Eight-channel parallel transmit-receive system for 7 T MRI with optically controlled and monitored on-coil current-mode RF amplifiers

Abstract

Purpose:

To demonstrate a practical implementation of an eight-channel parallel-transmit system (pTx) for brain imaging at 7 T based on on-coil amplifier technology.

Methods:

An eight-channel parallel transmit-receive system (pTx-Rx) was built with optimized on-coil switch-mode current RF power amplifiers. The amplifiers were optically controlled from an eight-channel interface that was connected to a 7 T MRI scanner. The interface also optically received a down-converted version of the coil current sensed in each amplifier for monitoring and feedback adjustments.

Results:

Each on-coil amplifier delivered more than 100 W peak power and provided enough amplifier decoupling (< −15 dB) for the implemented eight channel array configuration. Phantom and human images were acquired to demonstrate practical operation of this new technology in a 7 T MRI.

Conclusion:

Further development and improvement of previously demonstrated on-coil technology led to successful implementation of an eight-channel pTx system able to deliver strong B₁ fields for typical brain imaging applications. This is an important step forward towards implementation of on-coil RF amplification for high field MRI.

Introduction

Ultra-high field MRI offers unique opportunities for neurological studies that may allow the detection of pathology with increased resolution, sensitivity and contrast. However, some technological challenges remain that affect image quality, safety, and robustness. A major challenge is the compensation of B₁ field inhomogeneities that result from the interaction of the shorter wavelength excitation with the tissue (1). These inhomogeneities lead to spatially varying contrast, and more focal energy depositions that may impose safety limitations on some applications.

Parallel transmission (pTx) (2–4) has been the preferred approach for shimming the B₁ field (5–10) as well as for control of the transmission electrical field responsible for tissue heating (11, 12). To facilitate pTx, new hardware has been developed that allows independent control of the amplitude and phase of the multiple fields generated in a Tx coil array (13–18). Recently, pTx hardware has been fully integrated into commercial high-field scanner infrastructure (19–21). Like most pTx hardware, this is designed with a conventional, and not necessarily optimal, Tx architecture similar to that used for quadrature Tx coils (22). In this approach, each coil is tuned and impedance matched for maximum power transfer to a nominal load. Power is transmitted to each coil from remote voltage-mode amplifiers through long coaxial cables (23), and monitored with directional couplers also remotely connected to the coils (24). This hardware is expensive and lossy, and B₁ control is limited because coil current is affected by complex, subject dependent factors.

Different methods have been presented to improve pTx systems built with conventional Tx architecture. To reduce cable losses and facilitate cable decoupling, the amplifiers have been redesigned and located in the scanner room near the bore (25–28). To compensate for load variations and improve system performance, automatic matching and tuning of pTx coils were demonstrated in the MRI scanner (29–31). Despite these measures, in practice most pTx coils are not dynamically adjusted to individual subjects, and therefore, are often operated suboptimally.

By moving away from conventional Tx architecture, we previously demonstrated that optically controlled and monitored on-coil current-source RF power amplifiers (RFPA) allow the implementation of a pTx chain for 7 T MRI brain imaging with excellent B₁ control and reduced power losses at a lower implementation cost (32–34). Minimal load sensitivity and elimination of cable losses were demonstrated by direct optical control of the coil current (32). As shown in a previous four-channel parallel transmit-receive system (pTx-Rx) implementation, this technology also allows direct sensing of coil current, information that can be used for safety monitoring and feedback (33, 34). Here, we further developed this method and improved its fiber-optic control and monitoring to allow whole-brain 8-channel pTx.

Methods

Hardware Design and Implementation

The eight-channel pTx system (Figure 1) was based on our previous 300 MHz on-coil current-mode RFPA design (33). The new design aimed to minimize wired connections, circuitry, and heat dissipation to allow a higher channel count implementation. The main focus was on the envelope amplifier and monitoring circuitry located on top printed circuit board (PCB) (Supporting Information Figure S1). In addition to the optical envelope and carrier control signals implemented previously (32–34), sync (CS) and clock (SCLK) control signals for on-coil digital-to-analog (decoding of envelope signal) and analog-to-digital (encoding of RF sensed signal) conversions were also implemented optically in this new prototype. This allowed the elimination of buffering circuitry, used in previous designs, to distribute these control signals through wired connections to each amplifier board. To further reduce power dissipation on the envelope amplifier PCB, down-conversion of the RF current sensed on the amplifier board was implemented with a RF passive mixer (pSemi PE4141) instead of the active mixer used previously (33). Furthermore, all monitoring electronics were RF shielded to avoid degradation of the small signal from interference with the high-power RF signal. Miniaturization allowed improvements without increasing the RFPA size (80[L]× 53[W] × 20[H] mm³).

Figure 1:

(a) Schematic of the pTx setup in the MRI equipment and magnet rooms. (b) Plug-in eight-channel interface box. (c) Eight-channel pTx-Rx system with sliding RF shield. (d) Monitored Tx signals displayed in control room.

In this first eight-channel implementation with focus on the performance of the electronics, each amplifier was connected to a 6 cm diameter circular loop operated in combined Tx-Rx mode through a small PCB that contains a Tx-Rx switch and a low-noise amplifier (LNA) for amplification of the NMR signal (35). Inter-element decoupling was achieved through the low input impedance of the LNA and the high output impedance of the RFPA during signal reception and transmission, respectively. Coils and electronics were assembled on a 265 mm inner-diameter cylindrical former with a sliding RF shield to reduce interference with the surrounding hardware in the 7 T scanner bore.

We also evaluated the performance of the amplifier driving a larger, 131 × 82 mm², rectangular coil. These coil dimensions were chosen to investigate the potential for future field penetration/sensitivity improvements in 8-channel arrays with overlapping elements on this same 265 mm diameter former.

The amplifiers were controlled by an in-house built eight-channel pTx interface. The essential part of this interface is a modular two-channel control board that performs vector modulation, carrier recovery, and envelope detection and digitization of a single RF input pulse (32, 36). Stacking four of these boards allows generation of eight optical carriers, eight optical envelopes, four CS and four SCLK signals transmitted to each amplifier through 1:2 or 1:4 custom plastic optical fiber (POF) splitters. With this, the interface allows independent amplitude and phase control for eight optically controlled on-coil amplifiers from a single RF input signal. The interface also receives the digital, optical down-converted RF coil currents sensed in each amplifier, which are sent to a controller (Linux computer), through USB ports, for monitoring and feedback (33). The controller updates the 16 in-phase and quadrature (IQ) baseband signals of the vector modulator, sent to the interface through a 68-pin connection (VHDCI cable), which set amplitudes and phases for each RF pulse. All electronics were assembled in a 2U rackmount enclosure of dimensions 430(L) × 87(H) × 328(D) mm³ with a commercial switch-mode power supply (5 V, 5 A) for direct plug-in to a standard outlet. The interface was connected to the scanner (Siemens Magnetom 7T) RF small signal (scanner RFPA input), 10 MHz reference and RF unblank. A schematic diagram and pictures of the optical transmit setup are shown in Figure 1.

Bench Measurements

Peak power delivered to the coil and coil current were confirmed with a B₁ field calibrated probe measurement. To evaluate device heat dissipation, a single amplifier was set to deliver 100 W peak power with 10% duty cycle excitation during 2 min. Immediately after, temperature was measured by taking an infrared picture (E6 Thermal Camera; FLIR Systems, Wilsonville, Oregon, USA). Dynamic range of the down-converted RF sensed current was measured at the RF mixer differential output using a 3.5 GHz active differential probe (N2751A InfiniiMode, Keysight, Santa Rosa, California, USA), connected to an oscilloscope (Infiniium DS0-S254A, 2.5 GHz Keysight). In the array, transmit decoupling was measured by setting the B₁ amplitude of one channel to 22 μT (at coil center) and all others to 0 μT. B₁ field was measured with the calibrated probe coupled to each channel. Receive decoupling was initially assessed on the bench through S21 measurements using a network analyzer (NA E5061B 100 kHz-3 GHz, Keysight, Santa Clara, California, USA) and during MRI experiments through noise correlation measurements as described below.

Phantom MRI

Images of a large spherical silicon oil phantom (24 cm diameter) and a head shaped phantom containing a solution with similar electrical conductivity and permittivity as average brain tissue (37, 38) were acquired using a gradient echo sequence (TR/TE=250/10 ms, FOV= 350 × 350 mm², matrix size 128 × 128 and slice thickness 5 mm) while the RF transmit current in each loop was monitored in real time (Figure 1d). Amplitudes and phases were set to generate a circularly polarized (CP) B₁ field in both forward (rotating with the spins) and reverse directions. Coil transmit efficiency of the small loop array was estimated from flip angle maps of the brain phantom acquired with the AFI method (39) during CP mode excitation. For comparison, a similar experiment was performed using a commercial 1Tx/32Rx head coil (Nova Medical Wilmington, MA). To assess preamplifier decoupling, the array was loaded with the brain phantom and noise images were acquired with all channels receiving simultaneously and Tx voltage set to zero. Inter-channel coupling was evaluated by calculating cross-correlation coefficients from these data. B₁⁺ field distribution was evaluated in the scanner with the larger Tx loop placed 1 cm above a spherical oil phantom with diameter similar to the human head (175 mm). In this setup, signal was received with a separate Rx volume coil. B₁⁺ distribution was estimated from large flip angle excitation (> 3 full cycles) using a 2.2 ms pulse while bias voltage of the amplifier power stage was 38 V (Vdd in Figure S1) for full power operation. The resistive loads seen from the amplifier with the brain phantom were ~3 Ω and ~7 Ω for the small circular loop and large rectangular loop, respectively, both assembled on the cylindrical former.

Preliminary volunteer MRI

A preliminary human study was performed under IRB approval with peak and average power restricted by both software and hardware. Restricting the supply voltage to each amplifier power stage limited total peak output power to 143 W. A 200 ms minimum TR was hard-coded in the pulse sequence to limit the excitation duty cycle to 1%, such that maximum average power could not exceed 1.5 W. For additional safety, supply current to the amplifier power stage was limited by a time-lag fuse and sensed by a shunt resistor connected to the oscilloscope located in the electronics room and accessed remotely from the control room. For this preliminary study, channel amplitudes and phases were set to generate a CP B₁ field. Axial, coronal and sagittal slices of the brain were acquired with a gradient echo sequence (TR/TE=250/10 ms, FOV= 240 × 240 mm², matrix size 128 × 128 and slice thickness 2 mm).

Results

The implementation of passive mixing allowed lower supply current (~1 A total for the array compared to ~1.8 A with active mixing) to the monitoring, envelope detection and envelope feedback electronics and yielded a 10 dB increase in dynamic range (~45 dB total) of the monitored RF current. As can be seen from Figure 2, maximum peak power delivered to the coil exceeded 100 W. The maximum temperature measured on the backside of the PCB was 65 °C after the amplifier delivered 100 W pulse power to the coil at 10% duty cycle for two minutes (Fig. 2). As expected, the hot spot location was right below the power enhancement-mode GaN field effect transistors (eGaN FETs) assembled on the PCB top layer. Without any heat sink or forced air cooling, heat was convectively dissipated from PCB copper to still air.

Figure 2:

Amplifier output. (a) Peak power with R_L=7.7 Ω and (b) temperature of hotspot on the backside of PCB after running the amplifier at 100 W peak power (red dot in panel a) with 10% duty cycle excitation for 2 min.

An axial B₁⁺ map of the brain phantom acquired with the eight-channel pTx in CP mode excitation is shown in Supporting Information Figure S2. From these maps, the estimated B₁⁺ efficiency at the center of array was ~0.3 μT/√W. Maximum B₁⁺ at the center of the 175 mm spherical oil phantom was around 7 μT for single channel excitation with the larger rectangular loop (Supporting Information Figure S3), which was more than threefold the B₁⁺ value generated at same depth with the smaller circular loop. Tx decoupling and Rx noise correlation are shown in Figure 3. The Tx element decoupling matrix is shown in Figure 3a, indicating that decoupling was below −15 dB. Channels 4 and 7 were “leaking” RF signal due to a DC bias error in the control interface, see Discussion. The noise correlation matrix shows overall low correlation coefficients, with maximum correlation between elements 5 and 6. On the bench, S21 for all element pairs was lower than −18 dB. Single channel and eight-channel simultaneous transmission was successfully performed with the array. Figure 4a shows images of the 240 mm diameter oil phantom reconstructed from the eight received signals while RF power was transmitted with each channel individually (outer images), and with all channels simultaneously (center image). Eight channel amplitude and phase control with the in-house built pTx optical interface were demonstrated by acquiring images of the brain phantom using a forward (top) and reversed (bottom) CP excitation mode as shown in Figure 4b. Single slice images of the in-vivo brain acquired with amplitudes and phases set for CP mode excitation (without B₀ shimming) are shown in Fig. 5. This demonstrates that the array yields homogenous artifact-free coverage of the upper brain.

Figure 3:

Tx and Rx element decoupling by amplifier and preamplifier methods respectively. (a) The eight-channel Tx decoupling matrix was measured with a calibrated probe in a benchtop experiment with one channel transmitting at the time. (b) Eight-channel noise correlation matrix obtained in the MRI with transmit voltage set to 0 V.

Figure 4:

Phantom MRI. (a) Image of an oil phantom obtained with each channel transmitting individually (outer images) and with all channels transmistiing simultaneously (center image). (b) Sagittal slice through the brain phantom acquired with a forward and reversed CP Tx mode.

Figure 5:

Coronal, axial, and sagittal MRI of human brain.

Discussion

Further optimization of optically controlled on-coil amplifier technology allowed implementation of an eight-channel pTx system for 7 T MRI with a minimal number of electrical connections. Optical clock and chip-select inputs to each analog-to-digital converter (ADC) and digital-to-analog converter (DAC) control on the amplifier board allowed a more stable digital data transmission than in the previous four-channel implementation (33). The local oscillator (LO) signal used for down-conversion of the sensed coil current was the only electrical input to the amplifier. The LO signal was received optically on the array side and converted back to an electrical signal and buffered to multiple differential outputs. Balanced LO outputs allow connections to amplifiers using twisted pair instead of coaxial cables. Even though optical splitting of the ADC/DAC control signals allowed us to reduce the number of fiber connections, we found that POF splitters were not robust with the frequent disconnections/connections that were required for our non-permanent installation.

Additional shielding of the electronics was necessary for accurately monitoring coil current at high power operation. Even with the shield added to each amplifier board we detected some residual instability of the monitored signal from some channels. This was resolved by adding an amplification stage following optical reception of the LO signal. Accurate monitoring of coil current phase and amplitude per channel during high power transmission is advantageous for SAR supervision (42).

Because of the amplifier decoupling method and current source amplification (16, 43), the coil array operates simply as a set of resonant loops without any matching, decoupling and detuning circuitry. The optimal geometry of these loops as well as other possible types of RF antennas will be subject of further study. For this first eight-channel implementation we chose simple non-overlapped circular loops. Circular loops allowed easy calibration of coil current based on B₁ field measurements. However, field penetration of the implemented array was poor, having the center of the array at a distance larger than 4 times coil radius. Field penetration was improved by connecting the same amplifier, without any modification or tuning, to a larger rectangular loop. Results suggest that a well decoupled array built with these loops will provide higher B₁⁺ efficiency at the center (>0.5 μT/√w) as reported elsewhere for other 7 T head coils (40, 41) and as experimentally obtained with a commercial head coil (Nova Medical Head Coil 1Tx/32Rx). B₁+ generation at adequate power levels seems possible with this on-coil technology, at a cost of less than $5 per Watt delivered to the coil. Considering only price of RFPAs, this is at least four times lower than the price per Watt generated at the coil plug in a 7 T MRI with a commercial remote voltage-mode 7 T amplifier (COMET Technologies USA Inc.).

While on-coil amplification allows high power efficiency (in this implementation >70%), a potential disadvantage is that power losses, and associated heat generation, occur inside the bore. These losses and consequent heat generation are dominated by the FET on-state resistance (R_FET__ONI²). At full power and 10% duty cycle, the temperature of the small PCB area below FET footprints exceeded 60 °C after 2-minute operation. Given the small area, and consequently high thermal resistance, higher temperatures could result in device failure. Besides performance of the electronics, it is important to consider how heat dissipation from the electronics, especially at higher duty cycles and/or longer excitation cycles, will impact temperature inside the coil housing. Here, additional thermal insulation between amplifier and inner side of the former (head side) was achieved by mounting the amplifier PCB ‘floating’ on an acrylic beam (1 cm height) attached to the outside of the former (electronics side). Further heat dissipation could be achieved by forced air injected in this 1 cm space through tubing.

A trade-off with the use of on-coil RFPAs is a higher demand on the valuable bore space near the object under study. This requires the use of compact electronics and its careful layout. Thanks to miniaturization, all electronics and RF shield fitted inside a modestly-sized housing (380[L]× 378[W] × 386[H] mm³) and it is expected that integration with a separate 32-channel or even 64-channel receiver will not be problematic.

When not transmitting, some amplifiers were leaking RF signal due to a non-zero envelope amplitude (Figure 3a). This leak was the result of a small DC bias error in the corresponding vector modulator baseband inputs. This error propagates as a nonlinear error at the modulator output and needs to be compensated. As shown previously, we can use the optical feedback signals to compensate for phase errors, residual channel coupling and safety monitoring (33). Following the same approach, a fast algorithm that fits the monitoring signals to update the IQ control to automatically compensate for DC bias errors and any other hardware deviations is under development. Moreover, the modular design of the optical control interface (32, 33) allows easy expansion to a 16-channel Tx system by stacking additional two-channel control boards, provided the power supply is upgraded to the higher current load and space and ventilation of the electronics enclosure are appropriate for the added boards.

The use of on-coil RFPAs may be particularly advantageous for use at field strengths of 7 T and above. At the higher RF frequencies associated with these fields, the implementation of a pTx system using remote amplification becomes more challenging due to higher losses and more difficult control of common mode currents in cables, and higher variations in coil loading. The latter complicates accurate control of RF phase and amplitude. Optically controlled on-coil current-source RFPAs eliminate cables losses and coupling, and reduce load sensitivity by more direct control and sensing of coil current (33). On the other hand, with increasing channel count, the technology requires valuable bore space and demands additional engineering to handle in bore heat dissipation. Moreover, the design of a RFPA prototype that can deliver more than 100 W of peak power with efficiency around 70% at frequencies above 300 MHz is challenging mainly because of parasitic elements in PCB layout and FET switch. We expect advances in FET technology and miniaturization to ease these engineering challenges, and make on-coil amplification a competitive technology for the implementation of pTx systems at MRI field strengths of 7 T and above.

Conclusion

Further development and improvement of on-coil technology led to a successful implementation of an eight-channel optically controlled and current-driven pTx system that can deliver sufficient power for most brain imaging applications. As in our previous four-channel implementation, B₁ per channel at the center of each loop was directly controlled and sensed. Channel decoupling for parallel transmission and reception was successfully implemented exclusively by amplifier impedance methods. Artifact-free brain images were acquired on a human volunteer.

Supplementary Material

supp fig1-3

Figure S1: Schematic and photo of the updated 7 T on-coil amplifier with optical ADC/DAC control and passive mixing for down conversion of RF sensed coil current.

Figure S2: Axial B₁⁺ map (left) of the brain phantom with amplitudes and phases of the eight-channel pTx set for CP excitation and current per amplifier ~ 2 A (right). The distance between the 6 cm diameter loop and the center of the phantom is indicated in the B₁⁺ map.

Figure S3: Estimated B₁⁺ field penetration (left) and gradient-echo image (right) obtained with a large flip angle excitation. The amplifier was connected to a 131 × 82 mm² rectangular loop located 1 cm above the silicon oil phantom.