Received: 9 February 2024 Revised: 29 August 2024 Accepted: 1 October 2024

DOI: 10.1002/mrm.30349

TECHNICAL NOTE

High dynamic range B+ 1

mapping for the evaluation of
parallel transmit arrays

Jörg Felder1,2 Markus Zimmermann1 N. Jon Shah1,3,4,5

1
Institute of Neuroscience and
Medicine – 4, Forschungszentrum Jülich, Abstract
Jülich, Germany Purpose: Demonstration of a high dynamic-range and high SNR method
2
Faculty of Medicine, RWTH Aachen for acquiring absolute B+1 maps from a combination of gradient echo and
University, Aachen, Germany
actual-flip-angle measurements that is especially useful during the construction
3
Institute of Neuroscience and
Medicine – 11, Forschungszentrum
of parallel-transmit arrays.
Jülich, Jülich, Germany Methods: Low flip angle gradient echo images, acquired when transmitting
4
JARA–BRAIN–Translational Medicine, with each channel individually, are used to compute relative B+1 maps. Instead
Aachen, Germany
of computing these in a conventional manner, the equivalence of the problem to
5
Department of Neurology, RWTH
the ESPIRiT parallel image reconstruction method is used to compute B+1 maps
Aachen University, Aachen, Germany
with a higher SNR. Absolute maps are generated by calibration against a single
Correspondence actual flip-angle acquisition when transmitting on all channels simultaneously.
Jörg Felder, Forschungszentrum Jülich,
52425 Jülich, Germany.
Results: Depending on the number of receiver channels and the location of the
Email: [email protected] receive elements with respect to the subject being investigated, moderate to high
gains in the SNR of the acquired B+1 maps can be achieved.
Funding information
Helmholtz Association Conclusions: The proposed method is especially suited for the acquisition of
B+1 maps during the construction of transceiver arrays. Compared to the original
method, maps with higher SNR can be computed without the need for additional
measurements, and maps can also be generated using previously acquired data.
Furthermore, easy adoption and fast estimation of receiver channels is possible
because of existing highly optimized open-source implementations of ESPIRiT,
such as in the BART toolbox.

KEYWORDS
B+1 mapping, ESPIRiT, high SNR, ultrahigh-field

1 I N T RO DU CT ION sequences has also increased. Prominent examples are the

double angle method,1 actual flip angle imaging (AFI),2
With the introduction and increased application of the Bloch-Siegert shift method,3 the dual refocusing echo
ultrahigh-field (UHF) imaging systems, parallel trans- acquisition mode (DREAM) technique,4 and the presatu-
mission (pTX) techniques have found wide interest. rated turboFLASH method.5
Because all pTX methods require the acquisition of 2D Another driving factor for transmit field mapping has
or 3D transmit sensitivity maps, interest in B+1 mapping been the evaluation of novel transmit arrays where B+1

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the
original work is properly cited.
© 2024 The Author(s). Magnetic Resonance in Medicine published by Wiley Periodicals LLC on behalf of International Society for Magnetic Resonance in Medicine.
1298 wileyonlinelibrary.com/journal/mrm Magn Reson Med. 2025;93:1298–1305.
 15222594, 2025, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/mrm.30349 by CAPES, Wiley Online Library on [09/01/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FELDER et al. 1299

mapping is used to validate the simulation model by com- separation of signal and noise, possibly by exploiting a
paring simulated versus measured transmit field distri- larger null space.
butions. This step is essential as the simulation model is
subsequently used to derive the specific absorption rate
(SAR) properties of the array.6 Validation is conducted 2 METHODS
on phantoms. Generally, water-based phantoms are used
because they should provide a high SNR and a single reso- In the original ESPIRiT algorithm, the receive field maps
nance line.6 However, because of the complex interactions CRX,j found in a SENSE-like problem are estimated by
between the RF fields and the phantom at UHF, the maps
acquired from phantoms usually show strong interference P F CRX,𝑗 x = d𝑗 , (1)
patterns that require high dynamic range field maps.
Because the dynamic range of many mapping with the underlying image, x, the sampling matrix, P, the
sequences is not sufficient to depict B+1 maps of individ- discrete Fourier transform, F, the measured k-space data,
ual transmit elements of a pTX array,7 several solutions d, and the (receive) channel index, j, ranging from one
have been presented. One method proposes the acqui- to the total number of receive channels. It can be shown
sition of several maps when transmitting with multiple that a system matrix can be constructed by running a slid-
elements in an interferometric way and the subsequent ing window through a fully sampled auto calibration (AC)
decomposition of these maps into those for the individual region of the k-space data and that the sensitivity maps are
coil elements.8–10 Another method, presented by van de those eigenvectors of the system matrix that correspond to
Moortele et al.,11 used low flip-angle gradient echoes to the eigenvalues of one.
derive relative maps, which were then calibrated to abso- The problem of estimating the transmit field maps,
lute values using a single AFI acquisition that transmits CTX,I , from 2D complex (magnitude and phase), low
on all channels. This method has been used to evalu- flip-angle and spoiled or fully relaxed gradient echo (GRE)
ate the field distribution of a number of pTX arrays, for images, acquired when transmitting with one channel at
example.12–14 However, the method shows low SNR and a time, can be rewritten in the same way. Given that the
potentially requires signal averaging.15,16 Similarly, Sete- excited signal, xi , can be approximated as
sompop et al.17 used spoiled gradient echoes (SPGRs)
whereby relative B+1 maps are acquired in combination sin(𝛂i )M0 ≈ 𝛂i M0 = xi , (2)
with a single absolute B+1 map. This method has since
been extended by Padormo et al.18 to include multiple RF with the flip angle, 𝛂i , given by 𝛂i = γ B+1 ,i TRFE , and B+1 ,i
drive levels for the SPGRs, which renders the acquisition being the transmit field of the (transmit) channel i within
of an absolute map obsolete. the range 1 to total number of transmit channels, γ is
In this work, we extend the method developed by van the gyromagnetic ratio, TRFE is the hard pulse equivalent
de Moortele et al.11 by replacing their post-processing tech- pulse duration, M0 the equilibrium magnetization, and xi
nique for estimating the relative transmit fields of individ- is the GRE image acquired when transmitting on chan-
ual coils with the ESPIRiT algorithm. Although originally nel i. By substituting the left-hand side of Eq. (2) with the
designed to estimate receive coil sensitivities in parallel relative transmit field CTX,i , then applying a Fourier trans-
imaging, we show that ESPIRiT’s mathematical frame- formation on both sides and multiplying with a “sampling
work is equally applicable to transmit fields. ESPIRiT uses matrix” P = 1, one obtains.
an eigenvalue decomposition of the system matrix to dis-
tinguish between signal and noise, creating relative B+1 P F CTX,i M0 = P F xi = di , (3)
maps with enhanced SNR. This is achieved by isolating
the dominant eigenvectors, which most accurately repre- where di = F xi is the Fourier transform of the recombined
sent the true sensitivity profiles while suppressing com- reconstructed image xi . It can be seen that Eqs. (1 and 3)
ponents associated with noise. We further compare this are equivalent. Thereby, we assume that the transmit field
approach to the method presented by Brunner et al.,19 maps, CTX,i , obtain features similar to those of the receive
which uses a voxelwise eigenvalue decomposition of the field maps, CRX,j , in terms of spatial smoothness.
outer product of the transmit and receive channels to In this formulation, the acquisition of uncombined
simultaneously determine the transmit and receive sensi- GRE images is not required, because only combined
tivities. Although the method proposed by Brunner et al.19 images are required for map generation. The images are
also reduces noise by keeping only the dominant eigen- then projected back into k-space by 2D Fourier trans-
values, our experiments demonstrate that ESPIRiT pro- formation, and the k-space data is used as input for
duces superior results, likely because of a more effective the ESPIRiT algorithm. Assuming an array with a total
 15222594, 2025, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/mrm.30349 by CAPES, Wiley Online Library on [09/01/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1300 FELDER et al.

number of n transmit elements and a matrix size of nx × ny , the following parameters: FA = 68◦ , TE1/TE2/TE3 =
the ESPIRiT algorithm returns a coefficient matrix, CTX , 2.04/6.12/10.20 ms, TR2 = 1000 ms, TR2/TR1 = 5, matrix
with shape (n, nx , ny ). size = 64 × 64, 16 slices, slice thickness = 4 mm, aver-
This matrix is equal to the transmit field maps of all ages = 1, slices = 16, and a total duration = 9 min 51 s. The
channels in arbitrary units. A normalization to obtain used sequence enabled simultaneous B+1 and ΔB0 map-
magnitudes of the relative B+1 maps in analogy to11 can be ping,21,22 where the static field map was used to correct the
derived by flip angle maps.23 The AFI images were acquired when
transmitting in CP mode, with all TX elements active,
CTX,i
RelB1Magi = ∑ . (4) and only the slice located at the position identical to the
|CTX,CP,i |
| | GRE images was considered. Combining the relative B+1
maps of all channels linearly with the appropriate phases
Normalization in the above equations refers to circu-
to achieve CP allows the computation of a scalar scal-
lar polarized (CP) transmission using all channels so that
ing factor that is valid for all voxels in contrast to other
CTX,CP,i is given by
techniques24 where each voxel needs to be normalized
CTX,CP,i = CTX,i ei𝜑CP.i , (5) individually. This scale factor was subsequently used to
convert relative B+1 maps into absolute ones.
with φCP,i containing the CP mode phase of the i-th trans- The wavelet-based approach described in Donoho
mit array element, and the sum is taken over all transmit and Johnstone25 computes very high SNR values for the
channels. Normalization of the transmit efficiency phase ESPIRiT-based maps, therefore, restricting its applica-
to that of the first array element, if desired, is achieved by bility to Gaussian distributions only. SNR comparisons
can also be based on statistical measures, like the
(
)
CTX,i Kullback–Leibler-divergence (KL-divergence),26 which
RelB1Phasei = angle . (6) measures the relative difference between two probability
CTX,1
distributions. However, the singular value deomposition
To evaluate the proposed reconstruction method, rela- (SVD)-based maps show a slightly different distribution to
tive B+1 maps were acquired using an 8TX/32RX coil array those computed with the original and the novel method,
(Nova Medical) and an eight-channel transceiver array. which renders KL-divergence unsuitable in this case. This
The second array uses off-center fed dipoles arranged on is why we compare the SNR of the B+1 maps generated with
a cylindrical former with a free diameter of 260 mm and the different algorithms using the autocorrelation meth-
is intended for use in a combined PET/MR system. B+1 ods described in Sim et al.27 It should be noted that the SNR
maps for each of the arrays were acquired with the orig- values obtained with this method depend on the way the
inal method proposed by van de Moortele et al.11 and by images are cropped. Therefore, we refer to the SNR values
the ESPIRiT techniques proposed here. For comparison, obtained in this way as relative SNR, and all calculations
the approach proposed by Brunner et al.19 was used to are based on consistently cropping the outer 10 pixels.
derive maps from the same input data. GRE images were The dynamic range of the proposed B+1 mapping was
acquired when transmitting with each channel individu- evaluated using simulations. For this purpose, the B+1 and
ally, using the following parameters: FA = 10◦ , TE = 4 ms, B−1 maps of the PET/MR array were simulated in 3D elec-
TR = 200 ms, matrix size = 128 × 128, averages = 1 and a tromagnetic simulations using the finite element method
total duration = 27 s with the shortest TR according to the as implemented in the frequency domain solver of CST
range suggested in Cloos et al.20 Both phantom and in Microwave Studio (Dassault Systems, Vélizy-Villacoublay,
vivo images were acquired from the 8TX/32RX coil array. France). The field maps were fed into a Bloch equation
Because the PET/MR array has not yet been approved for simulator (JEMRIS28 ) where AFI and GRE images were
in vivo measurements, it was only used to acquire phan- simulated for a concentric two-sphere phantom with
tom images. The phantom used was a cylindrical water T1 = 1000 ms/2000 ms and T2 * = 25 ms/30 ms. The over-
phantom (diameter: 170 mm) doped with 50 mmol NaCl. all simulation workflow is described in more detail in
All measurements were carried out in a 7 T TERRA sys- Horneff et al.29
tem (Siemens Healthineers) equipped with eight transmit
channels. In vivo images were acquired after approval of
the study by the local ethics committee and after having 3 RESULTS
obtained written informed consent from the volunteer
subjects. Figure 1 shows absolute B+1 maps and correspond-
The calibration to the absolute maps used data ing unreferenced phase maps acquired with the recon-
acquired with a 3D AFI method acquisition with struction described in van de Moortele et al.11 These
 15222594, 2025, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/mrm.30349 by CAPES, Wiley Online Library on [09/01/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FELDER et al. 1301

FIGURE 1 Absolute B+1 maps and unreferenced phase maps obtained with the method described by van de Moortele et al.,11 using the
ESPIRiT-based reconstruction and the singular value deomposition (SVD) approach described by Brunner et al.:19 (A) for the Nova 8TX/32
array in vivo (A) and in a phantom (B) and for the PET/MR array prototype (C). All maps are subjected to threshold masking to remove
background noise.

maps were obtained from the same measurement, but from our novel method mostly displays removed noise.
were reconstructed with the ESPIRiT algorithm and the The quantitative comparison of the relative maps shown
SVD-based reconstruction method described by Brunner in Figure 2 is given in Table 1 as the relative SNR based
et al.19 for two different RF coil arrays. An SNR gain in the on the autocorrelation method. Again, it can be seen that
magnitude maps created using the ESPIRiT reconstruc- our novel method consistently provides higher SNR by a
tion is visible compared to those obtained with the original factor of 3 to 4 for all SNR estimators given in Sim et al.27
method and, to a lesser extent, compared to those cre- compared to the methods proposed by van de Moortele
ated with the SVD-based approach. It can further be seen et al.11 and Brunner et al.19 and achieves an infinitely
that exploitation of spatial smoothness through the appli- larger SNR for the National Electrical Manufacturers Asso-
cation of a noise-dominated null-space within the ESPIRiT ciation (NEMA) standard method. The SVD based method
reconstruction creates higher SNR and smoother phase described by Brunner et al.19 produces a lower SNR for all
maps. The consistency of the phase maps is demonstrated SNR estimators given in Sim et al.,27 compared against the
in Figure S1 where the phases have been normalized to the original method described by van der Moortele et al.11
first array elements according to Eq. (6). Simulated B+1 maps for a single transmit element of
The SNR gain of the magnitude maps also holds for the PET/MR array are shown in Figure 3. Compared with
the CP mode relative B+1 maps generated by linear super- the directly computed map from the AFI simulation, both
position, as described above and shown in the first row of B+1 maps based on GRE images for the relative maps and
Figure 2. In addition, difference images from the CP mode calibration with a single CP mode AFI acquisition show
maps are given in the second row of Figure 2. It can be seen a higher dynamic range—as can be seen by the larger
that the SVD-based approach creates systematic deviations range of isocontour lines in these maps. Direct compari-
from the original method, whereas the difference image son of the B+1 maps computed with the original method
 15222594, 2025, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/mrm.30349 by CAPES, Wiley Online Library on [09/01/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1302 FELDER et al.

FIGURE 2 (First row) Circular polarized (CP) mode relative B+1 maps reconstructed with the method proposed by van de Moortele
11
et al., the modified method based on the ESPIRiT algorithm presented here and the singular value deomposition (SVD)-based method
described by Brunner et al.19 (Second row) Difference maps with respect to the original method of the CP mode maps shown above. All maps
were subjected to threshold masking to remove background noise.

TA B L E 1 Comparison of the quality of the relative B+1 maps obtained with the three reconstruction methods for different coils and
objects being imaged.

Relative SNR original method Relative SNR novel method Relative SNR SVD-based method
Nova coil (phantom) 3.59–3.65 (2.75) 11.37–12.62 (∞) 2.51–2.56 (2.53)
Nova coil (in vivo) 2.6.78–6.90 (4.03) 15.35–16.10 (∞) 4.75–4.81 (3.76)
PET/MR coil (phantom) 5.11–5.25 (3.45) 4.93–5.10 (∞) 4.33–4.40 (3.63)

Note: The range of relative SNR given spans the range computed with the three SNR estimators given in Sim et al.27 The value in parentheses represents the
SNR value computed according to the National Electrical Manufacturers Association standard method 4.

FIGURE 3 Simulated single element B+1 maps obtained with the AFI method (nominal flip angle [FA] = 75◦ ) (left), the novel method
based on ESPIRiT (middle), and the original method (right). The images show isocontours in red for 30%, 40%, and 50% of the maximum value
for the AFI map and 10% to 50% in 10% steps for the other two maps. All three maps have been threshold-masked to remove background noise.

by van de Moortele and the ESPIRiT-based reconstruc- 4 DISCUSSION AND

tion reveal the smoothing effect of the ESPIRiT algorithm. CO NC LUSIONS
Although both maps are plotted for a matrix size of 64 × 64,
the middle map shows smoother isocontour lines, which We have demonstrated the application of the ESPIRiT
also contributes to the visual impression of a higher SNR. algorithm, widely used in parallel imaging reconstruction,
This is the case even though no noise was added to the for computing relative B+1 maps from a set of gradient echo
k-space data, which was simulated using the JEMRIS images. Similar to the established method proposed by
Bloch equation simulator. van de Moortele et al.,11 the ESPIRiT-based reconstruction
 15222594, 2025, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/mrm.30349 by CAPES, Wiley Online Library on [09/01/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FELDER et al. 1303

normalizes the relative transmit sensitivity maps to the for this case was derived in an identical way to the one
CP-mode. This normalization enables the efficient gener- used to generate Figure 3 and are shown in Figure S2. It
ation of absolute sensitivity maps using a single absolute can be seen that the dynamic range obtained in this way
flip-angle measurement in CP-mode, for instance, via the is increased compared to when transmitting with a sin-
AFI sequence. gle element, but does not reach the range of the original
However, unlike the method by van de Moortele or novel GRE-based mapping method. Figure S3 shows
et al.,11 the ESPIRiT-based approach significantly enhan- the influence of varying the nominal flip angle of the AFI
ces the SNR in the transmit B+1 fields. This improvement sequence on the dynamic range of the computed maps.
is attributed to the ESPIRiT model’s ability to account Again, it can be seen that the GRE-based methods retain
for both intra- and inter-voxel relationships, whereas higher fidelity.
the van de Moortele method considers only intra-voxel Although our implementation used the relatively slow
relationships, treating each voxel independently. but accurate AFI method to acquire a CP mode flip angle
When comparing the ESPIRiT-based approach to the map, it is worth noting that a faster mapping sequence,
SVD-based algorithm suggested by Brunner et al.19 , we such as DREAM,4 has previously been used for this pur-
observe superior noise suppression with ESPIRiT. This pose in Beqiri et al.30 In this case, a total acquisition
advantage is likely because of the fact that similarly to time (TA) in the range of TA = 2 min for an eight-channel
van de Moortele’s method, Brunner’s method does not TX array and a matrix size of 64 × 64 can be realized.
consider inter-voxel relationships. More importantly, we Speed enhancement, including the absence of averaging
found that the method proposed by Brunner et al.19 sys- and faster absolute mapping, is essential when using the
tematically deviates from the results of van de Moortele described mapping method as input for B1 -shimming or
et al.,11 whereas the ESPIRiT method does not. This parallel-transmit pulse calculation.20,31 A combination of
discrepancy may arise from the way in which Brun- the ESPIRiT-based reconstruction of B+1 maps in con-
ner’s method normalizes the relative sensitivity maps— junction with a fast DREAM acquisition might also be
ensuring that the TX/RX matrices are voxelwise useful in body applications where SAR does often not
unitary—whereas both the van de Moortele method and limit the achievable flip angle, but rather, the available
our ESPIRiT-based approach normalize voxelwise to the transmit power is the limiting factor. Therefore, the low
physics-based CP-mode, defined as the voxelwise sum of FA gradient echoes in combination with a fast DREAM
the channel magnitudes. or time-interleaved acquisition of modes24 -based acqui-
Improved SNR using this technique is highly benefi- sition of absolute maps in CP mode present promising
cial as it enables fast acquisition of high dynamic range avenues.
B+1 maps. The lower SNR of the previous implementa- Finally, normalization of B+1 phase maps to an arbi-
tion often required the use of signal averages to obtain trary channel may result in errors in parallel transmit pulse
high-quality maps in both phantom (e.g., Zhang et al.16 ) design (e.g., as reported in Krueger et al.32 ) This is why nor-
and in vivo measurements (e.g., Wu et al.15 ). In contrast, malization is often carried out with respect to two virtual
high-quality maps could be obtained with the ESPIRiT coil elements. The smoothness of the phase maps obtained
technique without averaging, as demonstrated for mul- using the ESPIRiT reconstruction without normaliza-
tiple coil arrays in Figure 1. The gain in SNR is espe- tion might be sufficient to avoid the normalization step
cially useful when evaluating transmit arrays during con- before pulse computation. This will be investigated in the
struction when no multi-channel RX array is present. In future.
this case, the ESPIRiT-based generation of B+1 achieves
the highest SNR gain in the phantom measurements.
ACKNOWLEDGMENTS
Because highly optimized open-source implementations
This work was funded, in part, by the “Innovation Go” ini-
of ESPIRiT are available (e.g., in the BART toolbox) our
tiative of the Helmholtz Association. We like to thank C.
proposed approach can be readily adopted. The fast imple-
Rick for proofreading the manuscript and J. Dammers for
mentation also reduced the computation time, which is
discussions on SNR evaluation. The AFI sequence used in
critical in pTX applications, because preparation time to
this study was supplied CEA NeuroSpin.
estimate the transmit sensitivity maps is a limiting factor
in widespread adoption.
It should be noted that interferometric acquisitions of CONFLICT OF INTEREST STATEMENT
+
B1 maps, as mentioned in the introduction, promise a N.J.S. and J.F. are co-founders of Affinity Imaging a
higher dynamic range then measurements obtained when spin-off company that manufactures high field MRI coils
transmitting with a single array element. Simulated maps for research purposes.
 15222594, 2025, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/mrm.30349 by CAPES, Wiley Online Library on [09/01/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1304 FELDER et al.

ORCID using a 16-channel transceiver coil at 7T. Magn Reson Med.

Jörg Felder https://orcid.org/0000-0001-6431-4322 2013;69:1285-1296.
Markus Zimmermann https://orcid.org/0000-0003 17. Setsompop K, Alagappan V, Gagoski B, et al. Slice-selective RF
-1273-5218 pulses for in vivo B+1 inhomogeneity mitigation at 7 tesla using
parallel RF excitation with a 16-element coil. Magn Reson Med.
N. Jon Shah https://orcid.org/0000-0002-8151-6169
2008;60:1422-1432.
18. Padormo F, Hess AT, Aljabar P, et al. Large dynamic range
REFERENCES relative B+1 mapping. Magn Reson Med. 2016;76:490-499.
1. Stollberger R, Wach P. Imaging of the active B1 field in vivo. 19. Brunner DO, Pruessmann KP. SVD analysis of Array transmis-
Magn Reson Med. 1996;35:246-251. sion and reception and its use for bootstrapping calibration.
2. Yarnykh VL. Actual flip-angle imaging in the pulsed steady Magn Reson Med. 2016;76:1730-1740.
state: a method for rapid three-dimensional mapping of 20. Cloos MA, Boulant N, Luong M, et al. kT-points: short
the transmitted radiofrequency field. Magn Reson Med. three-dimensional tailored RF pulses for flip-angle homogeniza-
2007;57:192-200. tion over an extended volume. Magn Reson Med. 2012;67:72-80.
3. Sacolick LI, Wiesinger F, Hancu I, Vogel MW. B1 mapping by 21. Amadon A, Boulant N. Simultaneous measurement of B0 - and
Bloch-Siegert shift. Magn Reson Med. 2010;63:1315-1322. B1 -maps with modified Actual Flip Angle Imaging sequence.
4. Nehrke K, Börnert P. DREAM-a novel approach for Proceedings of the 14th ISMRM; 2008; Toronto. p. 1248.
robust, ultrafast, multislice B1 mapping. Magn Reson Med. 22. Witschey WR, Reddy R, Elliott MA. Simultaneous B1 and B0
2012;68:1517-1526. mapping at 7T. Proceedings of the 16th ISMRM; 2010; Stock-
5. Chung S, Kim D, Breton E, Axel L. Rapid B+1 mapping using holm. p. 2852.
a preconditioning RF pulse with TurboFLASH readout. Magn 23. Boulant N, Cloos M, Amadon A. A simple and analytical way to
Reson Med. 2010;64:439-446. correct for ΔB0 inhomogeneity in the evaluation of B1 maps rely-
6. Zanche ND, van den Berg C, Brunner D, et al. ISMRM ing on flip angle measurements and non-selective square pulses.
Best Practices for Safety Testing of Experimental RF Hard- Proceedings of the 16th ISMRM; 2010; Stockholm. p. 4918.
ware. 2022 Available at: https://www.ismrm.org/safety/RF 24. Brunheim S, Gratz M, Johst S, et al. Fast and accurate
_Hardware_Safety_Testing_2022-03.pdf multi-channel mapping based on the TIAMO technique for 7T
7. Morrell GR, Schabel MC. An analysis of the accuracy of mag- UHF body MRI. Magn Reson Med. 2017;79:2652-2664.
netic resonance flip angle measurement methods. Phys Med Biol. 25. Donoho DL, Johnstone IM. Ideal spatial adaptation by wavelet
2010;55:6157-6174. shrinkage. Biometrika. 1994;81:425-455.
8. Brunner DO, Pruessmann KP. B1 + interferometry for the 26. Kullback S, Leibler RA. On information and sufficiency. Ann
calibration of RF transmitter arrays. Magn Reson Med. Math Stat. 1951;22:79-86.
2009;61:1480-1488. 27. Sim KS, Lai MA, Tso CP, Teo CC. Single image signal-to-noise
9. Malik SJ, Larkman DJ, Hajnal JV. Optimal linear combina- ratio estimation for magnetic resonance images. J Med Syst.
tions of array elements for B1 mapping. Magn Reson Med. 2009;35:39-48.
2009;62:902-909. 28. Stöcker T, Vahedipour K, Pflugfelder D, Shah NJ.
10. Tse DHY, Poole MS, Magill AW, Felder J, Brenner D, Jon SN. High-performance computing MRI simulations. Magn Reson
Encoding methods for B+1 mapping in parallel transmit systems Med. 2010;64:186-193.
at ultra high field. J Magn Reson. 2014;245:125-132. 29. Horneff A, Eder M, Hell E, et al. An EM simulation-based design
11. Van de Moortele P-F, Snyder C, DelaBarre L, Vaughan J, Ugur- flow for custom-built MR coils incorporating signal and noise.
bil K. RF Shimming at Very High Field with Multiple Element IEEE Trans Med Imaging. 2018;37:527-535.
+∕− 30. Beqiri A, Hoogduin H, Sbrizzi A, Hajnal JV, Malik SJ.
RF Coils: Calibration Tools from Ultra Fast Local Relative B1
Phase to Absolute Magnitude B+1 Mapping. Proceedings of the Whole-brain 3DFLAIR at 7T using direct signal control. Magn
13th ISMRM; 2007; Berlin. p. 1676. Reson Med. 2018;80:1533-1545.
12. Adriany G, Auerbach EJ, Snyder CJ, et al. A 32-channel lattice 31. Hsu Y-C, Lattanzi R, Chu Y-H, Cloos MA, Sodickson DK,
transmission line array for parallel transmit and receive MRI at Lin F-H. Mitigation of B+1 inhomogeneity using spatially selec-
7 tesla. Magn Reson Med. 2010;63:1478-1485. tive excitation with jointly designed quadratic spatial encod-
13. Sengupta S, Roebroeck A, Kemper VG, et al. A specialized ing magnetic fields and RF shimming. Magn Reson Med.
multi-transmit head coil for high resolution fMRI of the human 2017;78:577-587.
visual cortex at 7T. PLoS One. 2016;11:e0165418. 32. Krueger F, Aigner CS, Hammernik K, et al. Rapid estimation of
14. Solomakha G, Andreychenko A, Moortele P-F, et al. A coax- 2D relative B+1 -maps from localizers in the human heart at 7T
ial RF applicator for ultra-high field human MRI. IEEE Trans using deep learning. Magn Reson Med. 2022;89:1002-1015.
Biomed Eng. 2019;66:2848-2854.
15. Wu X, Schmitter S, Auerbach EJ, Moeller S, Uğurbil K, SUPPORTING INFORMATION
Van de Moortele P-F. Simultaneous multislice multiband Additional supporting information may be found in the
parallel radiofrequency excitation with independent slice- online version of the article at the publisher’s website.
specific transmit B1 homogenization. Magn Reson Med. 2013;70:
630-638.
16. Zhang X, de Moortele P-FV, Schmitter S, He B. Complex B1 Figure S1. Figure 1 of the manuscript repeated but with
mapping and electrical properties imaging of the human brain phases normalized to that of channel #1, thus showing
 15222594, 2025, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/mrm.30349 by CAPES, Wiley Online Library on [09/01/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FELDER et al. 1305

the consistency of the phases obtained with the different acquisition obtained interferometrically, novel method
reconstruction algorithms. based on the ESPIRiT algorithm and original method.
Figure S2. Simulated B+1 maps of the PET/MR array using
AFI in all-but-one encoding. (top row): raw maps and
(bottom row) maps decomposed for individual transmit How to cite this article: Felder J,
channels. Zimmermann M, Shah NJ. High dynamic range B+1
Figure S3. Simulated B+1 maps of the PET/MR array with mapping for the evaluation of parallel
varying nominal flip angles of 50◦ (top row), 75◦ (middle transmit arrays. Magn Reson Med.
row) and 90◦ (bottom row). The columns show, from right 2025;93:1298-1305. doi: 10.1002/mrm.30349
to left: AFI acquisition with a single transmit channel, AFI