Magnetic Resonance in Medicine 64:725–733 (2010)

T1 Corrected B1 Mapping Using Multi-TR Gradient

Echo Sequences
Tobias Voigt,1* Kay Nehrke,2 Olaf Doessel,1 and Ulrich Katscher2

This work presents a new approach toward a fast, simultane- techniques highly desirable. Several approaches to-
ous amplitude of radiofrequency field (B1) and T1 mapping ward fast and accurate B1 mapping have been devel-
technique. The new method is based on the ‘‘actual flip angle oped in the last decades.
imaging’’ (AFI) sequence. However, the single pulse repetition
The B1 transmit field can be measured via the double-
time (TR) pair used in the standard AFI sequence is replaced
by multiple pulse repetition time sets. The resulting method
angle method (8), which has been refined repeatedly to
was called ‘‘multiple TR B1/T1 mapping’’ (MTM). In this study, increase accuracy and/or reduce scan time (9–12). Fur-
MTM was investigated and compared to standard AFI in thermore, separate images based on multiple flip angles
simulations and experiments. Feasibility and reliability of are used for fitting suitable signal equations (13,14).
MTM were proven in phantom and in vivo experiments. Error Another approach comprises the application of stimu-
propagation theory was applied to identify optimal sequence lated echoes in multipulse sequences, where signals
parameters and to facilitate a systematic noise comparison with different flip-angle dependence are acquired (15–
to standard AFI. In terms of accuracy and signal-to-noise 17). The study presented in Dowel and Tofts (18) sug-
ratio, the presented method outperforms standard AFI B1
gests using the 180
signal void to map B1 fields. A
mapping over a wide range of T1. Finally, the capability
of MTM to determine T1 was analyzed qualitatively and
recently published B1 mapping method uses two RF
quantitatively, yielding good agreement with reference pulses to encode the flip-angle information into the spa-
measurements. Magn Reson Med 64:725–733, 2010. V C 2010 tial phase distribution (19). Finally, actual flip angle
Wiley-Liss, Inc. imaging (AFI) maps the B1 transmit field using a fixed
Key words: RF-field mapping; quantitative MRI; B1 mapping; flip angle in a dual pulse repetition time (TR) steady-
T1 mapping; actual flip angle imaging state gradient echo sequence (20). In recent years, AFI
has gained increasing popularity because it represents a
fast and potentially robust B1 mapping approach. Several
In MRI applications, spatially varying sensitivity pro- studies have been reported concerning different aspects
files of the radiofrequency (RF) transmit coils and of AFI, e.g., spoiling of transverse magnetization compo-
local variations of the conductive and dielectric prop- nents (21,22).
erties in the tissue may result in a spatially inhomo- Besides speed and signal-to-noise ratio (SNR), system-
geneous amplitude of the RF transmit field (B1). atic errors caused by T1 relaxation may be a problem in
Therefore, accurate B1 mapping is required for a vari- accurate B1 mapping. Some approaches suggest using
ety of applications. For instance, nonuniform B1 dis- repetition times much longer than T1 to eliminate these
tributions are an essential source of errors in quanti- errors (8,12,14), however, leading to impractical long
tative imaging. Thus, B1 mapping is frequently used imaging times. Other approaches are independent of T1
for postcorrection of, e.g., spin density (1), T2 (2), T1 (9,10,17–19). The AFI approach relies on a linearization
(3), or magnetization transfer measurements (4). More- of T1 relaxation and thus produces a systematic error in
over, B1 mapping allows the estimation of the electric the flip-angle estimation that depends on T1. On the
conductivity and local specific absorption rate (SAR) other hand, a variety of quantitative T1 mapping techni-
(5). In the framework of parallel transmission, RF ques involves postcorrection for B1 field inhomogene-
shimming (6) and accelerated local (SENSE) (7) rely ities via additional measurements of the B1 transmit
on accurate B1 field information. Here, the need for field (1,3,23). A recently published study provides a
sequential mapping of the B1 distribution of multiple method for concurrent B1 and T1 mapping based on
transmit channels makes particularly fast B1 mapping analyzing signals from two pulse trains with different
initial magnetization (24).
In this work, a generalization of the AFI approach is
1
Institute of Biomedical Engineering, University of Karlsruhe, Karlsruhe, presented, which is based on multiple TR interleaved
Germany. in a single steady-state sequence and/or distributed
2
Philips Research Europe, Hamburg, Germany. over successive scans (25). Multiple TR B1/T1 mapping
Additional Supporting Information may be found in the online version of (MTM) allows the simultaneous determination of both
this article.
*Correspondence to: Tobias Voigt, PhD, University of Karlsruhe, Institute of
B1 and T1 maps from the acquired data. Hence, the
Biomedical Engineering, Kaiserstrabe 12, Karlsruhe, Germany. resulting B1 maps are corrected automatically for T1-
E-mail: [email protected] induced errors and vice versa. The focus of this study
Received 8 July 2009; revised 19 October 2009; accepted 20 November is placed on the B1 mapping capabilities of MTM,
2009.
DOI 10.1002/mrm.22333
regarding the simultaneously acquired T1 map as cor-
Published online 17 June 2010 in Wiley Online Library (wileyonlinelibrary.com). rection for the B1 map. The B1 mapping performance
V
C 2010 Wiley-Liss, Inc. 725
726 Voigt et al.

of the new method is compared with standard AFI The corresponding gradient-echo signals Sn can be
mapping in the framework of error propagation theory. derived from the normalized, transversal, steady-state
Furthermore, phantom and in vivo measurements sup- equilibrium magnetization Mt,n directly after the nth RF
porting the new method’s superiority regarding accu- pulse:
racy are presented.
TTE
Sn ¼ e 2 M0 Mt;n ½1
THEORY
with the equilibrium magnetization M0. Subsequently
Signal Equation
applying the Bloch equation for the z-component of the
A steady-state RF-spoiled gradient-echo sequence based magnetization
on N different, arbitrary repetition times TRn, n [ {1. . .N}, Mz ðtÞ ¼ Mz ðt0 Þeðtt0 Þ=T1 þ M0 ð1  eðtt0 Þ=T1 Þ yields an an-
is depicted in Fig. 1a. Apart from the variable delay at alytical expression for Mt,n, i.e., the magnetization
the end of each TR, all sequence parameters are fixed. reached right after the nth pulse

PN h
 QN i
j¼1 cosðNjÞ ðaÞ  1  Emodðjþn2;NÞþ1 k¼jþ1 Emodðkþn2;NÞþ1
Mt;n ¼ sinðaÞ Q ½2
1  cosN ðaÞ  Nj¼1 E j

with En ¼ eTRn =T1 and mod(i,N) the modulus function angle a, the M0 and T* 2 dependence in Eq. 1 can be
with basis N. The analytical expressions [Eqs. 1, 2] are removed via the ratio ri,j of two signals Si and Sj
based on the assumption that there is no remaining  
transverse magnetization at the time of the next pulse. ri;j ¼ Si Sj ¼ Mz;i Mz;j : ½5
RF and gradient spoiling should be used to minimize
contributions of spurious spin echoes to the signal Combining Eqs. 5 and 2 yields
(21,22). X
N
 
In the case of N ¼ 1, Eq. 2 yields the standard steady- cosNk ai;j ri;j TRmodðkþj2;NÞþ1  TRmodðkþi2;NÞþ1 ¼ 0:
state normalized equilibrium equation (26): k¼1
½6
sinðaÞð1  E1 Þ
Mt ¼ : ½3 Here, TRn/T1  1 was assumed, which allows approxi-
ð1  E1 cosðaÞÞ
mating the exponential factors En in Eq. 2 by first-order
In the case of N ¼ 2, Eq. 2 yields the dual TR signal Taylor series. This causes a systematic T1-dependent
equation as used for AFI (20): error present in the analytical approach and thus also in
AFI B1 mapping. Equation 6 is a polynomial of grade N


1  E2;1 þ 1  E1;2 E2;1 cosðaÞ  1 of cos(a) and does not contain any other unknown
Mt1;2 ¼ sinðaÞ : ½4 quantities. In general, if all TRn values in the sequence
ð1  cos2 ðaÞE1 E2 Þ 
N
are different, ratios (Eq. 5) and polynomials (Eq. 6)
2
In general, various ways can be considered to extract can be computed. Each polynomial has N  1 solutions
the desired quantities B1 (assumed to be proportional to for ai,j, which are not all physically meaningful. The case
a) and T1 from the signal equations [Eqs. 1, 2]. In this N ¼ 2 yields one polynomial with one solution, repro-
report, two approaches are discussed: (1) a general ana- ducing the results from Yarnykh (20)
lytical method based on signal ratios, and (2) a numeri- 
cal method fitting measured data to Eqs. 1, and 2. An Rr2;1  1
a2;1  arccos ; ½7
overview of the discussed methods is given in Table 1. R  r2;1

General Analytical Method with R ¼ TR2/TR1. In the case N ¼ 3, the solution of Eq.
This section can be seen as a generalization of the work 6 can also be derived analytically. For the ratios ri,j ¼ Si/
published in Yarnykh (20). To determine the actual flip Sj, the physical solutions are found to be

"
ri;j TRmodðj;NÞþ1  TRmodði;NÞþ1
ai;j  arccos 
2ðri;j TRmodðj1;NÞþ1  TRmodði1;NÞþ1 Þ
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
u
u ðri;j TRmodðj;NÞþ1  TRmodði;NÞþ1 Þ2 ri;j TRmodðjþ1;NÞþ1  TRmodðiþ1;NÞþ1
þt 2  : ½8
4ðri;j TRmodðj1;NÞþ1  TRmodði1;NÞþ1 Þ ri;j TRmodðj1;NÞþ1  TRmodði1;NÞþ1
Multiple TR B1/T1 Mapping 727

Here, the flip angle a can be obtained by fitting the mea-

sured data to the theoretically expected signals (Eqs. 1
and 2). Additionally, this fit also yields the corresponding
T1 map. The mathematical structure of the signals reads

SðTR1 . . . TRn ; TE; T2 ; T1 ; a; M0 Þ

¼ S0 ðT2 ; TE; M0 Þ  Mt ðTR1 . . . TRn ; a; T1 Þ: ½9

Thus, the fitting algorithm requires at least three

images to determine three unknown parameters, S’, T1,
and a. S’ acts as a global scaling and will not be used in
MTM. These three data points can be obtained by using
at least N ¼ 3 different repetition times or by using an N
¼ 2 sequence and repeating the measurement M times,
with varying TR values (see Fig. 1b), yielding K ¼ N  M
data points. A corresponding, second index m is intro-
duced, with m [ {1. . .M} for the repetition times TRnm and
the signals Snm. Thus, also the K ¼ 4 data points obtained
from N ¼ M ¼ 2 are sufficient to determine the three
unknowns, S’, T1, and a. Of course, arbitrary N and M
with K 3 are possible. The fitting is performed compar-
ing the K measured signals Snm to the analytical expres-
sions given in Eqs. 1 and 2. In the next section, it will be
shown that MTM reconstructions, i.e., fitting K 3 data
points, achieve a substantial increase of the reconstruc-
tion accuracy of a. Moreover, the assumption TRn/T1  1
is not required in this approach, and thus the recon-
structed a is free of systematic, T1-induced errors.
In the following, the index m [ {1. . .M}, introduced for
MTM, will be used also in connection with standard and
generalized AFI imaging, with N ¼ 2 and N ¼ 3, respec-
tively. In this case, identical TR sets will be applied M times,
corresponding to averaging generalized AFI results, to facili-
FIG. 1. Sketch of sequence based on N different repetition times tate the comparison of the different considered schemes.
TRn (a). Sketch of M sequences containing N different repetition
times TRnm (b). Each pulse is applied with the same flip angle a. MATERIALS AND METHODS
Implementation
The three solutions (Eq. 8) are different with respect to
systematic errors induced by the relaxation time T1 and The MTM reconstruction of simulated and measured B1
can be used for averaging, lowering mean statistical mapping data was performed using in-house Cþþ soft-
errors. ware based on the Levenberg-Marquardt algorithm (27).
For N > 3, no analytical solution of Eq. 6 can be pro-
vided. In the following experiments and simulations, Simulation
only the case N ¼ 2 (standard AFI) and N ¼ 3 (called
The noise behavior of MTM fitting and the generalized
generalized AFI) are considered and compared with the
analytical approach for N ¼ 2 (standard AFI) and N ¼ 3
numerical fitting approach outlined in the next section.
(generalized AFI) was analyzed by means of simulations.
Since the goal of this paper is a T1-corrected B1 map,
MTM Fitting
this analysis was restricted to B1 performance only.
Alternatively to the analytical method outlined above, a The comparison of AFI and MTM comprises a system-
numerical fitting approach, dubbed MTM, is presented. atic and a statistic error. First, AFI suffers from a

Table 1
Overview of B1 Mapping Techniques Discussed in This Study: ‘‘Standard’’ AFI Described in Yarnykh (20), Generalized AFI Introduced in
This Study, and MTM, Also Introduced in This Study*
B1 mapping method Number of TRs N Number of measurements M Reconstruction technique
(Standard) AFI 2 1 (M > 1 used to denote averaging Analytically via signal ratio
multiple identical measurements)
Generalized AFI 3 1 (M > 1 used to denote averaging Analytically via signal ratio
multiple identical measurements)
MTM 2 1 (N  M 3) Numerically via model fitting
*The cases M ¼ 1, N 3 can be reconstructed alternatively via generalized AFI or MTM.
728 Voigt et al.

systematic error Da due to neglecting T1 effects. In this According to Eq. 14, the best settings for TRnm and
study, Da is defined as the difference between recon- anom will depend on the parameter a, T1 or S’ chosen to
structed flip angle aAFI and the real value areal. be optimized. Since this work is focused on B1 mapping,
the optimal parameters that minimize s2MTM;a have been
Da ¼ areal  aAFI : ½10 chosen. The minimization of s2MTM;T1 shall be investi-
gated in detail in a separate study.
The systematic error due to neglecting T1 effects is not For minimization, an absolute error function,
present in MTM. The second error component is due to
qffiffiffiffiffiffiffiffiffi
statistical noise, present in both techniques, AFI and dabs ¼ ðsa þ DaÞ Tacq ; ½15
MTM. This is described by the propagation of noise in
the input images to the reconstructed map. For the corre-
sponding mapping error, the standard deviation (SD) s and a relative error function,
of the reconstructed flip angle was used as a measure.
As described below, the systematic and statistical errors dabs
drel ¼ ; ½16
of AFI and MTM were compared using optimized sets of areal
scan parameters, i.e., using the combinations of TR and
anom leading to optimal accuracy of the considered were defined, with sa being the corresponding SDs of
approaches. AFI or MTM, respectively. Da is given in Eq. 10 and
The analytical framework of AFI B1 mapping (cf. Eq. thus is only nonzero in the case of AFI. The relative
7) allows for the estimation of the mapping error by the weight of sa and Da in Eqs. 15 and 16 is influenced by
law of error propagation the ratio M0/sin of the input images. For minimizing Eqs.
15 and 16, M0/sin ¼ 200 was chosen in accordance with
XN  2 typical experimental findings. Noise differences caused
@f
s2AFI;a ¼ sin : ½11 by different
P Ptotal acquisition times (proportional to
n¼1
@S n Tacq ¼ N M
TR
n¼1 m¼1 nm ) are compensated in Eqs. 15 and
16 by the factor HTacq. The described techniques allow
The function f is given by Eq. 7, and sin denotes the the determination and comparison of the absolute and
SD of the input signals Sn. The variance var(uAFI,a) of relative mapping errors of AFI and MTM B1 mapping as
the standard AFI (N ¼ 2) reconstruction is obtained from a function of the SD rin of the underlying fast field echo
Eqs. 11 and 7 and reads (FFE) images. The best settings for TRnm and anom for
MTM and AFI have been determined by minimizing the
varðuAFI;a Þ ¼ s2AFI;a error functions in Eqs. 15 and 16, respectively. The mini-
" 2
 #
ð1  R2 Þ S21 þ S22 1 mization was performed using in-house software written
¼ s2in    : ½12 in Cþþ, using the downhill simplex method (29).
S41 ðR  r Þ4 ðrR  12 ÞðR  r Þ2 M
In principle, the parameter space for comparing AFI
and MTM is infinitely large. However, restrictions can
Here, as discussed above, M denotes the number of be imposed for Tacq and for T1. For the sake of a realistic
AFI averages acquired and R ¼ TR2/TR1. A correspond- total acquisition time, an upper limit for Tacq < Tmax ¼ K
ing expression for N ¼ 3 was derived accordingly but  125 ms was chosen. The minimal TRnm allowed for the
omitted for the sake of brevity. simplex method was set to 20 ms. On the other hand,
The error propagation for the MTM fitting approach is only T1 < 1000 ms was regarded since larger T1 values
described in the more general framework of the Cramer- influenced optimization results only marginally, i.e.,
Rao theorem (CRT) (28). To this goal, it is first assumed resulting TR values and flip angles varied less than 5%
0
that the measured signals sk (k [ {1 . . . K}, K ¼ N  M dif- for 1000 ms < T1 < 2000 ms and less than 10% for 2000
0
ferent TRs) can be written as sk ¼ sk[h] þ wk. White gaus- ms < T1 < 3000 ms.
sian noise is denoted by wk; the unknown parameters by To start, standard N ¼ 2 AFI was compared with gen-
u ¼ [a,T1,S0 ] and the signals sk[h] are given according to eralized N ¼ 3 AFI. Two optimal sets of parameters were
Eq. 1. It is further assumed that there is no cross-correla- obtained for each method by minimizing error functions
tion between noise in measurements and that the input dabs (Eq. 15) and drel (Eq. 16), respectively. The first pa-
variance of each measurement s2in is independent of the rameter set, obtained from minimizing dabs, was used to
object parameters h. The Fisher information matrix I(h) compare the quantities
required for CRT then reads (28):
qffiffiffiffiffiffiffiffiffi
s ¼ sa Tacq ½17
1 X K
@sk ½h @sk ½h
½IðhÞu;v ¼ 2 : ½13
sin k¼1 @hu @hv and Da (Eq. 10). The second parameter set, obtained
from minimizing drel, was used to compare
According to CRT, the variance of an algorithm fitting
0
the parameter set h ¼ [a,T1,S ] possesses the lower SNR ¼ areal =s: ½18
bound
Then, standard AFI and MTM were compared analo-

 h i
1 gously. First, this was performed for N ¼ M ¼ 2 for the
var hMTM ;u ¼ s2MTM ;u I ðhÞu;u : ½14
chosen T1 range. A second comparison for N ¼ 2 included
Multiple TR B1/T1 Mapping 729

increasing M, indicating multiple subsequent measure-

ments in MTM and multiple averages in AFI. It was per-
formed for distinct T1 values and repeated for N ¼ 3. All
calculations were based on the assumption that noise on
the input images sin is equal for both methods and thus
cancels out by forming ratios (cf. Eqs. 11 and 13).

Experiments
All experiments were carried out on a 1.5-T Achieva sys-
tem (Philips Healthcare, Best, The Netherlands). The
quadrature body coil was used for RF transmission and a
quadrature head coil for reception. An optimized RF
spoiling scheme according to the approach presented in
Nehrke (21) was applied to each measurement m
separately. FIG. 2. Comparison of standard N ¼ 2 AFI and generalized N ¼ 3
AFI. Using N ¼ 3 different repetition times within one sequence is
advantageous only with regard to lower systematic errors for T1 >
Phantom Study
250 ms.
A phantom consisting of six compartments with varying
contrast agent concentrations resulting in different T1
field mapping was predicted for (TR11/TR21 ¼ TR12/TR22
values was used. To obtain an independent reference to
¼ 40/309 ms) and anom ¼ 58
. All other sequence param-
the MTM results, T1 was additionally measured via the
eters were equal for both methods; 96  96 scan matrix,
‘‘ratios and least squares’’ algorithm (RLSQ), based on a
FOV 210  210 mm2, and 10 slices (4 mm thickness).
combined spin echo (SE)/inversion recovery (IR)
The same total imaging time of 11 min was achieved in
sequence (30), with TRSE ¼ 1300 ms, TRIR ¼ 3500 ms,
both cases. Every sequence was run twice, and corre-
echo time (TE) spacing ¼ 50 ms, and eight echoes. An
sponding reconstructed B1 images were subtracted to
independent B1 reference was obtained using the double-
determine the noise level of each method.
angle method (8), with TR ¼ 5 s and a1,2 ¼ 60
/120
.
As a first experiment, a three-dimensional volume of
the phantom containing 30 slices (5 mm thickness) and Volunteer Study
field of view (FOV) 220  220 mm2 with spatial scan re- Experiments on healthy volunteers were conducted to
solution 96  98 pixels was acquired with TE ¼ 1.72 ms. proof the in vivo feasibility of the MTM approach.
A dual TR sequence N ¼ 2 with M ¼ 3 different TR com- Written consent was obtained from all volunteers. Three-
binations (TR11/TR21 ¼ 50/200 ms, TR12/TR22 ¼ 50/250 dimensional AFI images of the brain were acquired
ms, and TR13/TR23 ¼ 50/300 ms) was employed, with a (coronal scan orientation, flip angle anom ¼ 50
, spatial
nominal flip angle anom ¼ 50
. Resulting imaging times resolution ¼ 1.25  1.25  12 mm3, FOV 160  160 
were 12 min, 15 min, and 17 min, respectively. The B1- 228 mm3). Subsequent dual TR sequences (N ¼ 2)
related scan parameters anom and TRnm were chosen to were applied with M ¼ 3 different TR combinations
be in the optimal AFI range as given in Yarnykh (20). TR11/TR21 ¼ 40/80 ms, TR12/TR22 ¼ 40/200 ms, and
Thus, the data have been reconstructed twice: via averag- TR13/TR23 ¼ 40/400 ms, leading to imaging times of
ing analytical standard AFI reconstructions (Eq. 7) and 4 min, 8 min, and 15 min, respectively. Repetition times
via numerical MTM fitting (Eqs. 1 and 2). and nominal flip angle represent optimal settings with
For the six compartments, the systematic deviation respect to maximum SNR of B1 and T1 mapping
between AFI B1 map and MTM B1 map has been deter- anom =sMTM ;a þ T1 =sMTM ;T1 for given total imaging time
mined as a function of the T1 value measured by MTM and T1 ¼ 700 ms. TRs obtained via CRT have been
and compared with corresponding simulations. More- rounded to yield integer ratios between TR pairs to allow
over, the MTM T1 map was compared with the RLSQ T1 spoiling of transverse magnetization according to Nehrke
map, and the MTM B1 map was compared with the dou- (21). In a postprocessing step, the images were compen-
ble-angle B1 map. sated for translatory and rotational patient motion
Using the same phantom, a second experiment was between subsequent measurements, using a dedicated
performed to analyze the noise sensitivity of the two registration library (31).
methods and to compare the findings with theoretical
predictions. Two dual TR pairs (N ¼ 2, M ¼ 2) were RESULTS
acquired. Simulations were performed employing error
Performance Analysis
propagation theory according to the previous section to
find optimal scanning parameters with respect to lowest Noise analysis of AFI and MTM B1 mapping employing
noise in reconstructed B1 maps. According to the results error propagation theory is shown in Figs. 2–4. First,
of CRT noise minimization and target T1 ¼ 375 ms, a generalized N ¼ 3 AFI was compared with standard N ¼
nominal flip angle anom ¼ 88
and repetition times TR11/ 2 AFI (Fig. 2). As mentioned, the ratios sAFI,N ¼ 3/sAFI,N
TR21 ¼ 40/146 ms and TR12/TR22 ¼ 40/473 ms were cho- ¼ 2 and DaAFI,N ¼ 3/DaAFI,N ¼ 2 are shown for minimized
sen for MTM mapping. Optimal performance for AFI dabs (Eq. 15) and the ratio SNRAFI,N ¼ 3/SNRAFI,N ¼ 2 for
730 Voigt et al.

Table 2
Optimal Parameters Found for MTM Using CRT Error
Propagation Theory*
SNR Standard deviation s
T1 ¼ T1 ¼ T1 ¼ T1 ¼
600 ms 950 ms 600 ms 950 ms
TR11 20 ms 20 ms 20 ms 20 ms
TR21 105 ms 105 ms 55 ms 40 ms
TR12 20 ms 20 ms 20 ms 20 ms
TR22 355 ms 355 ms 405 ms 420 ms
a 118
125
82
83

*All results shown are based on two subsequent dual-TR sequen-
ces M ¼ N ¼ 2 (see Fig. 3). Scan parameters for maximum SNR
and minimal standard deviation in the reconstructed B1 maps,
respectively, are shown for distinct T1 values representing white
FIG. 3. Comparison of N ¼ M ¼ 2 MTM and standard N ¼ 2 AFI. (T1 ¼ 600 ms) and grey (T1 ¼ 950 ms) matter.
The relative SNR increase is found to be between 1.07 and 2 at
high T1 (1 sec) and low T1 (100 ms), respectively. On the other
hand, MTM is able to reduce noise by more than 10% over a
wide range of T1 values. been found (e.g., anom ¼ 83
for T1 ¼ 950 ms). The
obtained optimal repetition times vary only slightly over
minimized drel (Eq. 16). Normalized SNRs and SD are the investigated T1 range. As already observed for the
shown according to definitions (Eqs. 17 and 18). In the optimal flip angles, optimizing SNR and noise variance
parameter range investigated, generalized AFI does not yields different optimal TRnm.
yield improvement over standard AFI with regard to Third, more than M ¼ 2 subsequent measurements
SNR or SD. However, generalized AFI has significantly have been analyzed and compared with averaging AFI
lower systematic T1 errors for T1 > 250 ms. results (Fig. 4). In the case of N ¼ 2, SNRMTM outper-
Second, the comparison of Fig. 2 was repeated for forms SNRAFI for all investigated T1 and M. In the case
standard AFI and a dual TR sequence N ¼ 2 and two of N ¼ 3, SNRMTM is also higher than SNRAFI (except for
subsequent measurements M ¼ 2 (Fig. 3). For T1 400 M ¼ 1 and T1 950 ms) and similar to SNRMTM,N ¼ 2.
ms, SNRMTM is more than 30% higher than SNRAFI, and The maximum SNR gain has been found for M ¼ 4(2)
SD sMTM is more than 25% lower than sAFI. For larger with N ¼ 2(3), which is caused by a tradeoff between too
T1, SNR gain by MTM decreases to roughly 7% at T1 ¼ small SNR for smaller M and too large Tacq for larger M.
1000 ms. The highest systematic error found for AFI is Since MTM requires at least K ¼ N  M ¼ 3, the case N ¼
Da ¼ 3.8
at T1 ¼ 50 ms and decreases with increasing 2 and M ¼ 1 does not yield MTM results.
T1. The resulting optimal sequence parameters are dis-
cussed in Table 2 for distinct T1 values. To obtain maxi-
mum (normalized) SNRMTM, nominal flip angles between Phantom Experiments
115
and 125
seem to be the best choice. For minimal Experimental phantom results are shown in Figs. 5
variance, significantly lower optimal flip angles have and 6. Figure 5a,b shows the measured flip angle maps
of MTM and AFI. Both maps show a similar B1 distribu-
tion. The flip angle difference between the two methods
is clearly T1 dependent and increases with decreasing
T1, as discussed in connection with Fig. 2. Correspond-
ing quantitative results are shown in Fig. 6a. The com-
parison with simulations shows that the differences in
reconstructed flip angles between AFI and MTM are due
to the discussed systematic error in AFI B1 mapping,
which is not present in MTM. No significant systematic
flip angle differences between long TR double angle
method and MTM B1 mapping were found.
Predicted noise ratios via error propagation theory
could be confirmed in the second phantom experiment.
Noise reduction for the given T1 range was predicted
theoretically between 40% and 50% and found experi-
mentally between 45% and 55% (Fig. 7).
The MTM T1 map is shown in Fig. 5c. A quantitative
FIG. 4. Comparing SNR of standard AFI and MTM using N ¼ 2
comparison with the RLSQ measurement is given in Fig.
(solid symbols) and N ¼ 3 (open symbols). Ratios are shown as 6b. MTM T1 mapping is able to reproduce this reference
function of number of TRs K ¼ N  M for three different T1 set- measurement with sufficient accuracy. Fitting a straight
tings representing gray (T1 ¼ 950 ms) and white (T1 ¼ 600 ms) line through the origin, T1,MTM ¼ a  T1,RLSQ, yields an
matter, as well as fat (T1 ¼ 250 ms). inclination of a ¼ 0.96 6 0.02.
Multiple TR B1/T1 Mapping 731

FIG. 5. Flip angle map obtained with (a) MTM and (b) standard AFI reconstruction. Level and window were adjusted to highlight differ-
ences. The largest deviations can be observed in regions with low T1, i.e., the background and the top tube, where AFI noticeably
underestimates the actual flip angle. c: T1 map of the phantom, reconstructed with MTM. Quantitative values for B1 and T1 are shown
in Fig. 6.

In Vivo Experiments cal procedure, this technique yields not only B1 but also
T1.
MTM results of in vivo experiments are shown in Fig. 8.
The investigated extension of the analytical approach
The B1 distribution with high resolution and low noise
from N ¼ 2 to N ¼ 3 (‘‘generalized AFI’’, Fig. 2) exhib-
is shown in Fig. 8a. The T1 distribution is shown in Fig.
ited a significant drop of the systematic B1 error, which
8b, revealing good contrast between gray matter, white
is caused by the assumption T1 TR. However, this
matter, and cerebrospinal fluid (CSF). The quantitative
benefit is counterbalanced by a strong noise increase.
values are in good agreement with literature values (32).
Thus, in the framework of the analytical approach, no
The FFE image Snm for n ¼ 1, m ¼ 2 is shown in Fig. 8c.
clear advantage of N ¼ 3 is visible.
Alternatively, the data points obtained with the dis-
cussed N ¼ 3 sequence can serve as input for the numer-
DISCUSSION
ical model fitting (‘‘MTM’’, Fig. 3). This reconstruction
The B1 reconstruction of the previously published tech- method shows a more benign error propagation and a
nique AFI is based on two data points acquired per higher SNR than standard AFI, at least for a wide range
voxel, corresponding to the two different TRs forming of T1. Using more than one TR triple (i.e., M 2, N ¼ 3),
the AFI sequence. The presented study raises the num- this advantage increases and holds for all investigated
ber of data points per voxel by introducing multiple TR T1. Applying the same acquisition times, the SNR gain
sets and/or multiple sequences with differing TR sets. for N ¼ 3 is similar to N ¼ 2 (keeping in mind that N ¼
From the obtained data points, B1 can be reconstructed 2 is possible only for K 4). Again, a separate study
in two ways. On one hand, an analytical formula can be shall analyze if this observation also holds for higher N.
applied, which is derived from ratios of the obtained sig- In the performance analysis, M0/sin ¼ 200 was used,
nals, in analogy to the standard AFI technique. On the determining the relative weight of systematic and statis-
other hand, the data points can be fitted in the frame- tical error in the AFI optimization. This factor is related
work of a numerical model. In opposition to the analyti- to but not identical with image SNR. To translate the

FIG. 6. Quantitative analysis of phantom experiments. a: T1 dependent difference in reconstructed flip angle between MTM and AFI
(open symbols) and between MTM and the double-angle method (solid symbols). The straight lines, representing the systematic error in
AFI B1 mapping, were obtained from simulation of an AFI sequence with the same flip angle as observed in the experiment and different
repetition times. The curves represent longer (dotted), same (solid), and shorter (dashed) TRnm as used in the experiment. b: T1 values
reconstructed with MTM as a function of T1 values determined independently with a ratios and least squares sequence. Data points for
both plots were obtained by averaging circular areas inside the tubes and on the background.
732 Voigt et al.

largest at T1  150 ms, where, however, no corresponding

tissue types have been reported for main field strength B0
¼ 1.5 T used in the presented experiments (32). Here, it
might be worth mentioning that B0 enters MTM exclu-
sively via its impact on T1 (32), i.e., the presented calcula-
tions hold for all B0 as long as adequate T1 values are con-
sidered. For low-field systems B0 < 0.5 T, also T1 values in
the abovementioned range of 150 ms occur.
Strongly varying B1 fields, as can occur in parallel
transmission (6,7), are frequently problematic for B1
mapping techniques. This aspect has not been investi-
gated in this study. However, it can be circumvented by
the recently introduced ‘‘all-but-one-channel’’ approach
(34,35), yielding B1 maps with significantly reduced
FIG. 7. Comparison of SD of MTM and AFI for M ¼ N ¼ 2. Noise dynamic range.
reduction was predicted theoretically between 40% and 50% for In the experiments conducted in this study, TRnm
flip angles aAFI between 51
and 58
, as observed in the experi- T1 has been used. Considering even smaller TRnm in the
ment. Noise ratios from experiments are found in good agreement case of AFI B1 mapping would lead to lower systematic
to CRT calculations. errors on the one hand and an increase in noise on the
other hand. In the case of MTM, smaller TRnm would
lead to an increase in noise level for both B1 and T1
factor M0/sin into SNR, repetition times, flip angle, and maps. The relative noise increase would depend on the
T1 have to be taken into account. In the experiments con- optimization scheme chosen, i.e., minimizing s2MTM;a or
ducted in this study, the image SNR ranged from 8% to s2MTM;T1 . To take into account this noise dependence on
40% of M0/sin. The obtained results advise that the SNR imaging time, SNR and variance have been normalized
of AFI should not be increased by simply averaging mul- with imaging time in the numerical comparison pre-
tiple measurements, i.e., repeating the AFI sequence sented in this study (Eq. 15).
with identical TR pairs. Instead, available scan time MTM simultaneously provides a T1 map with good ac-
ought to be used for sampling M different TR sets, ena- curacy without requiring additional acquisition time.
bling B1 reconstruction via MTM fitting. For a given N, Thus, MTM surpasses subsequent B1 and T1 measure-
an optimum M can be determined, yielding maximum ment techniques, which are often performed to correct
SNR gain compared with straightforward averaging. In one measurement by use of the other. Furthermore, the
general, this optimum depends on the choice of the max- B1 transmit field determined with MTM is intrinsically
imum summed profile duration Tmax, set to K  125 ms T1 corrected and thus has no systematic error, as found
in this study, yielding M ¼ 4(2) as optimum for N ¼ in AFI. A systematic investigation of MTM’s T1 mapping
2(3). This advantage could also be used to accelerate B1 efficiency and versatility was postponed due to the focus
mapping maintaining SNR. Although presented investiga- of B1 mapping chosen for the present study.
tions employed quadrature coils exclusively, the combina- With regard to incomplete spoiling of remaining trans-
tion of MTM with parallel imaging (see, e.g., Pruessmann verse magnetization, it is assumed that different meas-
et al. (33)) seems to be particularly advantageous. At least urements with different settings for repetition times will
at low acceleration factors A, the SNR loss of parallel imag- result in varying contributions of spurious spin echoes
ing shows the statistical behavior proportional to HA. to the signal (21). A fit algorithm could possibly elimi-
According to the presented study, MTM is able to overcom- nate these effects and therefore could be advantageous
pensate this SNR loss. Of course, imaging time in general over systematic approaches. Thus, MTM could be less
will depend on the readout mode used, e.g., echo-planar prone to systematic errors from spurious transverse mag-
imaging and/or half Fourier techniques. While predicted netization. Regarding AFI, in case of incomplete spoil-
SNR gain is expected for all T1 values investigated, it is ing, Eq. 6 remains valid only if the ratio of the signals Si

FIG. 8. MTM in vivo results. B1 map (a), T1 map (b), and corresponding anatomy (c) are shown. Flip-angle maps with very low noise
level are obtained (a). Relaxation times reveal good gray and white matter contrast (b,c).
Multiple TR B1/T1 Mapping 733

and Sj is preserved. Otherwise, systematic deviations 9. Stollberger R, Wach P. Imaging of the active B1 field in vivo. Magn
will potentially result in biased flip-angle maps. It is Reson Med 1995;35:246–251.
10. Cunningham CH, Pauly JM, Nayak KS. Saturated double-angle method
assumed that a more careful spoiling scheme as opposed for rapid B1þ mapping. Magn Reson Med 2006;55:1326–1333.
to the MTM fitting approach should be used, since opti- 11. Wang J, Qui M, Constable RT. In vivo method for correcting trans-
mal spoiling settings depend heavily on applied object mit/receive nonuniformities with phased array coils. Magn Reson
and scan parameters. Besides remaining transverse mag- Med 2005;53:666–674.
12. Fernández-Seara MA, Song HK, Wehrli FW. Trabecular bone volume
netization, other problems predominantly encountered
fraction mapping by low-resolution MRI. Magn Reson Med 2001;46:
in in vivo imaging have not been addressed in the pres- 103–113.
ent study. Flow could be a source of error in B1 mapping 13. Hornak JP, Szumowski Bryant RG. Magnetic field mapping. Magn
results; particularly, CSF flow might account for altered Reson Med 1988;6:158–163.
B1 visible in Fig. 8. 14. Alecci M, Collins CM, Smith MB, Jezzard P. Radiofrequency mag-
netic field mapping of a 3 tesla birdcage coil: experimental and theo-
In principle, generalizing the AFI approach could also retical dependence on sample properties. Magn Reson Med 2001;46:
involve varying flip angles between different segments. 379–385.
In this case, however, a more sophisticated optimization 15. Jiru F, Klose U. Fast 3D radiofrequency field mapping using echopla-
scheme has to be used. An investigation is therefore nar imaging. Magn Reson Med 2006;56:1375–1379.
16. Carlson JW, Kramer DM. Rapid radiofrequency calibration in MRI.
postponed to a separate study.
Magn Reson Med 1990;15:438–445.
17. Akoka S, Franconi F, Seguin F, Le Pape A. Radiofrequency map of
CONCLUSION an NMR coil by imaging. Magn Reson Imaging 1993;11:437–441.
18. Dowel NG, Tofts PS. Fast, accurate, and precise mapping of the RF
A new simultaneous B1/T1 mapping approach named field in vivo using the 180
signal null. Magn Reson Med 2007;58:
MTM has been introduced. Its B1 mapping performance 622–630.
was thoroughly compared with AFI, a well-established 19. Morrell G. A phase-sensitive method of flip angle mapping. Magn
fast B1 mapping technique. The new approach is able to Reson Med 2008;60:889–894.
20. Yarnykh VL. Actual flip-angle imaging in the pulsed steady state: a
substantially decrease systematic and statistical mapping
method for rapid three-dimensional mapping of the transmitted ra-
errors compared to standard AFI, which is beneficial for diofrequency field. Magn Reson Med 2007;57:192–200.
various applications that rely on accurate and fast B1 21. Nehrke K. On the steady-state properties of actual flip angle imaging
mapping. (AFI). Magn Reson Med 2009;61:84–92.
The applied Cramer Rao Lower Bound estimation 22. Yarnykh VL. Optimal spoiling of the transverse magnetization in the
actual flip-angle imaging (AFI) sequence for fast B1 field mapping.
allows a fast determination of optimal scanning parame- In: Proceedings of the 16th Annual Meeting of ISMRM, Toronto, On-
ters for performing MTM sequences. tario, Canada, 2008. p 3090.
In vivo applications of MTM shown in this study were 23. Treier R, Steingoetter A, Fried M, Schwizer W, Boesinger P. Opti-
limited to imaging the head. In principle, this approach mized and combined T1 and B1 mapping technique for fast and accu-
rate T1 quantification in contrast enhanced abdominal MRI. Magn
is applicable for all anatomies, which will be subject of
Reson Med 2007;57:568–576.
further investigations. 24. Hsu JJ, Zaharchuk G, Glover GH. Rapid methods for concurrent mea-
surement of the RF-pulse flip angle and the longitudinal relaxation
time. Magn Reson Med 2009;61:1319–1325.
ACKNOWLEDGMENTS 25. Voigt T, Katscher U, Nehrke K, Doessel O. Simultaneous B1 and T1
mapping based on modified ‘‘actual flip-angle imaging.’’ In: Proceed-
The authors thank Ewald Rössl for valuable contribu- ings of the 17th Annual Meeting of ISMRM, Honolulu, HI, 2009. p 4543.
tions to the work on Cramer Rao lower bound calcula- 26. Haacke EM, Brown RW, Thompson MR, Venkatesan R. Magnetic res-
tions and Christian Stehning for help on motion onance imaging: physical principles and sequence design. New York:
correction of volunteer data. Wiley-Liss; 1999. p 451–512.
27. Marquardt D. An algorithm for least-squares estimation of nonlinear
parameters. J Soc Indust Appl Math 1963;11:431–441.
REFERENCES 28. Kay SM. Fundamentals of statistical signal processing: estimation
1. Venkatesan R, Lin W, Haacke EM. Accurate determination of spin- theory. Upper Saddle River, New Jersey: Prentice Hall; 1993. p 27–
density and T1 in the presence of RF-field inhomogeneities and flip- 83.
angle miscalibration. Magn Reson Med 1998;40:592–602. 29. Press WH, Teukolsky SA, Vetterling WT, Flannery BP. Downhill sim-
2. Sled JG, Pike GB. Correction for B1 and B0 variations in quantita- plex method in multidimensions. In: Numerical recipes: the art of
tive T2 measurements using MRI. Magn Reson Med 2000;43: scientific computing. 3rd edition. Cambridge University Press; 2007.
589–593. p 408–412.
3. Wang J, Qui M, Kim H, Constable RT. T1 Measurement incorporating 30. In den Kleef JJ, Cuppen JJ. RLSQ: T1, T2 and q calculations, combin-
flip angle calibration and correction in vivo. J Magn Reson 2006;181: ing ratios and least squares. Magn Reson Med 1987;5:513–524.
283–292. 31. Thévenaz P, Ruttimann UE, Unser M. A pyramid approach to sub-
4. Samson RS, Wheeler-Kingshott CAM, Symms MR, Tozer DJ, Tofts pixel registration based on intensity. IEEE Trans Image Process 1998;
PS. A simple correction for B-1 field errors in magnetization transfer 7:27–41.
ratio measurements. Magn Reson Imaging 2006;24:255–263. 32. Bottomley PA, Foster TH, Argersinger RE, Pfeifer LM. A review of
5. Katscher U, Voigt T, Findeklee C, Nehrke K, Weiss S, Doessel O. normal tissue hydrogen NMR relaxation times and relaxation mecha-
Determination of electric conductivity and local SAR via B1 map- nisms from 1–100 MHz: dependence on tissue type, NMR frequency,
ping. IEEE Trans Med Imaging 2009;28:1365–1374. temperature, species, excision, and age. Med Phys 1984;11:425–448.
6. Ibrahim TS, Lee R, Baertlein BA, Kangarlu A, Robitaille PL. Applica- 33. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sen-
tion of finite difference time domain method for the design of bird- sitivity encoding for fast MRI. Magn Reson Med 1999;42:952–962.
cage RF head coils using multi-port excitations. Magn Reson Imaging 34. Nehrke K, Börnert P. Improved B1-mapping for multi RF transmit
2000;18:733–742. systems. In: Proceedings of the 16th Annual Meeting of ISMRM, To-
7. Katscher U, Börnert P, Leussler C, van den Brink JS. Transmit ronto, Canada, 2008. p 353.
SENSE. Magn Reson Med 2003;49:144–150. 35. Brunner DO, Pruessmann KP. A matrix approach for mapping array
8. Insko EK, Bolinger L. Mapping of radiology field. J Magn Reson A transmit fields in under a minute. In: Proceedings of the 16th Annual
1993;103:82–85. Meeting of ISMRM, Toronto, Canada, 2008. p 354.