 14	−34.9	43.4	11.6	−41.8	53.5
Rx 15	−29.7	43.9	16.8	−41.8	46.8
Rx 16	−24.5	46.2	14.8	−40.9	48.4

Note that transmit coupling is shown as the S₂₁ between the detuned receive element and transmit coil operating in quadrature mode. In some cases, receive element coupling was slightly higher to individual transmit elements operated in linear mode though no case exceeded −32.4 dB.

The quality factor measurements from the test elements with various network replacements revealed significant losses associated with some of the added networks. The Q measurements for the large and small elements with all networks replaced by segmenting capacitors were 310 and 273 respectively, much higher than those for the actual array elements. Inclusion of the detuning networks decreased these values only marginally to 280 and 268. However, Q factors were reduced significantly to 137 and 128 for elements containing LCC traps, indicating large losses in these networks. Similarly, Q factors decreased to 121 and 104 in elements containing the match/tune/decoupling networks. Considering these two networks both contain inductors which invariably increase coil losses, a Q factor reduction was unsurprising; however, the magnitude of the reduction was higher than expected.

The coil construction process indicated very little coupling between receive coil elements with no obvious peak-splitting displayed in any configuration. With all other receive elements actively detuned, no tuning adjustments were necessary compared to their initially-tuned isolated state. Likewise, only small adjustments were needed when the receive elements were preamp decoupled but not actively detuned. Active detuning on all array elements measured greater than 43.4 dB and S₂₁ coupling between any combination of detuned receive element and either tuned ¹³C transmit element was less than - 32.4 dB.

Coupling was on average 5 dB higher for the larger receive elements compared to the small elements. Clean S₁₁ peaks were observed for all coils when appropriate detuning/decoupling was applied to the remaining elements.

Average preamplifier decoupling for all receive elements measured 16.2 dB (minimum - 11.6 dB) with matching of all coils better than −17 dB with all other receive elements preamplifier decoupled. Regardless of decoupling values, smaller receive elements required less retuning in this configuration and generally coupled less even when other receive elements were left resonant.

The transmit coil also showed matching of both ports of the ¹³C quadrature coil better than −30 dB and clean S₁₁ plots with the actively-detuned array inserted. Q factors of the Helmholtz and saddle coils were 146.9 and 156.1 respectively in this state. B₁ fields for the ¹³C Helmholtz were 2.8 dB higher than those for the saddle coil at the center of the coil though this difference decreased closer to the saddle elements. Though identical field intensities would be preferable for achieving true circular polarization, this difference was expected due to the significantly greater spacing between saddle elements (19 cm center-to-center) compared to the Helmholtz spacing (7 cm) which was necessary to accommodate the array. Active detuning of the coil was better than 32.4 dB for both transmit coil elements, and the detuned transmit coil did not cause observable peak splitting in any tuned receive coil element.

Regardless of whether any other coil was kept active or detuned, the ¹H transmit/receive coil could be matched to better than −20 dB. Isolation between the ¹H coil and ¹³C transmit elements was 26.4 dB and 23.4 dB for the ¹³C Helmholtz and saddle pair coils respectively at the ¹H frequency. Similarly, isolation was 17.1 dB and 26.6 dB at the ¹³C frequency. Isolation between the ¹³C Helmholtz and ¹³C saddle coils was 11.2 dB and 25.2 dB at the ¹³C and ¹H frequencies respectively.

The ¹³C field mapping shown in Figure 5 revealed relatively homogeneous B₁⁺ fields with less than 2.2 dB variation in average field measured across each slice for all slices that fell within the coil volume. This variation increased to 6.1 dB when including the two slices measured slightly outside the coil volume. Inclusion of the array with/without active detuning enabled decreased the average field in the central slice by 3.8 dB and 14.8 dB respectively due to increased transmit shielding while the array was in the detuned state and due to disruption of transmit coil matching when active detuning of the array was disabled. The decrease in average measured field when detecting B₁⁻ compared to B₁⁺ (i.e. when operating in anti-quad mode) in a central slice was 5.4 dB, indicating that the majority of the transmit power was used to produce the correctly-polarized B₁⁺ field.

Fig. 5.

B₁ field maps obtained from ¹³C transmit coil with the array removed (top) and inserted while actively-detuned (middle). The field maps indicate fairly good field homogeneity throughout the coil volume. A large decrease in field intensity is seen after insertion of the detuned array as characterized by the ratio of the fields before and after array insertion (bottom). This loss was attributed to shielding effects from the array elements and their respective circuitry.

B. Spectroscopy Experiments

Bulk spectroscopy experiments indicating the SNR improvement achieved with the array are shown in Figure 6. The signal level varied slightly between receive elements primarily corresponding to differences in coil element orientations with respect to the main magnetic field. For the bulk phantom, array coil SNR measured after Matlab processing was 198.7 compared to 62.1 for the volume coil acquisition indicating an SNR increase by a factor of 3.2. For comparison, the highest SNR from a single individual receive element was 105.7, and individual elements had an average SNR of 47.8, slightly lower than that of the quadrature volume coil which is expected considering the non-localized nature of the experiment.

Fig. 6.

Bulk spectroscopy experimental results. A greater than three times increase in SNR from the array coil (left) compared a similar acquisition using the volume coil in T/R mode (middle) was seen, indicating the array’s ability to improve SNR over the entire breast volume. The noise data from the scan shows little inter-element coupling as expected from bench measurements with the highest coupling of 13.7% shown between elements 15 and 16 and all other elements showing significantly lower values.

Noise correlation between array elements showed minimal coupling. Average noise correlation between all elements was 2.0% with maximum coupling of 13.7%. No noticeable reliance of noise correlation on coil positioning was observed, presumably due to geometric decoupling of nearest neighbor coils and the spherical arrangement of coil elements which meant that most coils had similar levels of coupling.

For the localized bicarbonate phantom experiments (Figure 6), average SNR of individual receive elements was 24.8. The most proximal element to the phantom (i.e. the Row 1 element) had an SNR of 66.7. Row 2 elements had an average SNR of 24.0, and Row 3 elements had an average SNR of 17.6. In general, the observed SNR for receive elements decreased with increasing distance from the phantom as expected due to typical surface coil sensitivity patterns. This effect was even more pronounced in the most distal small receive elements which had an average SNR of 6.3 and would be expected to have sensitivity patterns least receptive to signal at the phantom location.

IV. Discussion

The low coupling measurements between the transmit coil and array made tuning the array elements insensitive to its positioning within the transmit coils. Similarly, the array was insensitive to phantom loading due to both the low conductivity of the phantom and the low element quality factors. Achieving better sample-noise dominance would be worthwhile to further increase SNR gains but is challenging with sources of loss being the additional circuitry needed for ¹H decoupling and adequate preamplifier decoupling.

Results of quality factor testing of the various test coil configurations revealed that a “pure” resonant element can achieve a significantly higher quality factor than what was achieved in the array elements reported. While necessary components on the receive array, the addition of the LCC trapping networks, and match/tune/decoupling networks in particular, adds significant loss. Considering SNR is known to be proportional to $1/\sqrt{Q_{\mathit{\text{loaded}}}}$[32], these losses have a direct on achievable SNR for cases where the coil is not heavily sample-noise dominant. For this work, the large differences in Q for the fully-assembled elements versus the “pure” resonant elements and lack of sample-noise dominancesuggests that SNR could be roughly doubled if network losses were fully mitigated. As sample-noise dominance is increased and coil noise become negligible compared to body noise, these SNR improvements would become less substantial.

Future work to minimize these losses while maintaining similar levels of decoupling may require substantial array modifications such as changing LCC trap component values and reconfiguring the decoupling network to rely on tuning through receive cable phase length rather than the use of inductors. The issue is only exacerbated by the low frequency associated with the ¹³C nucleus which makes achieving sample load dominance additionally challenging.

Though other groups have reported the need for additional receive element decoupling strategies to maximize SNR in certain X-nuclei arrays [68, 69], these strategies were not pursued in this work due to the low loaded quality factors and coupling coefficients (0.025 for the most tightly coupled coils) of our array elements. While both of these factors should minimize expected SNR gains from further decoupling methods, further study of these effects could still be beneficial. It should also be noted that the low noise correlation values obtained during scanner experiments could be partially attributed to the lack of sample noise dominance. Because coil losses seemed to be the dominant noise source and these losses do not necessarily show the same inter-element correlations as sample noise, noise correlation values may increase as sample noise dominance is approached.Though preamplifier decoupling was generally greater than 15 dB for the array, integration of the preamplifiers onto the coil itself could further enhance SNR and improve preamplifier decoupling values. However, the tight-fitting array geometry made the use of a modular interface box a convenient alternative which could be used for future ¹³C coils.

Ideally, characterization of transmit homogeneity for the ¹³C coils could be performed on the scanner using standard B₁⁺ mapping techniques such as the double-angle method [70],DREAM [71], or Bloch-Siegert shift [72] techniques. However, the inherently low NMR sensitivity of the ¹³C nucleus makes obtaining useful maps using these scanner-based techniques challenging. Use of the benchtop measurement system for measuring transmit fields provided a simple alternative mapping method suitable for the current work. Though the fields could only be measured on this system in an unloaded configuration, the light loading effects of the phantoms used and low frequency of the ¹³C nucleus suggest that the benchtop-measured maps provided an accurate representation of the fields produced during scanner experiments for this work.

Further SNR improvements are expected if the use of the proton decoupling [73] and the Nuclear Overhauser Effect (NOE) [74, 75] is pursued, though these were not attempted due to the scope of the project. It is likely that improvements in ¹H coil transmit efficiency would be necessary to make proton decoupling possible, but this could be accomplished by modifying the ¹H coil to use a quadrature implementation.

Though high power requirements were noted for the ¹H coil, this had also been seen previously with similar 7T coil designs [54, 55]. Reduced ¹H B₁⁺ efficiency was likely due to shielding effects of the added ¹³C array and transmit coil system [57] as well as impedance mismatches at the element feed points and standing wave effects due to use of the FCE technique [53]. These issues could likely be somewhat alleviated by modification of the current ¹H coil to a quadrature design to improve efficiency, but these modifications were not undertaken at this point considering the focus of this work was primarily on the ¹³C coil system. However, these alterations would undoubtedly be worthwhile for future studies.

Further electromagnetic simulation work would be needed to confirm how best to improve coil efficiency and would also be especially beneficial in considering potential SAR implications of the present coil. SAR guidelines are often a limiting factor in ¹H 7T studies [24–26], especially those involving SAR intensive techniques such as proton-decoupling. Simulated SAR maps must be created and verified in further phantom studies before undertaking in vivo studies to verify that the coil can stay within both local and global SAR guidelines. Considering the possibility of SAR-related issues for ¹H coils operating at 7T [24–26], this is an important concern, and should be examined closely to ensure patient safety. However, it should be noted that while the FCE technique can lead to decreases in B₁⁺ efficiencies due to impedance mismatches at the element feed point and standing wave effects [53], these losses are not deposited in the patient are not expected to impact SAR efficiency [54]. A loss of efficiency was also noted for the ¹³C transmit coil system after insertion of the array. Though SAR limitations are less likely a major concern at this frequency, these effects still need to be more well-characterized through simulation work verified through bench and phantom measurements before human studies are pursued.

The bulk phantom spectroscopy experiments indicate the significant SNR benefits of the array over the breast volume. The increased SNR when compared to volume coil reception would be essential for decreasing voxel sizes for localized spectroscopy experiments, and the array geometry means that deeper lesions may be studied more effectively than when using single loop elements for reception.

The results from the array localization experiment using single elements indicated a degree of signal localization ability from the array elements even when performing bulk spectroscopy experiments. The relationship of spectral SNR to coil element positioning suggests that looking at spectra from individual elements could provide a method of coarse signal localization. The low inter-element coupling needed for this was further demonstrated by low array noise correlation values which suggest low g-factors if accelerated, hyperpolarized studies are pursued. True gradient-based spectral localization techniques were not attempted in this work but could be attempted in future work. With the frequency-translation approach used in this work, the scan procedure is performed with the stock system operating as if performing a multichannel ¹H scan, so gradient strengths and pulse bandwidths would need to be adjusted to account for the difference in the ¹³C and ¹H gyromagnetic ratios, but adjusting for these changes would be straightforward.

V. Conclusion

The array coil demonstrated superior SNR performance compared to a typical volume coil covering a similar field-of-view. Furthermore, the array design adds the potential benefit of coarse localization through the distinct reception profiles of the individual receive elements. The array performance was demonstrated both on the bench and through phantom spectroscopy studies.

The most obvious future direction for this work is in vivo studies assessing lipid profiles of various populations to better understand the relationship of the lipidome to cancer. The ability of the coil to potentially provide an objective, noninvasive means of lipid analysis makes this particularly advantageous when compared to dietary surveys or biopsies. We hope to pursue such studies, which will of course require IRB approval and a detailed characterization of the B₁⁺ fields and SAR distribution through both electromagnetic simulations and further phantom studies at both the ¹³C and ¹H frequencies. The high channel count and low coupling between array elements also suggest that the coil would be well-suited for use in accelerated dynamic studies using hyperpolarization techniques such as DNP [76, 77] to retain polarization while providing high temporal resolution.

Fig. 7.

Spectra from example receive elements and the combined array spectra using the localization phantom. A bicarbonate spectrum from each row of receive elements is shown with spectral intensity levels generally decreasing for receive elements more distal to the phantom. Note that scaling was kept consistent between all spectra for easy comparison. The results indicate that spatial sensitivity patterns of individual receive elements may be used for coarse signal localization even without using localized sequences.