The restricted SAR protocol: A method to assess MRI coil prototypes in an unconditionally safe manner

Testing an RF coil prototype on subjects involves laborious verifications to ensure its safety. In particular, it requires preliminary electromagnetic simulations and their validations on phantoms to accurately predict the specific absorption rate (SAR). For coil design validation with a simpler safety procedure, the restricted SAR (rS) mode is proposed, enabling representative first experiments in vivo. The goal of the developed approach is to accelerate the transition of a custom coil system from prototype to clinical use.


INTRODUCTION
The MRI evolution along with the interest in strong magnetic fields demands new technological solutions for instrumentation and image acquisition.][3] The transmit coil design considers the resulting excitation profile, transmission efficiency, and deposed RF energy.One aspect to face is the inhomogeneity of the transmitted RF magnetic field B 1 + .It arises from the variation in the dielectric properties of tissues and the interference of RF waves. 4The latter becomes especially pronounced for ultrahigh-field MRI, as the Larmor frequency increases with the field strength, and the electromagnetic (EM) wavelength shortens and reaches the dimension of the imaged organ.The B 1 + inhomogeneity leads to changes in the signal intensity and contrast, complicating image evaluation and diagnosis.6][7] Another state-of-art solution is parallel transmission (pTx), [8][9][10] providing control over the B 1 + of several individual transmit elements to achieve more uniform excitation under given power and SAR constraints.
For receive coil engineering, the key factors consist of the SNR gain as well as acceleration capabilities that can be attained by using receive coil arrays.Multiple receiving elements provide a higher SNR due to a better filling factor, especially in the case of flexible arrays. 11,12In addition, the parallel-imaging technique often used with receive coil arrays reduces MRI exam time. 13ustom antenna designs can be of interest to research organizations seeking to test new concepts beyond the scope of commercial offers.Before being introduced into clinical practice, the coil prototype undergoes multiple tests on subjects to assess its applicability for in vivo studies.These exams verify that intersubject anatomical variations (age, geometry, and anatomy) do not affect coil performance and confirm the validity of computer simulations for the prototype.In vivo examinations with a specific antenna design must be approved as safe.For a transmit coil, the safety concerns involve tissue heating due to the RF energy deposition in the patient's body.The core temperature depends on various biophysical parameters, and its local increase occurring in MR exams cannot be easily assessed, despite recent efforts to measure it in vivo. 14Therefore, safety rules consider the heating measure directly related to the EM nature of the energy dissipation-the specific absorption rate (SAR). 4,15,16It corresponds to the average RF power absorbed by a unit mass of an object and directly depends on the electrical field.
The subject's security requires monitoring not only the global SAR value for the entire body, but also the local SAR for each partial (small) tissue region.While the former reflects an increase in overall core temperature and poses a threat to patients with thermoregulatory disorders, the latter indicates local heating and hotspots that could cause tissue damage.Safety committees, such as the International Electrotechnical Commission (IEC), recommend global and local SAR restrictions to be respected during all in vivo experiments.Guidelines generally consider local SAR for every 10 g of tissue as a reasonable tradeoff for most volume and surface coil; however, 1-g SAR averaging is proposed in some cases. 17n addition to the nonuniform B 1 + and excitation profile at ultrahigh field mentioned previously, the wavelength reduction exacerbates the inhomogeneity of the electric fields' amplitude in the tissue and make the SAR highly irregular. 18Although the global SAR can still be estimated through the total transmitted and reflected power, the local SAR assessment represents a challenging task.][21] Its spatial distribution intricately depends on body size and composition as well as coil configuration and cannot be directly measured during in vivo studies.3][24] The gold standard, however, remains the SAR prediction based on computer simulations of the EM fields using numerical methods such as finite-difference time domain, 25,26 finite integration, 27 or finite element. 28s a result, ensuring safe SAR levels when testing a new coil prototype in humans usually requires preliminary checks and simulations of the transmitted EM field and SAR.Their main purpose is to find the relation between the peak local SAR and the input power to control the SAR level during experiments by restricting the MR sequence power.
The coil safety evaluation usually takes a long time and depends on the requirements of the research organization and ethics committees. 29Some works 30,31 propose a general procedure and distinguish the following steps.To find the SAR distribution, one first needs to introduce the numerical model, including all the necessary coil elements, such as conductors, lumped elements, and feed structures.The accuracy of the predicted EM fields is evaluated on phantoms with simplified geometry and known dielectric and thermal properties.For example, one can verify the agreement between the B 1 + fields calculated for the phantom model and those obtained using MR experiments.In addition, using near-field probes makes it possible to directly measure the amplitudes of RF fields at specific phantom points and compare them with the simulated values. 31Thermometric studies can also evaluate the model used.In this case, the temperature distribution is calculated using the heat diffusion equation starting from a SAR distribution as power source and checked against experimental values estimated using the proton-resonance frequency shift method or fiber optic temperature sensors. 32,33fter the coil numerical representation is verified, simulations are performed for various human body models to determine the most critical local SAR values.Finally, SAR safety margins consider possible errors in modeling and determining input power as well as the intersubjective variability. 34,35The results are then documented and can be used to request approval to use the prototype coil in vivo from the relevant organizations in accordance with the country regulations.
However, in vivo experiments may indicate necessary changes in the antenna design.The improved geometry then defines a different EM field distribution and thus requires repeating SAR simulation and verification.These numerous cycles instigated by the coil modifications consume a lot of time and resources and slow down the implementation of a custom prototype.
To overcome the described procedure complexity (i.e., to avoid the repeated SAR calculations and approvals during the development process), we propose the restricted SAR (rS) exam mode.It limits global and local SAR regardless of the EM field distribution by significantly reducing RF input power at the coil plug.This unconditional approach allows the safe testing of different coil configurations without any time-consuming prior simulations.The presented work develops the rS protocol based on modified MR sequences for the qualitative and quantitative coil assessment.The ability of the rS protocol to quantitatively characterize the antenna performance within the time corresponding to the standard MR exam is tested on commercial and pre-industrial RF coils.An appropriate choice of parameters and a complete pipeline for coil evaluation with the rS protocol are described.

rS protocol
The rS approach 36 considers the worst-case local SAR based on the energy conservation law only and defines the RF power constraints to keep this value safe following IEC guidelines.The highest possible SAR local corresponds to the hypothetical and unattainable in a practice situation, in which a single 10-g mass of tissue absorbs all the transmitted RF power.To meet the safety limits for these extreme conditions, the total power P supplied to the coil must satisfy P∕10g ≤ SAR local limit . (1) This work uses safety regulations prescribed by the IEC for head examinations in a normal mode, 37 in which the global and local SAR limits are 3.2 W/kg and 10 W/kg for an averaging period of 6 min (long-term threshold) and twice as high for 10 s (short-term threshold).In this case, considering the local SAR limitation, the rS mode imposes long-term and short-term input RF power constraints of 0.1 W and 0.2 W according to Eq. (1).Such conditions automatically limit the global SAR for the head and result in values two orders of magnitude below the IEC threshold (Table 1).
To run the measurements under the restricted SAR mode, we adapted several MRI pulse sequences needed for the coil assessment.They were modified to calculate the pulse train power for 6-min and 10-s periods and determine the available range of parameters (e.g., flip angle [FA], TR, number of slices) based on the rS mode conditions (Table 1).For sequence durations (TA) under 6 min, we calculated the long-term power limit for TA, which is a more stringent condition, allowing the sequences to be played sequentially more safely.

Protocol implementation
Evaluating a coil first requires a quantitative description of its transmitting and receiving performance.The former is characterized by the transmission profile (B 1 + map) and the transmission efficiency, which is determined, for Power constraints and specific absorption rate (SAR) values for the restricted SAR (rS) mode compared with SAR limits for normal mode for head studies (the average head weight of an adult m head = 5 kg was considered).

Max power, rS mode 6 min (10 s)
SAR, rS mode 6 min (10 s) SAR limit, normal mode 6 min (10 s) instance, by the reference voltage (V ref ) required to generate a rectangular inversion π-pulse of 1 ms.The latter incorporates the coil reception profile or B 1 − map and average SNR.The fact that the patient load variability cannot be reproduced properly with phantoms implies in vivo measurements of these coil characteristics.The prototype evaluation also needs some anatomical measurements to assess the resulting image quality, such as with T 2 *-weighted images.The selected parameters for each sequence should maintain a sufficient SNR to provide meaningful quantitative data without significantly increasing acquisition time to comply with clinical protocols.
Moreover, the rS mode requires some automatic adjustments performed by the scanner to be replaced with adapted custom procedures.Determining the transmitter reference voltage V ref and measuring the B 0 field inhomogeneity for the shimming procedure can violate the imposed power constraints.Therefore, it is necessary to include in the rS protocol the appropriate sequences to customize B 0 mapping and perform manual V ref adjustment.At the same time, the default frequency adjustment and forward power measurement with directional couplers (DICO) are characterized by low power supply (less than 1 W per 10 s) and do not interfere with rS conditions, at least on the 7T Magnetom Siemens scanner used.
To provide a coil testing "kit," we implemented the rS protocol on two customized MR sequences compatible with the restricted SAR mode: gradient-recalled echo (GRE) and presaturated Turbo-FLASH (hereinafter referred to as satTFL [custom name XFL]).The necessary constraints on the pulse train power were implemented in the sequence source codes using IDEA, the sequence programming environment provided by Siemens Healthineers.The rS-adapted GRE provides localizer, B 0 maps for shimming, T 2 *-weighted images, and coil reception profiles.The satTFL sequence yields maps of the transmitted magnetic field (B 1 + map) in the form of the FA distribution in the subject and the corresponding V ref map.

2.3
Coil assessment methodology

Presaturated Turbo-FLASH sequence for characterizing the coil transmission profile
The satTFL sequence, 38 based on a turbo-FLASH readout series with a FA β, yields two images: (i) saturation (preparation) image where the magnetization is rotated by an angle α before readout; and (ii) reference image without magnetization preparation (Figure 1).The ratio of their amplitudes delivers the distribution of the saturation flip Simplified scheme of the presaturated TurboFLASH sequence.
angle α in the imaged volume: For magnetization preparation, a minimum-phase Shinnar-Le Roux pulse was chosen in this study to improve slice selectivity and minimize T 1 dependency between preparation and readout. 39

GRE VFA method for SNR measurement
To extract the characteristic SNR independent of the sequence parameters and relaxation, one can use the steady-state GRE signal with amplitude, as follows: where 2 , ρ 0 denotes the proton density, and B 1 − is the reception sensitivity.Their product, S 0 , corresponds to a signal acquired in the fully relaxed state (FA = 90 • , TE → 0, TR → ∞).The SNR map is thus defined by the ratio of S 0 and the noise SD σ and will be further referred to as SNR 0 .In practice, from images acquired with two (or more) GRE sequences with different parameters, one can find S 0 and T 1 as a concurrent value.The obtained T 1 maps can indirectly characterize the reliability of the method used to measure the SNR 0 , if they provide physically adequate values and remain constant for the same patient and different coils being compared.
The VFA method using the GRE signals with different FA and fixed TR finds a wide application (generally, for T 1 mapping) due to its simple implementation.Using two FAs (α 1 and α 2 ) and neglecting T 2 * relaxation, we can directly determine the parameters through the ratio of signal intensities r = S 1 /S 2 as follows:

2.3.3
Postprocessing for SNR and B 1 − assessment In the case of multichannel reception, a rigorous SNR estimation requires optimal channel combining to decorrelate or prewhiten the received noise.This implies combining receive elements to create uncorrelated virtual channels with unit noise variance. 40The necessary channel weights can be obtained from the noise covariance matrix where N is the number of data samples; i and j denote channels; and n j (k) is the 0 voltage signal for the reception element j in the kth data sample.The factor of 0.79 corresponds to the noise equivalent receiver bandwidth. 40A noise map acquisition needs additional scanning without RF power.Then, one can perform the Cholesky factorization of the resulting R as LL H , with L being the lower triangular matrix, and then use L −1 as the decorrelation matrix.The prewhitened signal is given by The GRE images are reconstructed by applying the sum of squares to these prewhitened signals.It should be noted that the prewhitening procedure directly gives SNR 0 in the GRE signal equation instead of S 0 .
Applying Eqs. ( 4) and ( 5) at each pixel of the reconstructed images yields the SNR 0 and T 1 maps.The reliable parameter calculation also requires correction for the inhomogeneous FA distribution in the volume using the B 1 + map.The real FA for map calculation is defined as , where α GRE nom. is the nominal FA used in the GRE sequence, α GRE (r) denotes the actual FA corresponding to the position r, α satTFL nom. is the nominal FA used in the satTFL sequence to obtain the B 1 + map, and α satTFL (r) represents the actual angle on the B 1 + map.In addition, α GRE (r) must be corrected for a nonideal excitation profile of the GRE pulse, the deviation of which from a rectangular shape reduces the resulting FA.In the present work, we consider 3D-GRE acquisition.The slab profile was calculated from the pulse shape using the Bloch equations and provided average correction factors for the nominal FA for each slice in the excited slab.
Finally, to derive the B 1 − profile, one can apply a low-pass filter (e.g., polynomial) to the resulting SNR 0 maps, assuming a smooth reception profile varying slower than proton density. 41In this study, we used an 8th degree polynomial filter.The fitting was performed separately for each slice to exclude the error propagation from the boundary slices to the entire fitted volume.

Protocol parametrization
The protocol parameters (TRs and FAs) to assess the transmission and reception coil profiles were chosen based on an analytical analysis of noise-induced error and limitations imposed by the rS mode.The rationale is described in the Supporting Information.
In the case of the transmission profile study using the satTFL sequence, the signal analytical expression provided an estimate of the saturation FA α uncertainty.The values of α and β were chosen to minimize this error for a dynamic FA range of ±30%.
To evaluate the reception profile with the GRE-VFA method, the parameters reducing the noise-induced SNR 0 error were chosen based on the error propagation law and Eqs. ( 3), (4), and (5).Experimental restrictions included a limited in vivo acquisition time of about 10 min for the entire SNR 0 map protocol, a minimum planar resolution of 2.0 × 2.0 mm 2 , and the maximum FAs available for a given TR in the rS mode.

Experimental setup
Sequence testing was performed on a Magnetom 7T MRI scanner (Siemens Healthineers, Erlangen, Germany) with a SC72 whole-body gradient (maximal amplitude 100 mT/m and slew rate 200 T/m/s).The rS protocol was tested on two coils: 1-transmit/32-receive Nova Medical Head Coil (Nova Medical, Wilmington, MA, USA) and 1-transmit/8-receive pre-industrial prototype at an intermediate stage of development referred to as MW1 (Multiwave Imaging, Marseille, France).The protocol used was drawn up based on the previous discussion; the selected parameters are presented in Section 3. The calibration setup procedure consisted of the following steps: localizer -(rS B 0 mapping -shim calculation) x2 -V ref mapping with rS-satTFL -estimation and manual adjustment of V ref .We performed at least two shimming cycles to homogenize B 0 to an extent that is consistent with the automatic procedure.Automatic adjustments of directional couplers and resonance frequency were launched before each measurement.To avoid an automatic V ref calibration and stay in rS conditions, we acquired localizer and B 0 maps with a preset V ref of 250 V.Then, using the reference voltage map obtained with the satTFL sequence at the same 250 V, we determined the optimal V ref .It corresponded to the value providing the target FA in the central region of the image, selected manually.This voltage was applied to the remaining sequence protocols in the experiments.We should note that to unconditionally comply with the 6-min limit for successively played sequences, delays before and after sat-TFL need to be added (TA/2 and TA, respectively, given that the first block is a less energetic reference one).
The rS protocol was first tested on a phantom with a commercial Nova coil.To ensure the correctness of the rS power limitations, we compared the energies of the played sequences measured by DICO and available via the log file with values calculated in the sequence source code.Then, the rS protocol was tested on 3 healthy volunteers in accordance with local internal review board regulations.Studies were approved by the local and national ethics committees (CPP Sud Méditerranée 4 [No.180913] and IDRCB 2018-A011761-53, respectively).All volunteers gave a written informed consent.
First, we evaluated the rS protocol in vivo applicability on the Nova coil on the first volunteer by performing a setup procedure followed by B 1 + mapping and image-quality check.In this case, in addition to the rS satTFL-based B 1 + maps, the distribution of the transmitted magnetic field was analyzed using the "gold-standard" actual FA imaging (AFI) sequence.The latter was played in a nonrestricted conventional mode, as the powerful pulses used violated the rS limitations.However, the fact that we performed this testing on a commercial coil that had undergone a full simulation and test procedure ensured the safety of the experiment.
Then, a complete rS assessment procedure was performed on Nova and MW1 coils, both in multichannel reception mode, on the 2 remaining volunteers.The study included all the necessary adjustment steps described earlier, as well as B 1 + mapping, T 2 *-weighted imaging, and T 1 /SNR 0 /B 1 − mapping.To allow for in vivo acquisition within the rS mode, the MW1 coil passed bench tests to ensure that the coil was mechanically and electrically stable after high load stress.In vitro testing on phantom also demonstrated absence of arcs and repeatable performance over time.

Anatomy of the rS protocol
The sequence parameters of the developed rS protocol are given in Table 2.They were selected based on those used in routine clinical practice and the optimization described in the Supporting Information.Being adjusted to the rS conditions, the protocol settings maintain sufficient spatial resolution to adequately evaluate images and reasonable TAs to prevent bulk motion.B 1 + mapping with the satTFL sequence was performed with  = 4 • and  = 60 • .For experimental SNR 0 (and concurrent T 1 ) mapping with the 3D-GRE-VFA method, we chose the parameter set of TR = 160 ms, FA 1 = 16 • , and Parameters of the restricted specific absorption rate (rS) protocol used in the experiments (the parameters of the presaturated TurboFLASH [satTFL] saturation pulse are shown in parentheses).

Protocol verification
Table 3 lists the sequence energies measured during the phantom experiments.They include the rS-adapted setup procedures as well as automatic adjustments given for comparison, all played at V ref = 250 V.In addition, we provide the energies of the characteristic sequence protocols (satTFL, 2D GRE, and 3D GRE) for two boundary V ref of 180 V and 300 V.The measured values correspond to the predicted values calculated in the sequence code with a relative deviation of less than 1%.
In the proposed rS approach, the maximum energy for every 10 s is restricted to 2 W⋅s, and for every 6 min to 36 W⋅s, and the implemented rS protocol does not exceed these limits.The automatic adjustment of the transmitter amplitude takes about 2 s and is characterized by an energy of the order of 2.5 W⋅s., which therefore exceeds the short-term power threshold.In contrast, manual setting of V ref respects the rS conditions.In the same way, the automatic shimming exceeds the 10-s limit and must therefore be replaced by the custom B 0 -mapping procedure.
For the in vivo protocol validation on the Nova coil, the FA distributions (after the V ref adjustment) for satTFL and AFI are shown on Figure 2.During in vivo experiments, the automatically estimated SAR level available for commercial coils did not exceed 1% of the limits established by the coil and scanner manufacturer.The measured power of the rS-adapted sequence for B 1 + mapping was about 100 times less than that of the conventional AFI method (3.7 J for 80.3 s vs. 1628.3J for 333 s at V ref 200 V).

3.3
Protocol performance for pre-industrial prototype assessment Figure 3 presents an example of data obtained with the restricted SAR protocol for a home-made prototype MW1 and Nova coil on one of the volunteers.It includes FA maps normalized to nominal FA of the satTFL sequence, T 1 maps, SNR 0 maps, and sensitivity profiles (B 1 − ) derived according to the procedure described previously.Images were acquired in the axial plane, and the maps were also reconstructed for sagittal and coronal orientations.Figure 4 shows an example of T 2 *-weighted images.The images acquired for the second volunteer are similar.
As measures for evaluating and comparing coils, we considered the coefficient of variation (CV) of the B 1 + field (CV = σ[B 1 + ]/< B 1 + >), coil efficiency, and average SNR 0 .The B 1 + CV describes the spread of the transmit magnetic-field inhomogeneity within the masked region.Coil transmit efficiency (TrE) is defined as B 1 + inv.targ / √ P. Here, P denotes the transmitted power equal to V ref 2 /R with coil impedance R = 50 Ω, and B 1 + inv.targ is a constant coefficient corresponding to the amplitude of a rectangular inversion RF pulse with the length of  = 1 ms: B 1 + inv.targ = π/(2πγ⋅τ) where γ = 42.57⋅10 6Hz/T -gyromagnetic ratio.These values characterizing the Nova and MW1 coils obtained for the 2 volunteers are provided in Table 4.

DISCUSSION
For the proposed rS protocol, the average transmitted RF power of the played sequences remains at least Measured energies of automatic calibration and restricted specific absorption rate (rS) sequences verified on a phantom.The measured values correspond to the calculated energy with a relative deviation of less than 1%.  100 times lower than that allowed by the conventional mode for the V ref range used in MRI exams, ensuring exam safety (Table 3).Replacing automatic procedures such as B 0 mapping for shimming and V ref adjustments guarantees that the entire experiment satisfies the imposed rS conditions.At the same time, the proposed rS protocol allows a proper assessment of the coil transmission and reception Examples of T 2 *-weighted images acquired under the rS mode on the first volunteer: Multiwave Imaging 1 (MW1) (A) and Nova (B).

T A B L E 4
Coefficient of variation (CV) of B 1 + , transmit efficiency, and average SNR 0 for MW1 and Nova coils.profiles.rS-adapted B 0 mapping is characterized by accuracy and coverage (20 slices with 100% spacing) that allow the shimming with a sufficient degree of homogeneity to image the region of interest.The rS satTFL-based B 1 + map shows a good consistency with the AFI results (correlation coefficient = 0.89) for two reasons.First, the protocol enables different pulse type, including Shinnar-Le Roux, which provides a sharp excitation profile minimizing the FA error.Second, the adjustment of the satTFL parameters (α and β angles) reduces the noise-induced error in the α determination.Thus, the rS protocol with properly chosen settings characterizes the coil excitation profile with an accuracy comparable to the nonrestricted standard mode (i.e., with AFI), being 100 times less powerful for the considered configuration and sequences played.The proposed protocol allows characterizing and comparing coils with different architectures and SNR regimes in terms of excitation homogeneity, transmit efficiency, and achievable SNR.Between subjects, we observed stable values of B 1 + CV, TrE, and SNR 0 obtained with the rS protocol for each coil (Table 4).Variations in the calculated V ref and thus TrE between subjects emerge due to different coil loading or differences in the manually selected V ref -averaging regions on the satTFL maps.The same analysis can be made on the CV of B 1 + and average SNR 0 , which depend on the subject as well as the chosen mask.Thus, with proper postprocessing, the range of parameter deviations reflects the immunity of the coil performance to the patient load.An unchanged T 1 map when examining the same volunteer with different coils (Figure 3) also indicates the reliability of the VFA method used to measure SNR 0 .
For the SNR 0 estimation, it should be noted that at the periphery of the imaged volume, B 1 + inhomogeneity leads to a decrease in FA (up to 2 times) and thus to a drop in accuracy.For the selected parameters, halving the FAs leads to an increase in the relative SNR 0 error up to 3 times.Moreover, the imperfection of the slab profile leads to an additional decrease in FA, which can reach 40% of the nominal FA value for boundary slices.For multichannel reception, the effect of lower FAs on the periphery is partially mitigated by increased B 1 − near the receiving elements.In our case, doubling the SNR 0 compensates for the increase in SNR 0 error due to the halving of FA.
To test the rS protocol for different SNR regimes, we chose the prototype of MW1 coil undergoing development.To fully evaluate the stability of our approach, a larger sample of patients and coil configurations needs be considered.However, the performed experiments show that the rS protocol provides meaningful data even with considerable power constraints.Thus, it can be used to monitor coil design without SAR simulations and phantom validations.Note that the manufacturer still has to perform some safety checks (i.e., electrical and mechanical stability).
We should mention that not all experiments admit the rS mode.Its limitations include, for example, sequences with powerful (especially inversion or refocusing) pulses, such as MPRAGE and spin echo.Such sequences must be replaced by compatible analogues, if they exist, when operating in the rS mode.
Additionally, the rS approach implies that the transmitted power cannot be focused into an area smaller than 10 g (a sphere with a diameter of 2.7 cm considering water density).This is valid for volume coils and most of transmit arrays; however, some specific coil types with potentially localized electric field must be considered more attentively.In fact, we use the regulations for the case in which a patient's body does not contain internal conductive objects and suppose that the transmit elements do not fit close to the head.The conditions used also need to be reconsidered for higher field strengths where more localized hotspots could potentially be reached.
The high resolution (0.5 × 0.5 × 4.0 mm 3 ) of T 2 *-weighted images (Figure 4) with a reasonable SNR achievable despite drastic power constraints raises the question of extending the rS protocol to other studies that may require a severe SAR limitation.][44] Here again, it is essential to consider the high possible localization of the electric field.In addition to RF energy restriction, safe implant studies include conditions on gradients to avoid additional heating of surrounding tissues, which should be considered separately.
To go further, it can be of interest to directly restrain the temperature increase under the assumption of the rS mode.We considered the heating of a 10-g tissue sphere absorbing all the transmitted RF power. 45In this case, even a conservative thermal model without perfusion will allow us to relax the limitations of the rS mode from 1.5 to 2 times, depending on the MRI exam time, thereby providing greater freedom in experiments and a further increase in the available resolution.
The rS protocol can also be adapted for pTx, which uses several transmit channels to homogenize the B 1 + field in the sample.SAR calculations for pTx represent a more complex issue and must consider the RF fields from independent channels. 46,47However, the implementation of the rS approach remains as simple as for the single-channel (sTx) mode, and relies on an upper local SAR estimation, assuming an unrealistic power absorption in 10 g of tissue. 29In this case, the rS protocol establishes constraints on the total energy of all channels.
The rS protocol transition to other systems requires recompilation of the corresponding pulse sequences.It is necessary to check the DICO and frequency-adjustment energies and maximum allowed pulse amplitudes in advance, and edit the coil file if needed.

CONCLUSION
The restricted SAR approach facilitates the design of RF coils, as it makes the in vivo testing phase unconditionally safe and eliminates prior validations.The proposed protocol limits the transmitted RF power such that even for highly localized power absorption, the worst-case local SAR remains below a safe threshold.The power restrictions used do not depend on the SAR distribution, so the rS approach can be safely played with various sequences (i.e., GRE, satTFL), regardless of coil design and volunteer anatomy.Coil-independent safety conditions also make the rS protocol an appropriate tool for in vivo comparison of different coil designs in the early development stages.
Starting with the simplest idea of energy conservation and applying the drastic power constraints, we are still able to obtain representative images.Human studies under the rS mode conducted in this paper demonstrate the ability to evaluate a coil prototype with several sequences in less than 17 min, even at a low power supply.The rS protocol enables parameter adjustment, and the proper settings provide high-resolution images and coil-transmission analysis with accuracy comparable to conventional mode.This approach, being versatile, can be translated to other field strength and body regions, as well as to parallel transmission.

F I G U R E 2 F I G U R E 3
Flip-angle (FA) maps (B 1 + maps) obtained with restricted specific absorption rate presaturated TurboFLASH (left) and actual FA imaging (AFI) (right), and comparison of the FA distributions for commercial Nova coil over all slices.From top to bottom: normalized B 1 + maps, T 1 maps, SNR 0 maps, and sensitivity profiles obtained with the restricted specific absorption rate protocol for the second volunteer: Multiwave Imaging 1 (MW1) coil (A) and Nova coil (B).

B 0 -map satTFL B 1 + -map 2D GRE, T 2 * 3D GRE VFA FA 1 /FA 2
2 = 8 • , as it provided the minimum error under the experimental constraints.To neglect T 2 * relaxation, the GRE-VFA experiments used the smallest possible TE (2.8 ms).An additional short scan of 3D GRE (TR = 6 ms, TA = 17 s) with zero pulse amplitude and preserved FOV and resolution (matrix size = 256 × 256) served to calculate the decorrelation matrix for data prewhitening.