FULL PAPER Magnetic Resonance in Medicine 74:589–598 (2015)

Comprehensive RF Safety Concept for Parallel Transmission MR Ingmar Graesslin,1* Peter Vernickel,1 Peter B€ ornert,1 Kay Nehrke,1 Giel Mens,2 Paul Harvey,2 and Ulrich Katscher1 Purpose: The goal of this study is to increase patient safety in parallel transmission (pTx) MRI systems. A major concern in these systems is radiofrequency-induced tissue heating, which can be avoided by specific absorption rate (SAR) prediction and SAR monitoring before and during the scan. Methods: In this novel comprehensive safety concept, the SAR is predicted prior to the scan based on precalculated fields obtained from electromagnetic simulations on different body models. The radiofrequency fields and the global and local SAR are supervised in real time during the scan. This concept is integrated into a 3 T pTx MR scanner and validated experimentally. Results: Phantom and in vivo experiments successfully validated the basic feasibility of the real-time SAR supervision concept. Supervising the SAR minimizes SAR overestimation. Monitoring the radiofrequency fields allows the detection of unsafe radiofrequency situations for the patient, which a SAR supervision system alone cannot detect. Conclusion: This study demonstrates safe scanning in a pTx system. This new safety concept is also applicable for field strengths above 3 T and represents an important step toward safe operation of pTx systems. Magn Reson Med 74:589– C 2014 Wiley Periodicals, Inc. 598, 2015. V Key words: comprehensive safety concept; radiofrequency patient safety; parallel transmission; specific absorption rate prediction; specific absorption rate supervision; waveform supervision

INTRODUCTION MR applications at main field strengths of 3 T and above suffer from radiofrequency (RF) field inhomogeneities (1) and, therefore, also from signal and contrast inhomogeneities. This phenomenon hampers the reliable evaluation of diagnostic images. To improve the signal and contrast homogeneity, parallel transmission (pTx) allows shaping the Bþ 1 transmit (Tx) field in a patient-specific way, for example, via basic RF shimming (2–4) and advanced transmit SENSE applications using multidimensional RF pulses (5–8). However, for pTx to be used 1 Department Tomographic Imaging Systems, Philips GmbH Innovative Technologies, Research Laboratories Hamburg, Hamburg, Germany. 2 MR Business Innovation Unit, Philips Healthcare, Best, The Netherlands.

*Correspondence to: Ingmar Graesslin, Dipl., Ing., Philips GmbH Innovative € ntgenstrasse 24-26, Technologies, Research Laboratories Hamburg, Ro 22335 Hamburg, Germany. E-Mail: [email protected] Received 26 January 2014; revised 10 July 2014; accepted 4 August 2014 DOI 10.1002/mrm.25425 Published online 22 August 2014 in Wiley Online Library (wileyonlinelibrary. com). C 2014 Wiley Periodicals, Inc. V

in clinical practice, current protocols for (single channel) specific absorption rate (SAR) management must be thoroughly revised to prevent harm to the patient from RFinduced tissue heating. In pTx, SAR is no longer proportional to the total forward RF power, as is the case for single-channel systems (9). Recent studies have revealed significant differences between single-channel and multichannel RF transmission that are highly relevant for patient safety, stimulating the search for new, satisfactory SAR management concepts (9–18). Properly managing SAR during a scan requires effective SAR prediction. Most of the proposed approaches for SAR prediction before a scan focus on rather conservative estimation techniques (11,19–22). These approaches often strongly overestimate SAR, as they are based on worst-case assumptions or require an overestimation due to tolerance margins to prevent scan termination in case of moderate patient movement. The SAR overestimation would hamper the application of pTx in a clinical environment. To ensure patient safety during the scan, this must be performed as estimated during the SAR prediction process. To monitor the coherence between the predicted and actual SAR, various supervision approaches have been proposed (14,15,19,23). These are primarily based on RF power supervision using directional couplers partly requiring two monitoring channels per Tx channel. As an alternative, RF current and voltage sensors have been proposed, which can also be used for system performance and diagnostic measurements (24). In a single-channel Tx system, the peak and average power are usually monitored to ensure patient safety during the scan. However, monitoring each Tx channel individually in pTx systems using directional couplers (19) may not ensure safety in all circumstances, as it does not consider the phase relation among Tx coil elements, which is relevant for the SAR. Therefore, this method was extended by monitoring forward and reflected power considering amplitude and phase, allowing the safety margin to be reduced (14). However, also in this case, the global phase offset among coil elements remains unknown, unless they are specially calibrated. The in vivo global SAR prediction approach by Zhu et al. (23) monitors forward and reflected power using directional couplers. It furthermore compares the predicted RF values (based on a prescan in situ calibration scheme) with the values measured during the scan; however, it cannot predict or verify the local SAR. One RF supervision approach uses channel multiplexing (15). While this is cost-effective, simultaneous supervision of

589

590

all channels is no longer possible, so some potentially unsafe situations might not be detected. Overall, existing supervision methods either strongly overestimate SAR or do not reliably detect all situations unsafe for the patient, hampering the adoption of pTx in a clinical environment. This article presents in detail a novel, comprehensive RF safety strategy for pTx MR. The new concept allows SAR prediction before the scan and supervision during the scan, therefore, reducing SAR overestimation while still guaranteeing patient safety. This concept consists of three steps: (i) SAR prediction, based on preprocessed numerical simulation data; (ii) a preparation phase, where active decoupling and load compensation are used to ensure the predicted SAR matches the actual SAR applied to the patient; and (iii) supervision during the scan, which ensures that the global and local SAR limits are not exceeded and that the RF waveforms applied are correct as predicted. The RF safety concept was developed and integrated on a 3 T MRI scanner equipped with an 8-channel parallel transmit/receive (Tx/Rx) body coil, and was verified in phantom and in vivo experiments. This new concept has the potential for a high permissible RF duty cycle due to precise SAR supervision, improves the detection of SAR limit violations and unsafe conditions, and reduces the number of unnecessary false-positive scan interruptions. METHODS Comprehensive RF Safety Concept and Implementation The comprehensive safety concept proposed here is described along its implementation on a 3 T MRI system (Achieva, Philips Healthcare, Best, The Netherlands) that was extended with eight independent RF transmit channels (25). As illustrated on the right hand side of Figure 1, for each RF transmit channel, the host system generates a description of the demand RF waveforms. The DAS converts it to RF transmit signals, which are then amplified by two 4-channel solid-state RF amplifiers (AN8135P8, 5kW peak, Analogic Corp., Peabody, MA). Each channel has its own power-monitoring unit for peak and average power detection. Circulators are integrated into each channel to protect the RF amplifier modules against reflected and coupled power from other channels. The amplified signals are then routed to a multichannel Tx/Rx RF coil, with each coil element having its own pick-up coil (PUC), a shielded figureeight design, to sense the actual RF current in the corresponding transmit coil element, while being insensitive to currents flowing in neighboring coil elements (26). The PUC and the coil element should be geometrically fixed. Moreover, it is essential that if the signal is “nulled” at the sensor, this corresponds to no field contribution by the transmit element or at least there is a very good correspondence between simulation and actual coil, which might not be easily achieved at higher field strength. Each PUC is connected to a monitor, essentially an MR receiver, which samples the measured signal and sends it for further processing to the supervision system.

Graesslin et al.

The proposed and implemented RF safety concept for pTx MRI systems has three major components (Fig. 1): i. Predicting the parallel Tx SAR prior to each scan and initiating the scan only when the SAR limits are met; ii. Measuring the system matrix (the coupling among Tx coil elements) for active decoupling and mapþ ping Bþ 1 to adjust to the prescribed B1 , thus ensuring the predicted SAR matches the actual SAR; and iii. Supervising patient safety during the scan and optionally terminating the scan in case of a hazardous situation for the patient. The prerequisites for a successful implementation of the concept are preparatory steps of electromagnetic modeling and system calibrations at the time of installation.

Scan Planning To prevent execution of scans exceeding the SAR limits defined by guidelines (27,28), the global and local SAR are predicted after the scan planning, as part of the scan parameter validation, shown in Figure 1a. If the predicted SAR exceeds the allowed limits, the sequence parameters, for example, the repetition time TR or maximum allowed Bþ 1 field amplitude, must be adapted accordingly. The SAR prediction is based on E and B, which are precalculated using the finite-difference time-domain method (29) with the commercial solver package XFDTD (REMCOM Inc., State College, PA). The 8-channel 3 T body coil (30) was modeled assuming ideally decoupled elements using one simulation per Tx element. The body coil was loaded with three different body models based on a priori measured segmented water-fat separated images (31). The models were created from healthy volunteers (male, age 32 to 40 years) in supine position with the arms parallel to the torso. These models had a 5-mm spatial grid resolution and were moved through the coil in steps of 10 cm to obtain the fields for different body positions (32). The simulated electric fields (E) and magnetic fields (B) of the loaded coil were normalized to the individual element currents of the empty coil, which produce an RF excitation field amplitude (Bþ 1 ) of 1 mT in the isocenter of the coil. These normalized fields of the models were calculated and stored in Q-matrices (6,31,32), which contain all non-RF pulserelated field and patient model information. In a subsequent step, they were significantly compressed (33–35), which has also been proposed in the context of SAR supervision (18), allowing an overestimation error of maximum 5%. This entire procedure is a one-time preparatory step. Prior to the scan, the appropriate Q-matrices of one preprocessed biomesh model are manually selected from the Q-matrix database (32,35,36) according to scan- and patient-specific information such as weight, sex, and position with respect to the Tx coil array. As part of the scan planning and parameter validation process, the host system obtains a prediction of the SAR from the

Comprehensive RF Safety Concept

591

FIG. 1. Comprehensive RF safety concept and implementation. Left: The extended parts of the control logic for a multi-Tx system are shown in the gray box surrounded by broken line. Those are elements of the (I) scan definition and planning, (II) preparation phase, and (II) actual scan execution (a) SAR prediction during scan definition and planning, (b) determination of load compensation and decoupling parameters during preparation phases, and (c) RF and SAR supervision during scan execution. Right: A simplified scheme of the integration of the safety concept into the system hardware.

supervision system based on the scan definition, and validates it against the limits. Preparation Phases The preparation phases aim to match the defined scan parameters (e.g., prescribed) to those of the actual scan,

thus cross-calibrating the SAR simulation and physical MR hardware. As a prerequisite for the preparation phases, the Tx drive scale and the PUC signals must be calibrated to the signal, which corresponds to a Bþ 1 of 1 mT, quantified via MR measurements with a small oil phantom in the

592

isocenter with active decoupling enabled. This calibration step must be carried out once during installation time and establishes the correspondence of the electromagnetic field simulations with the physical setup. The following data are determined to serve as references for subsequent safety checks: the system matrix Asys, which describes the coil loading, mutual inductive coupling among the Tx coils (or coil elements) (37,38), and load factors for several humans of different weights. In the first step, the load factors are calibrated, and the coil coupling is determined (Fig. 1b) by measuring the Asys with the patient located in the scanner. As the calibration is element-current-based, the RF field generated by one Tx element is proportional to the element current. Asys is measured for each patient and consists of Nc Nc complex values, the element currents, and is used for active decoupling. The matrix is acquired by transmitting a low power RF pulse sequentially on each Tx element and simultaneously sampling the induced signals of each PUC, as shown in Figure 2 for one RF channel. For the measurements to be less sensitive to interRF pulse variations and other potential error sources, 50% of the samples centered on the maximum of the RF pulse are averaged. To compensate for load-induced changes and residual coil coupling, the inverse system matrix A1 sys is applied to all pTx samples before transmission for the entire scan session (37). To verify that the system is not malfunctioning, the preparation phases include a consistency check of the patient-specifically measured Asys and load factors. If the load factors per coil element are less than 1 or exceed the maximum of the reference load factors per coil element acquired from multiple volunteers at installation time, this is considered to be a fault condition. In this case, the scan is aborted before execution (Fig. 1b), and the operator is informed. The scan is also aborted in case the magnitudes of the normalized Asys, measured during the preparation phase, are not in the range of the minimum and maximum of each corresponding off-diagonal element of the normalized reference Asys. The termination criteria are coil and system dependent and may need to be fine-tuned to the actual setup. In the second step, the Bþ 1 is adjusted to the prescribed value by determining a global scaling factor. In the current implementation, this is done via very fast Bþ 1 mapping using the dual refocusing echo acquisition mode approach (39), measuring a single transversal slice per Tx element. The total acquisition time was 8 s.

Scan Execution Monitoring not only the SAR (16) but also the transmitted RF fields during the scan (40); Fig. 1c ensures the validity of the prescan SAR prediction. The PUC signals are monitored at a dwell time of 6.4 ms, the resolution at which the RF samples are defined, with RF transmission and PUC signal reception synchronized. The PUC samples are sent to the supervision system (Fig. 1), which superposes the preprocessed electric fields from the Q-matrix database according to

Graesslin et al.

FIG. 2. PUC monitoring. a: The transmitted RF signal (green) received by the PUCs (red) is sampled by a dedicated monitoring system during RF transmission. Only one of the RF transmit and PUC signals is shown as an example. At the echo-time (TE), the actual MR response (blue) is sampled by the DAS. b: Schematic of the RF coil setup with the integrated PUC connected to the monitoring hardware, which samples the PUC signal to be supervised.

the received PUC samples to calculate the “supervision SAR” in real time. The SAR calculation is performed by highly optimized software using the parallelized Compute Unified Device Architecture (CUDA, Nvidia Corporation) and running on a high performance graphics card (EVGA GeForce GTX280SSC, EVGA Corporation) with 240 parallel processors in the supervision system (Intel Xeon processor 5430 operating at 2.67 GHz with 16 GB of system memory). The SAR calculation time only depends on the length of the RF pulses and is the same for all RF pulses including complex tailored RF pulses. For example, the global and local SAR calculation for a 4 ms long pTx excitation takes about 0.6 ms for the PUC samples of all 8 RF pulses. The SAR supervision uses a 10 s and 6 min sliding window, as defined by the IEC regulations. If the local SAR is exceeded, the system waits a certain time (i.e., a few hundred milliseconds), and then rechecks if the SAR limits are violated to prevent undesired scan terminations resulting from patient movements, such as coughing or muscle twitching. This waiting time must be tailored to the existing boundary conditions for safety,

Comprehensive RF Safety Concept

which vary with the coil, coil loading, RF chain properties, and field strength. In case of a SAR violation, the supervision system initiates the termination of the scan. In addition to the SAR supervision, RF waveform supervision verifies that the absolute value of the complex distance between demand and measured samples does not exceed the error margin of 10% of the value occurring at maximum RF power. The selected value must be large enough to account for breathing and normal patient movement. PUC based respiration monitoring experiments (41,42) showed that typical motion induced amplitude deviations are 624%. An error margin of 6% was added to the maximum reported value of 4% to account for normal patient movement. The error margin might need further fine tuning depending on the coil used. A software-implemented watchdog mechanism secures the proper operation of the safety-critical supervision software and aborts the scan if a fault is detected. Experimental Verification Active Decoupling To demonstrate the effects of coil coupling and their compensation by the preparation phase, a 1-liter oil bottle was placed in the isocenter of the body coil. In a first experiment, the preparation phase was turned off, and a low power RF triangle pulse was transmitted sequentially on each Tx element. Asys was then composed of the simultaneously sampled signals induced in each individual PUC. In a second experiment, the preparation phase was turned on, and the pulses were transmitted with precompensation A1 sys . The second experiment was repeated using sinusoidal-shaped RF pulses. Load Compensation To verify the compensation of load changes, a low power RF triangle pulse (not precompensated) was transmitted sequentially on each Tx element while the preparation phase was turned off. From each PUC, corresponding to the element sending the triangle pulse, the maximum sample was used, resulting in a single value per Tx element. A volunteer was then placed in supine position with the arms touching the bore covers. The set of RF pulses used for the oil bottle was repeated and the PUC signals were acquired. The experiment was run a second time with the volunteer’s arms next to the body and a third time with the volunteer’s arms resting on the torso with the hands not touching it. Finally, the load compensation preparation phase was turned on, as required by this safety concept, and the experiments were repeated for the three different volunteer’s arm positions. All in vivo experiments were performed on healthy volunteers according to the ethical rules of the institution. SAR Prediction and Supervision To test the SAR supervision based on transmitted RF pulses via PUC monitoring, phantom experiments were performed, and RF shimming and local excitation were used as examples. The Q-matrices previously generated for a sphere phantom (diameter ¼ 20 cm, s ¼ 0:37 Sm1 ,

593

er ¼ 80) (32) were selected for the SAR prediction and supervision. The phantom was placed in the isocenter of the MR system. Bþ 1 shimming was performed based on maps measured with dual refocusing echo acquisition mode (39) (nominal flip angle a ¼ 50 , imaging flip angle b ¼ 3 , repetition/echo times TR=TEFID =TESTE ¼ 3:4=2:1=1:2 ms, scan resolution 3:5  3:5  10 mm3 ). For a fast gradient echo scan, TR and a were varied, for example, from 5.3 to 40 ms and 20 to 150 , respectively, resulting in a predicted global SAR of SARWB < 4 Wkg1 and a local SAR of SARL < 15 Wkg1 . In a second set of experiments, spatially selective RF pulses were used to compare the predicted and supervised SAR. A disc-shaped 2D region of 5-cm diameter was excited in the middle of the phantom. A spiral kspace trajectory with a numerical field-of-excitation of 32 32 pixels was used. The transmit SENSE pulses were calculated using a pulse calculation algorithm based on Lagrange multipliers (43) (TR ranging from 7 to 170 ms, and a ranging from 20 to 100 , resulting in a predicted global SAR of SARWB < 4 Wkg1 and local SAR of SARL < 8 Wkg1 ). For all scans, the SAR was calculated per RF pulse, and then averaged for all RF pulses. The standard deviation was calculated for the SAR of all RF pulses, and the difference was determined between the predicted SAR and the supervision SAR calculated from the PUC sample monitoring.

System Fault Situations To validate the RF safety concept, the system matrix and SAR were analyzed for different, intentionally introduced, fault scenarios (i)–(vii). For these tests, a healthy volunteer was placed inside the MR scanner in supine position, with the abdomen in the isocenter of the scanner. A safe, low-SAR fast gradient echo scan with TR=TE ¼ 5:7=2:5 ms and a ¼ 10 resulted in a predicted local torso SAR of SARLT < 2 Wkg1 , in order for the scans to be safe for the volunteer. For all test cases, the body coil was driven with quadrature excitation and was actively decoupled at the beginning of the scan. The normalized root mean square error (NRMSE) was calculated for the element of the magnitudes of Asys acquired at the beginning and end of the test. Furthermore, the SAR calculated by the SAR monitor was logged for each RF pulse. SAR deviations resulting from the following list of intentionally induced fault situations were investigated. The maximum-intensity projections of the local SAR for the abdominal region were calculated for test case (vii) before and after introducing the fault. i. The volunteer moved his arms from resting on the body to next to his body. ii. The volunteer moved his body closer to the right side of the body coil. iii. The patient table was moved 10 cm in feet-head direction. iv. A malfunction of a surface receive coil (1 ¼ 10 cm) was imitated by switching the coil to the

594

Graesslin et al.

tuned state during the scan. The coil was covered by foam and placed 6 cm above the volunteer for his safety. v. To mimic a fault situation of the RF chain, the phase of one Tx channel was changed by 377 when removing a 3-m coaxial cable. vi. To mimic a fault situation of the RF chain, the amplitude and phase of Tx channel 1 was increased by 25% and 10 , respectively, by removing a 3-dB attenuator at the RF amplifier input. vii. To mimic a broken cable, broken Tx/Rx switch, or similar, the amplitude of Tx channel 1 was set to zero by unplugging the RF demand signal at the RF amplifier input. RESULTS Active Decoupling To determine the effectiveness of the preparation phase to compensate for coil coupling effects, phantom experiments were carried out using an oil bottle. Figure 3a shows the received RF transmit signals, which are time interleaved, via the PUCs. For elements that transmit an RF signal, the received normalized RF amplitudes are at unity, while the PUC signals of the remaining channels also show signal due to the residual mutual inductive coupling among the coil elements. The induced opposing currents in the other coil elements result from the magnetic field generated by each Tx element and, according to Lenz’s law, is also referred to as coupling. When applying the A1 sys to the RF demand, the received PUC signals show nulled or compensated currents, thus leading to a decoupled system (Fig. 3b). This active decoupling works independently from the RF pulse shape as shown for the example of sinusoidalshaped RF pulses in Fig. 3c. Load Compensation We next determined the change in coil currents due to different loading conditions, and tested how well the preparation phases could compensate for these. Figure 4 shows the results of the loading experiments. While all values are close to unity for the oil bottle used to calibrate the system, large deviations from unity occur when the volunteer is loading the coil. This is particularly evident for elements 2 and 7. The closer the arms are to these elements, the larger the loading and the smaller the resulting current in the elements. The normalized current values without compensation are in the range of 0.35 to 0.86. After turning the preparation phase on, the loading is compensated for (Fig. 4b). The normalized current values with compensation are between 0.96 and 1.03.

FIG. 3. PUC-based active decoupling of RF transmit (Tx) signals. a: During RF transmission of eight individual, triangular test RF pulses on each Tx channel, the PUC signals of all 8 RF Tx channels are sampled simultaneously by the PUC monitoring system. b: Based on the measurements in (a), compensation parameters are determined to correct the RF Tx signals for the desired target distribution of the currents in the transmit elements, which is obtained if the RF transmit coil elements are decoupled. The received PUC signals show almost no remaining residual coupling. c: This active decoupling is RF pulse shape independent as shown for the example of sinusoidal-shaped RF pulses.

SAR Prediction and Supervision A wide range of global and local SAR values was tested for RF shimming and local excitation with varied repetition times and flip angles in a phantom study. The global and local predicted SAR correlate very well with the supervision SAR for various RF signal amplitudes

and durations (Fig. 5). This result demonstrates the functioning of the entire RF Tx chain, the detection of the PUC signals, as well as their processing by the supervision system. However, this test does not guarantee a match between simulation (simulated sphere) and reality

Comprehensive RF Safety Concept

595

FIG. 4. Effects of coil loading and preparation phases. a: The normalized element currents of the 8 RF Tx elements of the body coil in different loading conditions (different positions of volunteer’s arms) and for the 1-liter oil bottle phantom (black circle) used for system calibration. Coil elements 2 and 7, the closest to the volunteer’s arms, are most strongly loaded and show the largest change in amplitude. b: The polar diagram represents the data differently in (a) for the preparation phase turned off and shows the volunteer’s arms in three different positions and the physical Tx element locations (#1–#8). The element currents, when the preparation phases were turned off (solid lines) and on (broken lines), with the circles representing the values 0.0–1.0. In case the preparation phases are turned on to account for loading effects, all values are close to unity (outer circle).

(actual measurement of sphere phantom) and, in turn, the actual SAR. Care must be taken during coil simulation and design to ensure that nulling the signals induced in the PUCs is indeed sufficient to reliably match the simulated fields with reality. Therefore, the SAR prediction was previously verified for this phantom using calorimetric experiments (32). The averaged supervision SAR of all individual RF pulses only differs from the supervision SAR of all PUC samples within the range of numerical rounding errors. The largest standard deviation among all RF pulses of a scan is 7:0  103 . The maximum difference between predicted and calculated SAR values is 65% for the global SAR and 64% for the local SAR. The errors result from calibration and hardware inaccuracies.

fit the actual SAR of the scan. While the supervision SAR is reduced by about 15% in case (i), the local extremity SAR generally increases when extremities are moved closer to the coil elements. The same applies to case (iv), where a resonant structure close to the body leads to additional local E-fields. This may cause local patient heating, in which case the scan should be aborted. Because it is not easy to reliably distinguish between coil “external” and coil “internal” variations (Fig. 6a) a

System Fault Situations To test how the system behaves in various fault situations, we measured the system matrices before and after the fault situation. We calculated the NRMSE of the elements of the system matrices (Fig. 6a) and the deviations of the global SAR, local torso SAR, and local extremities’ SAR induced by the fault situations (Fig. 6b). Cases leading to load changes inside the coil [cases (i)–(iv)] and cases mimicking faults in the RF chain driving the coil [cases (v)–(vii)] were tested. While the NRMSE is a good predictor for load changes, it is a poor predictor for the SAR deviation because there is no correlation between NRMSE and SAR. For test cases (iv) and (v) the supervision SAR is increased, while for test case (iv) the NRMSE in percent is 4.6 but it is 2.2 for test case (v). In situations in which the load changes, the predicted SAR is incorrect because the electromagneticfield simulations used by the SAR supervision no longer

FIG. 5. Validation of predicted and calculated SAR from PUC samples. For various RF shimming and local excitation experiments using a water-filled sphere, the global SARWB and the 10-g averaged local SARL are shown. The predicted SARWB (open circles) and SARL (gray circles) based on the demand RF waveforms are compared to those calculated from the sampled RF signals acquired via the PUCs.

596

FIG. 6. Normalized system matrices, SAR deviation, and SAR plots. a: Plots of the elements of the normalized system matrices shown for the different cases mimicking different system fault situations (i)–(vii) (lower left corner). For each case, the NRMSE in percent is shown, which represents the deviation of the magnitudes of elements before and after introducing the fault (upper right corner). Only the first system matrix, as a reference, represents the case in which no fault was introduced during the scan, so that the deviations between the system matrices are close to zero. For visualization purposes, the elements of all system matrices are scaled such that the largest element is one. b: SAR deviation (%) for the global SAR (red), the local torso SAR (green), and the local extremities’ SAR (blue) for the different cases. c: Comparison of maximum-intensity projections of the local SAR of the abdominal region for the undisturbed scan (left) and disturbed scan (right) for test case (vii).

scan is terminated if the NRMSE in percent exceeds a value of 1.0, which is an error margin of a factor of 10 compared to the undisturbed case, thus allowing for a certain amount of normal subject movement. The SAR maximum-intensity projections in Fig. 6c show a local extremities’ SAR decrease from 1:7 to 1:4 Wkg1 , which is visible in the arms’ regions. DISCUSSION The key issues for the safety of pTx imaging systems are to correctly predict the actual SAR prior to the scan, to

Graesslin et al.

supervise that the SAR limits are not exceeded during the scan, and to abort the scan in case of hazardous situations for the patient. These issues have been addressed by the realization of the proposed concept. Fast and accurate SAR prediction is important for the compulsory safety check before each scan. Using Qmatrices based on preprocessed numerical simulation data, SAR computation times can be significantly reduced. To ensure accuracy, which means that the predicted SAR matches the actual SAR applied to the patient during the scan, hardware and software measures are used. The hardware was designed with circulators to improve the correspondence between SAR simulation and experiment. Furthermore, the SAR related system parameters are adapted during the preparation phases to establish the essential correspondence between electromagnetic field simulations and the actual scan situation with the patient in the magnet. The active decoupling is effective because the residual currents were effectively nulled and load compensation is successful, as the preparation phase was able to lead to unity currents for all tested loading conditions. This work demonstrates that RF and SAR supervision can effectively verify that the predicted SAR is not exceeded during the scan. Correctly establishing the relationship between the (forward) power and the resulting RF field is complicated by the effects of the static RF chain (cable losses, coil geometry, and matching circuitry) and multiple patient-dependent factors (e.g., power reflection, coil resonance quality, and interelement coupling). One approach to comprehensively model these effects is to carry out port-based postprocessing after the electromagnetic simulation (44). During this processing, the S-parameter matrix of the superposed simulated electromagnetic fields should match the matrix obtained by measuring the loaded coil array. However, the commonly used indirect approach to RF power-based calibrations relies on directional coupler measurements of the RF power, either at the RF amplifier or at the physical coil, and subsequent scaling of the simulated power according to that value at the location of the measurements (11,18). In contrast, the element-current-based calibration approach presented here allows the Tx phases of all coil elements to be directly obtained, and these values are essential for accurate SAR prediction. This approach, therefore, has the advantage of being more direct than power-based calibrations, avoiding load-dependent uncertainties. As the Q-matrices are normalized to the Bþ 1 -field and the RF transmission is actively decoupled, the SAR can be accurately predicted prior to the scan because all required information is available. Alternatively, coil coupling could be taken into account during the SAR prediction by applying the system matrix Asys measured during the preparation phase to the input of the SAR prediction, which, in turn, would not require modification of the RF Tx signals. However, as Asys can only be determined, the exact SAR for the scan could only be determined afterward. RF supervision makes it possible to abort scanning in case of erroneous RF transmission, for example, due to RF chain failures or the presence of unsafe devices (26). This concept has been adopted

Comprehensive RF Safety Concept

recently for 7 T using PUCs (17) and directional couplers (45). The problem of SAR overestimation based on RF supervision alone is reduced significantly by directly supervising the global and local SAR during the scan. Consequently, a high permissible RF duty cycle is achievable. This supervision is done by sampling the RF signals via PUCs (or directional couplers) and predicting the SAR based on the transmitted RF, using the same model as for predicting the SAR before the scan. Thus, it is possible to significantly reduce the overestimation, which is important in clinical practice. Unfortunately, unsafe patient situations may be undetected because different RF inputs can lead to the same SAR. Such situations can be identified when supervising the RF signals as well. The supervision algorithms must determine if a potentially unsafe situation for the patient occurs, which is done by calculating the NRMSE of the diagonal elements between the reacquired system matrix with the matrix measured during the preparation phase. For all fault situations, the NRMSE was at least a factor of 5 larger than in the undisturbed case. Another important aspect of RF-related patient safety is the use of recently introduced current sources and RF feedback loops (46–50) with the aim of compensating for RF chain deficiencies or the use of (low-cost) amplifier modules for Tx arrays. However, care must be taken that, when resonant objects are present (i.e., failing Rx coils), deviations in amplitude and phase are not compensated for by the feedback system, as they could lead to a potentially very dangerous situation for the patient. Reliable and low-latency processing are essential boundary conditions that a supervision system should fulfill. This can be achieved for instance by integrating the RF and SAR supervision system with the reconstructor because it receives the PUC samples from the DAS with minimal delay and has an appropriate processing capability. To minimize processing delays and decouple image reconstruction from the SAR calculations, its processing is carried out on a graphics card. Graphics processing units are ideally suited for the required tasks because most operations have no data dependency and can thus be optimally parallelized on many cores. While the safety concept presented in this article has been demonstrated for a 3 T pTx system, the present example of a body coil is applicable to any field strength and Tx array geometry, as recently shown by (18), where the concept of SAR supervision was adopted for Tx arrays at 7 T. A very important aspect in terms of SAR prediction accuracy, although beyond the scope of this article, is the selection of an appropriate body model to match the scan situation. A global SAR verification for the phantom used here was previously carried out using calorimetric experiments to validate the SAR prediction concept (32). Furthermore, the SAR prediction concept was validated in vivo using a subject specific model created from segmented water/fat separated MR data by comparing simulated and measured Bþ 1 fields. Although only three body models of volunteers have been used in this article for illustrative purposes, the proposed concept supports any number of body models.

597

While this dual supervision concept reduces the SAR safety margin, a major remaining source of potential error is the discrepancy between the modeled and the real patient. More future effort is necessary to bridge this gap. CONCLUSIONS Safe operation in pTx was successfully demonstrated on a pTx MR research system. The SAR is predicted before the scan, and the global and local SAR are monitored in real time during the scan, while the RF fields are supervised. This dual supervision concept allows minimal SAR overestimation while still detecting unsafe RF situations for the patient, which would be impossible to detect by SAR supervision alone. This new safety system could also find applications at field strengths higher than 3 T. Although further research efforts are necessary in this important field of patient-related RF safety, including in the area of patient modeling and selection, this concept can contribute to the safe operation of pTx systems. ACKNOWLEDGMENTS The paper is the result of a long research activity; therefore, the authors would like to cordially thank many people: Dennis Glaesel, Sven Biederer, and Henk Dingemans for supporting concept implementation; Hanno Homann for subject-specific model creation; Hanno Homann, Christoph Leussler, and Zhiyong Zhai for modeling the body coil; Bjoern Annighoefer for supporting SAR simulations; Oliver Lips, Daniel Wirtz, Johan Overweg, Christian Findeklee, Joachim Schmidt, Holger Eggers, and Johan van den Brink for technical discussions and support. REFERENCES 1. R€ oschmann P. Radiofrequency penetration and absorption in the human body: limitations to high-field whole-body nuclear magnetic resonance imaging. Med Phys 1987;14:922–931. 2. Ibrahim TS, Lee R, Baertlein BA, Kangarlu A, Robitaille PL. Application of finite difference time domain method for the design of birdcage RF head coils using multi-port excitations. J Magn Reson Imaging 2000;18:733–742. 3. Hoult DI, Phil D. Sensitivity and power deposition in a high-field imaging experiment. J Magn Reson Imaging 2000;12:46–67. 4. Seifert F, Rinneberg H. Adaptive coil control: SNR optimization of a TR volume coil for single voxel MRI at 3T. In Proceedings of the 10th Annual Meeting of ISMRM, Honolulu, Hawaii, USA, 2002. p. 162. 5. Katscher U, B€ ornert P, Leussler C, van den Brink JS. Transmit SENSE. Magn Reson Med 2003;49:144–150. 6. Zhu Y. Parallel excitation with an array of transmit coils. Magn Reson Med 2004;51:775–784. 7. Grissom WA, Yip CY, Zhang Z, Stenger V, Fessler JA, Noll DC. Spatial domain method for the desgin of RF pulses in multicoil parallel excitation. Magn Reson Med 2006;56:620–629. 8. Ullmann P, Junge S, Wick M, Seifert F, Ruhm W, Hennig J. Experimental analysis of parallel excitation using dedicated coil setups and simultaneous RF transmission on multiple channels. Magn Reson Med 2005;54:994–1001. 9. Graesslin I, Falaggis K, Vernickel P, R€ oschmann P, Leussler C, Zhai Z, Morich M, Katscher U. Safety considerations concerning SAR during RF amplifier malfunctions in parallel transmission. In Proceedings of the 14th Annual Meeting of ISMRM, Seattle, Washington, USA, 2006. p. 2041. 10. Graesslin I, Falaggis K, Biederer S, et al. SAR Simulations and experiments for parallel transmission. In Proceedings of the Joint Annual Meeting of ISMRM-ESMRMB, Berlin, Germany, 2007. p. 1090.

598 11. Seifert F, Wuebbeler G, Junge S, Ittermann B, Rinneberg H. Patient safety concept for multichannel transmit coils. J Magn Reson Imaging 2007;26:1315–1321. 12. Graesslin I, Glaesel D, Biederer S, et al. Comprehensive RF safety concept for parallel transmission systems. In Proceedings of the 16th Annual Meeting of the ISMRM, Toronto, Canada, 2008. p. 74. 13. Collins CM. Numerical field calculations considering the human subject for engineering and safety assurance in MRI. NMR Biomed 2009; 22:919–926. 14. Brote I, Orzada S, Kraff O, et al. A multi-channel SAR prediction and online monitoring system at 7T. In Proceedings of the 17th Annual Meeting of the ISMRM, Hawaii, USA, 2009. p. 4788. 15. Kudielka G, Vogel M, Loew W. Realisation of a flexible power measurement application for parallel transmit. In Proceedings of the 17th Annual Meeting of the ISMRM, Honolulu, Hawaii, USA, 2009. p. 4789. 16. Graesslin I, Biederer S, Annighoefer B, Homann H, Stahl H, Vernickel P, Katscher U, Glaesel D, Harvey P. Real-time global and local SAR monitoring for parallel transmission systems. In Proceedings of the 17th Annual Meeting of the ISMRM, Honolulu, Hawaii, USA, 2009. p. 302. 17. Gagoski BA, Gumbrecht R, Hamm M, et al. Real time RF monitoring in a 7T parallel transmit system. In Proceedings of the 18th Annual Meeting of the ISMRM, Stockholm, Sweden, 2010. p. 781. 18. Gumbrecht R, Fontius U, Adolf H, Benner T, Schmitt F, Adalsteinsson E, Wald LL, Fautz H-P. Online local SAR supervision for transmit arrays at 7T. In Proceedings of the 21st Annual Meeting of the ISMRM, Salt Lake City, Utah, USA, 2013. p. 4420. 19. Fontius U, Nistler J, Alagappan VA, Adalsteinsson E, Setsompop K, Zelinski A, Hebrank F, Schmitt F, Wald LL. Multichannel excitation. In ISMRM Workshop on Advances in High Field MR, Pacific Grove, California, USA, 2007. p. 7. 20. Cloos MA, Luong M, Ferrand G, Amadon A, Le Bihan D, Boulant N. Specific absorption rate monitor for in-vivo parallel transmission at 7 Tesla. In Proceedings of the 18th Annual Meeting of the ISMRM, Stockholm, Sweden, 2010. p. 3871. 21. Nistler J, Fontius U, Wald LL, Adalsteinsson E, Setsompop K, Alagappan VA, Hebrank F, Schmitt F, Renz W. Parallel transmission with an 8 channel whole body system at 3 T. In Proceedings of the Joint Annual Meeting of ISMRM-ESMRMB, Berlin, Germany, 2007. p. 1027. 22. Collins C, Wang Z, Smith M. A conservative method for ensuring safety within transmit arrays. In Proceedings of the Joint Annual Meeting of ISMRM-ESMRMB, Berlin, Germany, 2007. p. 1092. 23. Zhu Y, Alon L, Deniz CM, Brown R, Sodickson DK. System and SAR characterization in parallel RF transmission. Magn Reson Med 2012; 67:1367–1378. 24. Stang P, Pauly J, Scott G. A versatile in-line sensor for power monitoring and calibration of transmit arrays. In Proceedings of the 17th Annual Meeting of the ISMRM, Honolulu, Hawaii, USA, 2009. p. 3024. 25. Graesslin I, Vernickel P, Schmidt J, et al. Whole body 3T MRI system with eight parallel RF transmission channels. In Proceedings of the 14th Annual Meeting of ISMRM, Seattle, Washington, USA, 2006. p. 129. 26. Graesslin I, Krueger S, Vernickel P, Achtzehn J, Nehrke K, Weiss S. Detection of RF unsafe devices using a parallel transmission MR system. Magn Reson Med 2013;70:1440–1449. 27. International Electrotechnical Commission. Medical Equipment - IEC 60601-2-33: Particular Requirements for the Safety of Magnetic Resonance Equipment. 3rd ed. Geneva, Switzerland: International Electrotechnical Commission; 2010. 28. US Food and Drug Administration. Guidance for Industry and FDA Staff: Criteria for Significant Risk Investigations of Magnetic Resonance Diagnostic Devices. Silver Spring, MD: US Food and Drug Administration; 2003. 29. Kunz KS, Luebbers RJ. The finite difference time domain method for electromagnetics. Boca Raton, Florida: CRS Press; 1993. 30. Vernickel P, R€ oschmann P, Findeklee C, Luedeke KM, Leussler C, Overweg J, Katscher U, Graesslin I, Schuenemann K. Eight-channel transmit/receive body MRI coil at 3T. Magn Reson Med 2007;58:381–389. 31. Homann H, B€ ornert P, Eggers H, Nehrke K, D€ ossel O, Graesslin I. Toward individualized SAR models and in vivo validation. Magn Reson Med 2011;66:1767–1776.

Graesslin et al. 32. Graesslin I, Homann H, Biederer S, B€ ornert P, Nehrke K, Vernickel P, Mens G, Harvey P, Katscher U. A specific absorption rate prediction concept for parallel transmission MR. Magn Reson Med 2012;68: 1664–1674. 33. Gebhardt M, Diehl D, Adalsteinsson E, Wald LL, Eichfelder G. Evaluation of maximum local SAR for parallel transmission (pTx) pulses based on pre-calculated field data using a selected subset of “Virtual Observation Points”. In Proceedings of the 18th Annual Meeting of the ISMRM, Stockholm, Sweden, 2010. p. 1441. 34. Eichfelder G, Gebhardt M. Local specific absorption rate control for parallel transmission by virtual observation points. Magn Reson Med 2011;66:1468–1476. 35. Homann H, Graesslin I, Eggers H, Nehrke K, Vernickel P, Katscher U, D€ ossel O, B€ ornert P. Local SAR management by RF shimming: a simulation study with multiple human body models. MAGMA 2012;25: 193–204. 36. Graesslin I. Parallel transmission: a comprehensive RF safety concept. In Progress in Electromagnetics Research Symposium, Cambridge, USA, 2008. pp. 698–700. 37. Vernickel P, Findeklee C, Eichmann J, Graesslin I. Active digital decoupling for multi-channel transmit MRI Systems. In Proceedings of the Joint Annual Meeting of ISMRM-ESMRMB, Berlin, Germany, 2007. p. 170. 38. Scott G, Stang P, Overall W, Kerr A, Pauly J. Signal vector decoupling for transmit arrays. In Proceedings of the 16th Annual Meeting of the ISMRM, Toronto, Canada, 2007. p. 168. 39. Nehrke K, B€ ornert P. DREAM-a novel approach for robust, ultrafast, multislice B1 mapping. Magn Reson Med 2012;68:1517–1526. 40. Graesslin I, Biederer S, Falaggis K, et al. Real-time SAR monitoring to ensure patient safety for parallel transmission systems. In Proceedings of the Joint Annual Meeting of ISMRM-ESMRMB, Berlin, Germany, 2007. p. 1086. 41. Graesslin I, Glaesel D, B€ ornert P, Stahl H, Koken P, Nehrke K, Dingemans H, Mens G, G€ otze J, Harvey P. An alternative concept of non-sequence-interfering patient respiration monitoring. In Proceedings of the 16th Annual Meeting of the ISMRM, Toronto, Canada, 2008. p. 202. 42. Graesslin I, Stahl H, Nehrke K, Harvey P, Smink J, Mens G, Senn A, B€ ornert P. An alternative concept for non-sequence interfering, contact-free respiration monitoring. In Proceedings of the 17th Annual Meeting of the ISMRM, Honolulu, Hawaii, USA, 2009. p. 753. 43. Graesslin I, Schweser F, Annighoefer B, Biederer S, Katscher U, Nehrke K, Stahl H, Dingemans H, Mens G, B€ ornert P. A minimum SAR RF pulse design approach for parallel Tx with local hot spot suppression and exact fidelity constraint. In Proceedings of the 16th Annual Meeting of the ISMRM, Toronto, Canada, 2008. p. 621. 44. Kozlov M, Turner R. Fast MRI coil analysis based on 3-D electromagnetic and RF circuit co-simulation. J Magn Reson 2009;200:147–152. 45. Brote I, Solbach K, Orzada S, Kraff O, Maderwald S, Ladd ME, Bitz AK. RF monitoring of the complex waveforms of an 8-channel multitransmit system at 7T utilizing directional couplers and I/Q demodulators. In Proceedings of the 19th Annual Meeting of the ISMRM, Montreal, Canada, 2011. p. 3849. 46. Hoult DI, Kolansky G, Kripiakevich D, King SB. The NMR multitransmit phased array: a Cartesian feedback approach. J Magn Reson 2004;171:64–70. 47. Hoult D, Foreman D, Kolansky G, Kripiakevich D. Overcoming highfield RF problems with non-magnetic Cartesian feedback transceivers. MAGMA 2008;21:15–29. 48. Stang P, Kerr AB, Pauly J, Scott G. An Extensible transmit array system using vector modulation and measurement. In Proceedings of the 16th Annual Meeting of the ISMRM, Toronto, Canada, 2008. p. 145. 49. Stang P, Zanchi M, Kerr A, Pauly J, Scott G. Active coil decoupling by impedance synthesis using frequency-offset cartesian feedback. In Proceedings of the 19th Annual Meeting of the ISMRM, Montreal, Canada, 2011. p. 332. 50. Zanchi MG, Stang P, Kerr A, Pauly JM, Scott GC. Frequency-offset Cartesian feedback for MRI power amplifier linearization. IEEE Trans Med Imaging 2011;30:512–522.