Tissue Oxygenation Mapping by Combined Chemical Shift and T 1 Magnetic Resonance Imaging
Florence Franconi,
1,2* Laurent Lemaire,
1,2Herv e Saint-Jalmes,
1,3,4,5and Patrick Saulnier
2Purpose:To propose a method for determining tissue oxygen- ation via the measurement of fat T1. The method is based on a 2D fat/water chemical shift-encoded and T1-weighted acquisition.
Theory and Methods:A 2D data set was acquired with a fast spin echo sequence with several echo asymmetries and repe- tition times, wherein one dimension is related to the fat/water phase modulation and the other to the T1saturation recovery.
A joint magnitude-based process of phase modulation and T1
evolution allowed for the collection of the fat fraction and T1
maps with resolved fat or water dominance ambiguity while avoiding the phased error problem.
Results:In vitro imaging allowed for the attribution of fat con- tent for different water/oil emulsions that demonstrated longitu- dinal relaxation rate (R1) sensitivity to the oxygenated emulsion environment. The fat R1values were subsequently compared to reference values, which were measured using low receiver bandwidth acquisition to enhance water and fat signal separa- tions. In vivo feasibility of tissue oxygenation assessment was demonstrated by investigating interscapular brown adipose tis- sue modifications during an air/carbogen challenge in rats.
Conclusion: The proposed method offers a precise and robust estimate of tissue oxygenation illustrated by the meth- od’s ability to detect-brown adipose tissue oxygenation modifi- cations. Magn Reson Med 000:000–000, 2017. VC 2017 International Society for Magnetic Resonance in Medicine.
Key words:magnetic resonance imaging; chemical shift imag- ing; longitudinal relaxation time; fat fraction; oxygenation;
brown adipose tissue
INTRODUCTION
Oxygen supply is a crucial parameter to maintain cellu- lar integrity and function. Tissue oxygenation has a key role in a wide range of pathologies. In oncology tumor progression, therapy resistance and even patient progno- sis were shown to be associated with hypoxia (1–3). In
obesity-related metabolic diseases, hypoxia of adipose tissue has been suggested to be an underlying cause of diabetes, cardiovascular disease, or fatty liver disease (4).
There is, therefore, an obvious clinical need for a tech- nique capable of non-invasive imaging of tissue oxygena- tion for the improvement of prognostic and therapeutic monitoring to facilitate personalized medicine. Positon emission tomography (PET) imaging is the most widely established method in this regard, but MRI can be an attractive alternative because of its lack of ionizing radia- tion, its potential high spatial resolution, and its cost- effectiveness (5).
There is a wide range of MR-based oximetry methods to estimate tissue oxygenation. Combined with the hyper- oxic gas challenge, this allows for the in vivo mapping of hypoxia. Indirect oxygenation measurements such as the blood oxygen level-dependent (BOLD) method estimates the oxygen saturation (SO2) from the local transverse relaxation rate (R2) (6). Direct assessment of tissue oxygen pressure (pO2) is possible and circumvents the intrinsic sensitivity of the BOLD method to total hemoglobin con- tent. These methods are based on the paramagnetic effect of oxygen on the R1of a reporter probe. Exogenous19F or
1H probes, administered intravenously or locally, have demonstrated their ability to characterize tissue oxygena- tion (7–11). However, to reduce the invasiveness and potential toxicity of the injected probe, an endogenous source of contrast has also been explored. This method is based on the tissue R1sensitivity to dissolved oxygen in biological fluids (12–14). The sensitivity of the method has been further improved by focusing on lipid signals to assess oxygen status, as oxygen solubility is much greater in lipids than in water (15). This method, called mapping of oxygen by imaging lipids relaxation enhancement method (MOBILE) (16,17), selectively generates R1maps of fat using spectrally selective water suppression. Unfor- tunately, uniform and reliable water suppression may be challenging in areas of magnetic or radiofrequency field inhomogeneities.
Other strategies to select lipid signal exist, particularly, the method called iterative decomposition of water and fat with echo asymmetry and least squares estimation (IDEAL) (18). This method can accurately separate water and fat signals and has reduced sensitivity to magnetic field inhomogeneities effects by exploiting the difference in the chemical shifts between water and fat. However, to the best of our knowledge, the IDEAL method has never been applied to oximetry. Pure fat T1estimation has been previously generated with the IDEAL method but only for use in tissue characterization or T1-bias correction (19–21). Furthermore, fat T1estimates were generated in a
1PRISM Plate-forme de recherche en imagerie et spectroscopie multi- modales, PRISM-Icat, Angers et PRISM-Biosit CNRS UMS 3480, INSERM UMS 018, Rennes, UBL Universite Bretagne, Loire, France.
2Micro & Nanomedecines Translationelles-MINT, UNIV Angers, INSERM U1066, CNRS UMR 6021, UBL Universite Bretagne Loire, Angers, France.
3INSERM, UMR 1099, Rennes, France.
4LTSI, Universite de Rennes 1, Rennes, France.
5CRLCC, Centre Euge`ne Marquis, Rennes, France.
*Correspondence to: Florence Franconi, PhD, PRISM Plate-forme de recherche en imagerie et spectroscopie multi-modales, IRIS-IBS, Institut de Biologie en Sante, CHU Angers, 4 rue Larrey, 49933 Angers Cedex 9.
E-mail: florence.franconi@univ-angers.fr.
Received 20 March 2017; revised 22 June 2017; accepted 7 July 2017 DOI 10.1002/mrm.26857
Published online 00 Month 2017 in Wiley Online Library (wileyonlinelibrary.
com).
Magnetic Resonance in Medicine 00:00–00 (2017)
VC2017 International Society for Magnetic Resonance in Medicine 1
two-step manner: first, fat and water signals were sepa- rated accordingly to the IDEAL method, and second, each set of fat and water images were processed separately to calculate the corresponding T1values. A one-step fat T1
method for the diagnosis of pathologic bone marrow has been recently proposed (22).
We hereby submit a one-step method to map fat T1for the estimation of tissue oxygenation. The method is based on a 2D fat/water chemical shift-encoded T1- weighted data set acquired by varying the echo asymme- try and repetition times of a fast spin echo sequence.
Three features of the proposed method are noteworthy:
The optimized T1recovery curve sampling scheme, The simultaneous processing of the chemical shift-
induced phase modulation and T1evolution, and last, The magnitude-based processing to avoid any sensi-
tivity to phase errors in the source images while still resolving the ambiguity of fat or water dominance.
The method was validated in vitro using oil/water emulsions equilibrated with gas to represent a different oxygen level. Subsequently, fat R1values were compared to a reference-value obtained with a small bandwidth receiver acquisition to enhance water and fat signal sepa- rations. In vivo feasibility of tissue oxygenation assess- ment was also demonstrated by reporting interscapular
brown adipose tissue modifications during an air/carbo- gen challenge in rats.
THEORY
A multi-repetition time, multi-echo-shift, fast spin echo sequence (Fig. 1) was used to sample both T1 recovery and phase modulation induced by the chemical shift dif- ference. The signal model for a fast spin echo experiment with N-shifted echoes acquired at {TEn}¼TE1,. . .,TEN and M repetition times {TRm}¼TR1,. . .TRM including magnetic field inhomogeneity (in the absence of phase errors or noise) can be expressed as:
Sn;mðSo;FF;R1;W;R1;F;fF;cÞ ¼So½
ð1FFÞ
:
1expðR1;W:TRmÞ þFF:
1expðR1;F:TRmÞ
:expði2p:fF:TEnÞ
expði2pc:TEnÞ
for n¼1;. . .;N and m¼1;. . .;M;
[1]
where Sn;mðSo;FF;R1;W;R1;F;fF;cÞis the signal intensity, So includes the initial magnetization and the receiver FIG. 1. The pulse sequence diagram of the standard (a) and the asymmetric (b) fast spin echo sequence (the slice selection and the phase gradients were omitted). The dotted midline between the two successivepradio frequency (R.F.) pulses indicates where the cen- ter of the Hahn spin echo occurs. Negative or positive echo shifts can be created to induce a phase modulation of fat and water signals by shifting the readout frequency gradient. Six different echo time shifts TE were used to sample phase modulation corresponding to six relative fat-water phase shiftsf¼5p/6,p/2,p/6,p/6,p/2, and 5p/6 pointed as black circles in the geometric representation.fF
correspond to the fat chemical shift. Twelve repetition times TR were used to sample T1saturation recovery following a power-law spac- ing distribution.
gain, and FF represents the fat fraction. R1;W and R1;F
are the longitudinal relaxation rates of water and fat, respectively. c is the frequency shift because of local magnetic field inhomogeneity. A compromise had to be struck between an extensive model accurately probing the underlying physics and some simplification to improve the robustness and stability of the estimation.
Therefore, this model includes the following simplifying
approximations: spectral complexity was limited to a single peak for fat, fat chemical shift fF was fixed, and no local transverse relaxation rate R2 correction was applied.
Magnitude fitting was preferred to complex fitting because it greatly simplified the estimation by removing the effects of field inhomogeneity. The signal model fur- ther simplifies to
Sn;mðSo;FF;R1;W;R1;FÞ ¼So:
ð1FFÞ:
1expðR1;W:TRmÞ2
þ FF:
1expðR1;F:TRmÞ2
þ2:cosð2:p:fF:TEnÞ
:
ð1FFÞ:
1expðR1;W:TRmÞ
:
FF:
1expðR1;F:TRmÞ for n¼1;. . .;N and m¼1; . . .; M:
[2]
The consequential confounding fat/water dominance ambiguity, associated to magnitude fitting, can therefore be solved by the joint T1determination.
The measured signal is influenced by phase errors (e.g., as a result of eddy currents) at each echo, wn, and complex gaussian noise hn. The phase errors occur mainly in the first echo, i.e., wn¼0 for n>1. So, if u¼ fSo;FF;R1;W;R1;Fg, the measured signal can be expressed as
Sn;m;meas¼Sn;mðuÞ:eiwn;þhn; for n¼1;. . .;N and
m¼1; . . .; M: [3]
Fat and water T1maps, and consequently the fat frac- tion map, were obtained by estimating ðu^Þ that least- square best fits the acquired signal magnitude
u^¼arg min
u
XN
n¼1
XM
m¼1
jSn;mðuÞj jSn;m;measj2
: [4]
To estimate the unknown parameters u^, a number of images, larger than or equal to the number of unknown parameters, have to be acquired, therefore the greater the number of images the better the precision of the estima- tion. In both directions of the 2D data set (i.e., phase modulation and T1 saturation recovery), imaging sam- pling schemes have been defined with the aim of opti- mizing the measurement accuracy.
Along the phase modulation axis, sampling was based on literature recommendation. Pineda et al. (23) has shown that the number and the value of echo time shift influences the quality of fat and water magnitude estima- tion. The recommended pattern, of at least three equally 2p/3 phase shift(f) spaced echo times was then used as a starting point and condensed to six relative fat-water phase shifts (5p/6, p/2, p/6, p/6, p/2, 5p/6) to improve the quality of the estimation (Fig. 1). Corre- sponding echo shift delays were calculated asf=ð2p:fFÞ.
Along the longitudinal relaxation axis, the sampling strategy (number, range, and distribution of TR) was optimized according to the Cramer-Rao Lower Bound theory (24,25) to minimize the variance of the target
water and fat molecular species T1unbiased estimations.
Sampling times were taken as power-law spacing (26) according to (Fig. 1)
TRm¼TRminþ ðTRmaxTRminÞ m1 M1 r
;
m¼1; 2; . . .; M;
[5]
with r being the power-law factor. Briefly, the Cramer- Rao lower bound states that the minimum variancesðuiÞ of a parameter ui of an unbiased estimator is equal or greater than to the iith element of the inverse of the Fisher information matrix. It corresponds to the best achievable precision of parameters estimates
sðuiÞ ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðF1Þii q
: [6]
The ijthelement of the Fisher information matrix being defined by
Fij¼ 1 sG2
X
n
dSðuÞ dui :dSðuÞ
duj
; [7]
where sG is the standard deviation of zero-mean Gauss- ian noise, S the function dependent on the set of param- eters u and the sum is over all sampling points, n. The indices i and j run over all parameters of S. In the present case, u1 and u2 correspond respectively to So and T1.
The coefficient of variationsT1=T1 for a single molec- ular species can therefore be expressed as (26–28)
sT1
T1
2
¼ 1 SNR2: a
acb2¼ 1
SNR2:g2ðt1;t2;. . .;tMÞ; [8]
in which tm¼TRm=T1, a¼XM
m¼1ð1etmÞ2, b¼ XM
m¼1
tm:ð1etmÞ:etm
, c¼XM
m¼1tm2:e2tm and SNR is the signal-to-noise ratio. Assuming a signal-to-noise ratio that is sufficiently high, the optimal design is there- fore obtained for {TRm}, corresponding to the smaller G function
G¼g TR1
T1FAT; TR2
T1FAT;. . .; TRM
T1FAT
þg TR1
T1WATER
; TR2
T1WATER
;. . .; TRM
T1WATER
: [9]
In the context of proof-of-concept validation, the meth- od’s accuracy was favored, and the acquisition time of {TRm} was taken as 1 h. The corresponding in vitro and in vivo optimum {TRm} patterns have 12 sampling points with TRmin/max¼100 ms/10 s and a power-law factor of r¼2.
METHODS
Phantom Preparation
For the validation of the T1 recovery curve sampling optimization, a phantom composed of two compartments was used; one filled with water and the other with sun- flower oil.
For the rest of the in vitro experiments, oil–water emulsions were prepared based on the protocol adapted from the literature (29) with isopropyl myristate oil (CH3(CH2)12COOCH(CH3)2) and 4% w/w surfactant mix- ture (43% Span 60 and 57% Brij35), all compounds being provided by Sigma-Aldrich (Saint-Quentin Fallav- ier, France). To improve the constituents dispersion, Span 60 was dispersed in the oil phase and Brij35 in the water phase, with both being heated to 85C before being mixed. The system was then allowed to cool to 35C under mechanical stirring before being reheated to 85C.
After cooling to 45C, the mixture was transferred to a 50-mL Falcon tube for 4 min vigorous stirring using Ultraturrax 125 basic instrument (IKA, Werke Staufen, Germany) with a horn dimension of 8 mm100 mm and a speed of 22,000 rpm. Five emulsions, with variable oil content, were generated by varying the mass of fat mf and the mass of watermw. For emulsions A to E,mf=mw
were 4.0 g/15.6 g, 6.9 g/15.6 g, 7.9 g/11.7 g, 9.6 g/9.4 g, and 12.0 g/7.8 g, respectively. Fat mass fractionshM ¼mf=ðmf
þmwÞ were 20.4%, 30.8%, 40.2%, 50.6%, and 60.6%, respectively. The corresponding fat signal fraction was defined according to literature (30) as h¼SF=ðSFþSWÞ, SF and SW being, respectively, the signal from fat and water within a voxel. This signal is proportional to the proton mole number n¼ ðl:mÞ=MW with m the mass,l the number of protons per molecule, and MW the molec- ular weight. The fat signal fraction is therefore h¼nf= ðnf þnwÞwith MWW¼18.015 g/mol andlW¼2 for water and MWF¼270.45 g/mol and lF¼34 for the used fat, isopropyl myristate. As a result, theoretical fat signal fractions of the emulsions were 22.5%, 33.6%, 43.2%, 53.7%, and 63.5%, respectively (it should be noted that surfactants were not taken into account in the calcula- tion as a result of their low concentrations). Each emul- sion was equilibrated for 30 min with different gas, either nitrogen (0% O2), air (21% O2), oxygen (100%
O2), or a mixture of nitrogen and O2to reach respectively 10% or 40% O2. To better the equilibration, the emulsion was gently agitated using an oscillating plate and the gas phase ( 30 mL) present atop the emulsions was flushed every 5 min with the appropriate gas. Approximately 4 mL
of the emulsion was then transferred in a 1-cm diameter tube previously flushed with the adequate gas, and the tube was sealed.
MRI
7T 20 cm horizontal preclinical MRI scanner (70/20 Bio- spec, Bruker, Wissembourg, France), equipped with a 120-mm diameter gradient set (675 mT/m), was used to acquire spectra and images using ParaVision 6.0.1. For the in vitro T1 recovery curve sampling optimization, a 35 mm transmitter/receiver volume coil was used. The remaining in vitro experimentation was carried out using a 72 mm transmitter/receiver volume coil, while a 86 mm transmitter volume coil, with a 20 mm surface coil, was used for in vivo imaging. In vitro experiment were car- ried out at room temperature (21C).
T1Recovery Curve Sampling Optimization Simulation
The G function was calculated using MATLAB (Math- Works, Natick, MA) software for varying numbers of TR (M¼3, 6, and 12), different TRmax(2–10 s), and different distributions (power-law factor r¼1 to 6) for two esti- mated T1¼0.55 s (fat) and 2.9 s (water) and for TRmin¼100 ms.
In Vitro Measurement
Fast spin echo acquisitions with variable repetition times were carried out with the following parameters: one axial slice of 2 mm, field of view of 1515 mm, matrix of 6464, effective echo time¼11 ms, an echo train length of 4, a receiver bandwidth of 100 kHz, and one average.
TR sampling was varied accordingly to the simulation (54 acquisitions were carried out, each with a different TR scheme, calculated from Eq. [5] with the number of TR M¼3, 6, or 12, TRmin¼100 ms, TRmax¼2.5, 5, or 10s and the power-law factor r¼1, 2, 3, 4, 5, or 6). Fat and water T1and the standard deviations were estimated and the coefficients of variation (the ratio of standard devia- tion to the T1value) of the two molecular species were added.
In Vitro T1Measurements of Fat and Water
Before imaging, a localized 1H-spectrum from air equili- brated emulsion B was acquired with a point resolved spec- troscopy (PRESS) (31) sequence with an echo time¼16.5 ms, TR¼10 s, and a voxel size¼6 mm6 mm5 mm without water suppression. The combined T1-phase modu- lated and small receiver bandwidth data set was acquired to calculate the fat and water T1 maps. R1relaxation rates were calculated from T1as R1¼1/T1.
Combined T1-Phase Modulation MRI
The 2D water/fat chemical shift-encoded T1-weighted data set was obtained using a modified fast spin echo sequence to shift the readout gradient, with respect to the spin echo, to induce the phase modulation of fat and water signals. One 3-mm slice was acquired with a field of view of 46 mm12–46 mm and matrix of 12840–64
adapted to the number of tubes imaged at the same time, effective echo time¼5.4 ms, an echo train length of 8, a receiver bandwidth of 100 kHz, and one average. Several acquisitions were carried out with six different echo shifts of0.4, 0.24, 0.08, 0.08, 0.24, and 0.4 ms and with 12 different repetition times (power-law pattern with r¼2, TRmin¼100 ms and TRmax¼10 s).
Fat fraction, fat and water T1maps were obtained on a pixel-by-pixel basis using signal intensity threshold and lsqcurvfit procedure using MATLAB. Initial guess/lower bound/upper bound foru^were taken as fSo¼50=1=200;
FF¼0:7=0=1;R1;W¼0:3=0:2=0:7;R1;F¼2=0:7=20g.
T1Measurements by Small Receiver Bandwidth MRI To validate the combined chemical shift-T1method, ref- erence values of fat and water T1 were necessary.
Inspired from the T2 method from Salvati et al. (32), spin echo images with variable repetition time were acquired with a low receiver bandwidth to separate the water and fat signals along the frequency encoding direc- tion. Slice geometry was identical to those used for the combined T1-phase modulation MRI. Echo time¼12 ms, the receiver bandwidth of 20 kHz, one average, and 12 TR (power-law pattern with r¼2, TRmin¼100 ms and TRmax¼10 s) were used. T1maps were calculated from a mono-exponential fit of this data.
In Vivo Air/Carbogen Challenge
In vivo imaging was carried out on three rats (female Sprague-Dawley, 8 weeks old, Angers University/Hospital Animal Facility) anaesthetized with isoflurane, with the surface coil positioned on the interscapular area. The animal study protocol was approved by the local institu- tional animal care and use committee. Two series of 2D T1-phase modulation data sets were obtained, one under air inhalation and one under carbogen (95% O2 - 5%
CO2) inhalation, with a delay of 30 min of stabilization after gas switching. A fast spin echo sequence, modified to shift the readout gradient (with respect to the Hahn spin echo) to induce phase modulation of fat and water signals, was used. One sagittal slice of 2.5 mm was acquired with a field of view of 2040 mm, a matrix of 12896 zerofilled to 256192, echo time¼7.6 ms, an echo train length of 8, a receiver bandwidth of 25 kHz, and one average. Several acquisitions were carried out
using six different echo shifts of 0.4, 0.24, 0.08, 0.08, 0.24, and 0.4 ms and 12 different repetition times (power-law pattern with r¼2, TRmin¼100 ms, and TRmax¼10 s).
Fat fraction and fat T1maps were obtained on a pixel- by-pixel basis using lsqcurvfit procedure via MATLAB (MathWorks, Natick, MA). Initial guess/lower bound/
upper bound for u^were taken as ¼ fSo¼80=1=200; FF
¼0:8=0=1;R1;W ¼0:3=0:2=0:7;R1;F¼2:5=1=20g. All calcu- lated images were thresholded with a mask to eliminate meaningless pixels. The mask was defined to remove pixels whose signal was inferior to a threshold of 41 on the anatomical image, or with fat R1 values equal to 1.
Fat T1map variations, with the hyperoxic gas challenge, correspond to the oxygenation level variation map. The same data set was also processed with a reduced number of TR sampling values (keeping only the first five and the ninth values) corresponding to an acquisition time of
<10 min.
RESULTS
T1Recovery Curve Sampling Optimization
Figure 2a shows the simulated variation of the T1estimation error (G) for a fat and water system, as a function of recov- ery curve sampling parameters. G function decreases in response to number of TR and r, and the increase in TRmax. The plot indicates that the use of a high number of sam- pling values (M¼12), with a TR max of 10 s, and a power law coefficient value of 2 is the most favorable parameters.
Therefore, these parameters were selected for the combined chemical shift-T1 and small receiver bandwidth measure- ments both in vitro and in vivo. To validate the simulation, the experimental T1coefficients of variation were measured on the water and oil phantom (Fig. 2b).
In Vitro T1Measurements of Fat and Water
The resonance assignments, chemical shifts, and relative intensities from isopropyl myristate oil, as observed in the air equilibrated B-emulsion1H-spectrum, are summa- rized in Table 1.
Combined T1-Phase Modulation MRI
Fit examples of the 2D chemical shift T1-weighted data set is illustrated in Figure 3, representing two different FIG. 2. T1 sampling strategy
optimization, simulation, and experimental validation: calcu- lated G function variation (a) and experimental T1 variation coeffi- cients (b) for differing power law coefficient r, differing numbers of TR (M), and differing maximum TRs.
pixels located in two different emulsions with different fat fractions. The two dimensions were processed jointly to exploit the covariance with the effect of increasing the probability of successfully reducing the fat/water ambi- guity. Examples of fat fraction, and fat and water T1
maps, are shown in Figure 4. Whatever the gas environ- ment or the fat concentration, fat signal fractions, mea- sured with the chemical shift T1 MRI, were similar to theoretical calculated values as shown in Figure 5. The mean fat fraction coefficient of variation was close to 0.7%. Chemical shift T1 MRI R1 values show a linear evolution of R1 with pO2, therefore demonstrating the sensitivity of R1to oxygenation whatever the fat concen- tration in the emulsions (Fig. 6). The mean fat R1coeffi- cient of variation was<1.3%.
T1Measurements by Small Receiver Bandwidth MRI Direct fat and water T1 reference values were obtained by separating water and lipid signals along the direction of the frequency encoding using a small receiver band- width acquisition. An example of resultant T1 map is shown in Figure 7 for the air equilibrated emulsions.
Water and fat chemical-shift T1 R1 values (Supporting Table S1) are shown in Figure 8 to be in agreement with the small receiver bandwidth reference values for the five different emulsions and the five gases used (the identity curve coefficient of determination R2is 0.994).
In Vivo Air/Carbogen Challenge
In vivo applicability of the proposed combined chemical-shift T1oximetry method was demonstrated on rat adipose tissue (n¼3). Brown adipose tissue (BAT) is encountered in the interscapular area, as shown in the anatomical images in Figures 9a and b. The fat signal fraction (Fig. 9c) and the air and carbogen fat T1 maps (Figs. 9d and e) were calculated from the MRI acquisi- tions. Mean BAT fat signal fraction was found to be
6565%. The air/carbogen challenge induces a variation of fat T1in BAT deposits, as shown in red in Figures 9f and g, therefore indicating that BAT deposits are reactive to the gas challenge. Figure 9h is similar to Figure 9g but in 9h processing was carried out with a reduced number of TR sampling values.
DISCUSSION
In this work, insofar as tissue oxygenation measurement is concerned, we have demonstrated that accurate fat frac- tion and T1quantification can be obtained by a 2D water/
fat chemical shift encoded-saturation recovery method, using a multi-repetition-time multi-shifted-echo fast spin echo sequence. A fast spin echo sequence was preferred to a gradient echo (33) scheme as it offers many advantages at a high magnetic field. Gradient echo sequences have intrinsically limited spacing between consecutives echoes and are restricted to positive echo shifts referenced to zero echo time. Furthermore, at a high magnetic field the fat/
water phase modulation cycle is very short, taking only 0.95 ms with the 7T. Therefore, for the gradient echo sequence, the recommended 2p/3 spaced echo time shifts must be sampled over several phase modulation cycles, with differing T2weighting induced from one echo to the next. Contrarily, asymmetric fast spin echo allows for both negative and positive echo shifts referenced to the spin echo time. As a result, asymmetric fast spin echo offers the opportunity to use very small echo shifts, limiting the influence of T2 therefore making R2 correction superflu- ous. Additionally, as a result of its reduced sensitivity to magnetic field inhomogeneities, the fast spin echo sequence provides a better image quality.
A saturation recovery protocol was chosen to map T1 because of its compatibility with the chosen phase mod- ulation pattern and its applicability at high magnetic field. In regard to oxygen status evaluation, the method needed to detect small T1 variations (lower than 10%) Table 1
Chemicals Shifts and Relative Abundance Areas of Isopropyl Myristate Oil.
Structure Chemical group Chemical shift (ppm) Relative amplitude (%)
Methyl protons –CH3 0.8 9
Methylene protons –(CH2)n– 1.2 82
Methylene protonsato COO –CH2–COO 2.1 6
Methine protons –CH 3.5 3
FIG. 3. Example of least-square fit for combined chemical shift T1
data set (experimental data in red and fitted data in blue) for air equilibrated emulsion with 22.5%
(a) and 63.5% fat signal fractions (b).
therefore reinforcing the necessity of accurate T1 meas- urements. Therefore, a dense sampling in both directions of the data set plane was applied, leading to approxi- mately a 1 hr acquisition time for both in vitro and in vivo protocols. However, for the in vivo experiments, oxygenation mapping was still possible using a reduced number of TR sampling values data set process, corre- sponding to an acquisition of<10 min.
Several signal model simplifications must be granted, and the impact of these assumptions on the protocol per- formance have been published by Hernando et al. (34), which concluded on the importance of T2 modeling (35,36) and the superiority of multipeak over single peak fat modeling (37,38) as well as the advantage of complex over magnitude fitting (39–41). In the present work, our choices differ from the aforementioned recommendations FIG. 4. Example of combined chemical
shift T1 data for air equilibrated emul- sions A–E with fat signal fractions of 22.5%, 33.6%, 43.2%, 53.7%, and 63.5%, respectively. (a) Source image (TR¼10 s, echo shift 0.08 ms), (b) cal- culated fat fraction map (%), (c) calcu- lated fat T1map (s), and (d) calculated water T1map (s). A landmark tube was positioned to facilitate unambiguous tube assignments.
FIG. 5. Fat signal fraction (%) for different gas environments (with the hatched bar representing 0% O2, the black bar representing 10% O2, the white bar representing 21% O2, the gray bar repre- senting 40% O2, and the dotted bar representing 100% O2) and different fat fraction emulsions (A to E with fat signal fractions of 22.5%, 33.6%, 43.2%, 53.7%, and 63.5%, respectively), deter- mined by combined chemical shift T1MRI. Standard deviations on three values are plotted. The theoretical fat signal fraction values are represented by the red dash.
FIG. 6. Fat and water R1(s1) for different gas equilibrated (0%, 10%, 21%, 40%, and 100% O2) and different fat concentration emulsions (blue diamond¼emulsion A, red square¼emulsion B, green triangle¼emulsion C, orange circle¼emulsion D, and blue circle¼emulsion E with fat signal fractions for emulsions A to E of 22.5%, 33.6%, 43.2%, 53.7%, and 63.5%, respectively) deter- mined by combined chemical shift T1MRI. Standard deviations on three values are plotted.
to enable a compromise between the estimation accu- racy, fit robustness, and the absence of bias. Fat spec- trum complexity was limited to single peak models with a unique T1(even if it has been indicated previously that T1 values differ from one peak to another [42]) to sim- plify the calculation and increase fit stability. The T2 correction was omitted because of its meager impact when linked to the use of shifted fast spin echo, instead of gradient echo. The source image format was used as magnitude, which raised two interesting points; on the one hand, fitting the signal magnitude highly simplifies the calculation by eliminating the challenging field
inhomogeneity map estimation necessary to address phase errors (43), but on the other hand, magnitude fit- ting cannot resolve fat/water ambiguity. This dilemma was solved by new hybrid methods, mixing magnitude, and complex fitting (44,45). In the present work, the method, while based on magnitude fitting, still has the advantage of unambiguously separating water and fat sig- nals. This chosen multivariate approach exploits the covariance by fitting chemical shift and T1 data jointly.
Where other chemical shift and T1 coupled measure- ments have been previously proposed, they have been based mainly on a two-step manner (19–21) wherein each data set was treated independently. The ability to distinguish both molecular species (fat and water), on the basis of both their frequency and their T1in a one- step manner, was demonstrated recently by Le Ster et al.
(22). The method coupled fat fraction, T1, and T2 estima- tions of the bone marrow using breath-hold gradient echo acquired at two different flip angles in <40 s.
Unfortunately, this elegant method could be difficult to adapt at high field to measure oxygenation as the method is based on a gradient echo acquisition pattern that presents previously exposed drawbacks at high field.
Moreover the dual flip angle method to sample T1, although achieving a good coverage in a short amount of time, is sensitive to B1 field inhomogeneities. At high field, precise T1 determination over a large T1 range could be challenging (39).
Within the panorama of available fat T1 mapping methods to measure oxygenation in vivo, the proposed method offers an attractive alternative. A small receiver bandwidth method cannot be used in a complex system, such as an animal or a human, as it requires no adjacent tissue. The MOBILE method has proved its applicability on both animals (46,47) and humans (17,48). However, accurate measurements can be challenging in some ana- tomical areas because of the susceptibility effects, and the potentially delicate selective frequency water signal saturation. The dual flip angle gradient echo method proposed by Le Ster et al. (22), even if not targeted to oximetry, could be an alternative, however, it is difficult to transpose at high field. Furthermore, T1 estimates were only calculated from a mean region of interest sig- nal and not as a T1map with a variation coefficient for the in vivo repeatability experiment up to 25%. There- fore, its precision would have to be improved to enable the mapping of small T1 variations induced by a gas challenge. Our proposed method, while less sensitive to susceptibility effects, demonstrates its potential for oxim- etry but at the expense of a longer acquisition time.
BAT oxygenation modifications, in a gas challenge context, were carried out. BAT is a highly vascularized fatty tissue with the capacity to oxidize fatty acids to produce heat and represents an attractive therapeutic tar- get for obesity and diabetes (49). O2supply plays a key role in the metabolic demands of thermogenesis, there- fore hypoxia has been identified as a potential driver of the metabolic syndrome responsible for proinflammatory phenotypes (4,50). Determining adipose tissue oxygena- tion is, unfortunately, challenging owing to the difficulty to detect and characterize this tissue because of its spar- sity and a lack of a direct oxygenation measuring FIG. 7. T1 map obtained from small receiver bandwidth MRI for
the air equilibrated emulsions A to E with 22.5%, 33.6%, 43.2%, 53.7%, and 63.5% fat fraction, respectively. Fat and water T1
were measured directly in the areas where signals are separated along the frequency encoded direction. A landmark tube was positioned to facilitate unambiguous tube assignments.
FIG. 8. Comparison of fat and water R1(s1) for the different gas equilibrated (purple¼0%, blue¼10%, green¼21%, orange-
¼40%, and red¼100% O2) and different fat concentration emul- sions (A, circle; B, diamond; C, square; D, triangle; and E, cross) with fat signal fractions of 22.5%, 33.6%, 43.2%, 53.7%, and 63.5%, respectively, determined by the chemical shift T1method to the small receiver bandwidth MRI reference values. The dashed line represents the identity line.
methods. Indirect detection of BAT activity has been demonstrated in rodents by perfusion MRI (51), hyperpo- larized xenon gas MRI (52), and 18FDG-PET (positron emission tomography with fluorodeoxyglucose tracer) glucose uptake (52), and by the MRI blood oxygen level- dependent method (BOLD) (53). Hereby, we propose a method that was able to display BAT areas of modified oxygen level induced by a gas challenge. The plausibility of the method was demonstrated even in the challenging context of the rat interscapular BAT, in which the total volume is small (589681mL, n¼3) and enables the demonstration of potentially important partial volume effects. The measured BAT fat signal fraction was 6565%, consistent with the 37–70% range previously reported in mice by Hu et al. (54), keeping in mind that the BAT fat fraction and volume may vary largely with species, age, and weight (55). Beyond this in vivo proof- of-concept on rat BAT, this work indicates a greater potential for oxygenation measurements in any lipid con- taining tissue where oxygen and lipids play a key role including but not limited to cancer. Furthermore, no obstacle for the transposition of the method on clinical scanners was identified.
Although the current study has some limitations (e.g., the prolonged acquisition time), methods of improve- ment are proposed (e.g., a reduction in the number of TR sampling points, etc.). Furthermore, this study included a small sample size, but as our aim was showing the
feasibility of the method, this validation must be regarded only as a proof-of-concept.
CONCLUSIONS
This article presents a method to estimate fat fraction and fat T1mapping based on the simultaneous, 2D processing of the water and fat phase modulation and T1relaxation data sets. The precision and robustness of the method was demonstrated in vitro and validated by comparison to a reference method based on small receiver bandwidth MRI.
In vivo feasibility was also shown by evaluating the oxy- genation level variation in rat interscapular brown adipose tissue induced by a gas challenge.
ACKNOWLEDGMENTS
The authors acknowledge financial support from ‘Comite Inter-Regional Grand Ouest de la Ligue Contre le Cancer (CIRGO) and would like to thank Dr. Didier Wecker (Bruker Biospin) for his support on method modifica- tions, Olivier Thomas for providing the oil emulsions, and Janske Nel for editing. The authors would like also to thank the developers of the ISMRM (International Society for Magnetic Resonance in Medicine) fat water toolbox (http://www.ismrm.org/workshops/FatWater12/
data.htm) for providing access to said toolbox, which contributed to the inspiration of the chemical shift part of the process.
FIG. 9. In vivo air/carbogen chal- lenge MRI of the rat dorsal region: (a) axial anatomical image to localize brown adipose tissue area (BAT) in the inter- scapular area, (b) sagittal ana- tomical image in the BAT area, (c) corresponding fat signal frac- tion map in percentage, air (d) and carbogen (e) equilibrated fat T1 maps (in seconds), (f) corre- sponding T1percentage of varia- tion map following the air/
carbogen challenge, (g) gray anatomical image overlaid by colored T1 percentage of varia- tion following an air/carbogen challenge, (h) same as g, but processing was carried out only over six TR sampling values (keeping only the first five and the ninth one, corresponding to an acquisition time of <10 min).
All calculated images were thresholded to eliminate mean- ingless pixels (i.e., pixels whose signal is inferior to a thresholded value on the anatomical image) or with a fat R1value equal to 1.
REFERENCES
1. Hammond EM, Asselin MC, Forster D, O’Connor JP, Senra JM, Williams KJ. The meaning, measurement and modification of hypoxia in the laboratory and the clinic. Clin Oncol (R Coll Radiol) 2014;26:
277–288.
2. Tatum JL, Kelloff GJ, Gillies RJ, et al. Hypoxia: importance in tumor biology, noninvasive measurement by imaging, and value of its mea- surement in the management of cancer therapy. Int J Radiat Biol 2006;82:699–757.
3. Lemaire L, Nel J, Franconi F, Bastiat G, Saulnier P. Perfluorocarbon- loaded lipid nanocapsules to assess the dependence of U87-human glioblastoma tumor pO2 on in vitro expansion conditions. PLoS One 2016;11:e0165479.
4. Hodson L. Adipose tissue oxygenation: effects on metabolic function.
Adipocyte 2014;3:75–80.
5. Gaertner FC, Souvatzoglou M, Brix G, Beer AJ. Imaging of hypoxia using PET and MRI. Curr Pharm Biotechnol 2012;13:552–570.
6. Ogawa S, Lee TM, Kay AR, Tank DW. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc Natl Acad Sci USA 1990;87:9868–9872.
7. Lemaire L, Bastiat G, Franconi F, Lautram N, Duong Thi Dan T, Garcion E, Saulnier P, Benoit JP. Perfluorocarbon-loaded lipid nano- capsules as oxygen sensors for tumor tissue pO(2) assessment. Eur J Pharm Biopharm 2013;84:479–486.
8. Corroyer-Dulmont A, Chakhoyan A, Collet S, Durand L, MacKenzie ET, Petit E, Bernaudin M, Touzani O, Valable S. Imaging modalities to assess oxygen status in glioblastoma. Front Med (Lausanne) 2015;
2:57.
9. Zhao D, Jiang L, Mason RP. Measuring changes in tumor oxygenation.
Methods Enzymol 2004;386:378–418.
10. Sotak CH, Hees PS, Huang HN, Hung MH, Krespan CG, Raynolds S.
A new perfluorocarbon for use in fluorine-19 magnetic resonance imaging and spectroscopy. Magn Reson Med 1993;29:188–195.
11. Kodibagkar VD, Wang X, Pacheco-Torres J, Gulaka P, Mason RP. Pro- ton imaging of siloxanes to map tissue oxygenation levels (PISTOL):
a tool for quantitative tissue oximetry. NMR Biomed 2008;2:899–907.
12. Zaharchuk G, Martin AJ, Rosenthal G, Manley GT, Dillon WP. Mea- surement of cerebrospinal fluid oxygen partial pressure in humans using MRI. Magn Reson Med 2005;54:113–121.
13. O’Connor JP, Naish JH, Parker GJ, et al. Preliminary study of oxygen- enhanced longitudinal relaxation in MRI: a potential novel biomarker of oxygenation changes in solid tumors. Int J Radiat Oncol Biol Phys 2009;75:1209–1215.
14. O’Connor JP, Boult JK, Jamin Y, et al. Oxygen-enhanced MRI accu- rately identifies, quantifies, and maps tumor hypoxia in preclinical cancer models. Cancer Res 2016;76:787–795.
15. Lawrence JH, Loomis WF, Tobias CA, Turpin FH. Preliminary obser- vations on the narcotic effect of xenon with a review of values for solubilities of gases in water and oils. J Physiol 1946;105:197–204.
16. Jordan BF, Magat J, Colliez F, et al. Mapping of oxygen by imaging lipids relaxation enhancement: a potential sensitive endogenous MRI contrast to map variations in tissue oxygenation. Magn Reson Med 2013;70:732–744.
17. Colliez F, Safronova MM, Magat J, Joudiou N, Peeters AP, Jordan BF, Gallez B, Duprez T. Oxygen mapping within healthy and acutely infarcted brain tissue in humans using the NMR relaxation of lipids:
a proof-of-concept translational study. PLoS One 2015;10:e0135248.
18. Reeder SB, Pineda AR, Wen Z, Shimakawa A, Yu H, Brittain JH, Gold GE, Beaulieu CH, Pelc NJ. Iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL): appli- cation with fast spin-echo imaging. Magn Reson Med 2005;54:636–
644.
19. Rakow-Penner R, Daniel B, Yu H, Sawyer-Glover A, Glover GH.
Relaxation times of breast tissue at 1.5T and 3T measured using IDEAL. J Magn Reson Imaging 2006;23:87–91.
20. Karampinos DC, Yu H, Shimakawa A, Link TM, Majumdar S. T(1)- corrected fat quantification using chemical shift-based water/fat sepa- ration: application to skeletal muscle. Magn Reson Med 2011;66:
1312–1326.
21. Hu HH, Nayak KS. Apparent change in the T1 of lipids in mixture.
In Proceedings of the 17th Meeting of ISMRM, Honolulu, Hawaii, USA, 2009. p. 4448.
22. Le Ster C, Gambarota G, Lasbleiz J, Guillin R, Decaux O, Saint-Jalmes H. Breath-hold MR measurements of fat fraction, T1 , and T2* of
water and fat in vertebral bone marrow. J Magn Reson Imaging 2016;
44:549–555.
23. Pineda AR, Reeder SB, Wen Z, Pelc NJ. Cramer-Rao bounds for three-point decomposition of water and fat. Magn Reson Med 2005;54:625–635.
24. Jones JA, Hodgkinson P, Barker AL, Hore PJ. Optimal sampling strate- gies for the measurement of spin-spin relaxation times. J Magn Reson B 1996;113:25–34.
25. Spandonis Y, Heese FP, Hall LD. High resolution MRI relaxation measurements of water in the articular cartilage of the meniscectom- ized rat knee at 4.7 T. Magn Reson Imaging 2004;22:943–951.
26. Weiss GH, Gupta RK, Ferretti JA, Becker ED. The choice of optimal parameters for measurement of spin-lattice relxation times. I. Mathe- matical formulation. J Magn Reson 1980;37:369–379.
27. Kellman P, Xue H, Chow K, Spottiswoode BS, Arai AE, Thompson RB. Optimized saturation recovery protocols for T1-mapping in the heart: influence of sampling strategies on precision. J Cardiovasc Magn Reson 2014;16:55.
28. Zhang Y, Yeung HN, O’Donnell M, Carson PL. Determination of sam- ple time for T1 measurement. J Magn Reson Imaging 1998;8:675–681.
29. Shahin M, Abdel Hady S, Hammad M, Mortada N. Development of stable o/w emulsions of three different oils. International Journal of Pharmaceutical Studies and Research 2011;2:45–51.
30. Reeder SB, Hines CD, Yu H, McKenzie CA, Brittain JH. On the defini- tion of fat-fraction for in vivo fat quantification with magnetic reso- nance imaging. In Proceedings of the 17th Meeting of ISMRM Honolulu, Hawaii, USA, 2009. p. 211.
31. Bottomley PA. Spatial localization in NMR spectroscopy in vivo.
Ann N Y Acad Sci 1987;508:333–348.
32. Salvati R, Gambarota G. MRI-based direct measurements of the T2*
transverse relaxation time of water and lipid protons in water-lipid mixtures. Appl Magn Reson 2016;47:139–148.
33. Reeder SB, McKenzie CA, Pineda AR, Yu H, Shimakawa A, Brau AC, Hargreaves BA, Gold GE, Brittain JH. Water-fat separation with IDEAL gradient-echo imaging. J Magn Reson Imaging 2007;25:644–652.
34. Hernando D, Liang ZP, Kellman P. Chemical shift-based water/fat separa- tion: a comparison of signal models. Magn Reson Med 2010;64:811–822.
35. Hernando D, Haldar JP, Sutton BP, Ma J, Kellman P, Liang ZP. Joint estimation of water/fat images and field inhomogeneity map. Magn Reson Med 2008;59:571–580.
36. Yu H, McKenzie CA, Shimakawa A, Vu AT, Brau AC, Beatty PJ, Pineda AR, Brittain JH, Reeder SB. Multiecho reconstruction for simultaneous water-fat decomposition and T2* estimation. J Magn Reson Imaging 2007;26:1153–1161.
37. Reeder SB, Robson PM, Yu H, Shimakawa A, Hines CD, McKenzie CA, Brittain JH. Quantification of hepatic steatosis with MRI: the effects of accurate fat spectral modeling. J Magn Reson Imaging 2009;29:1332–1339.
38. Yu H, Shimakawa A, McKenzie CA, Brodsky E, Brittain JH, Reeder SB. Multiecho water-fat separation and simultaneous R2* estimation with multifrequency fat spectrum modeling. Magn Reson Med 2008;
60:1122–1134.
39. Bydder M, Yokoo T, Hamilton G, Middleton MS, Chavez AD, Schwimmer JB, Lavine JE, Sirlin CB. Relaxation effects in the quanti- fication of fat using gradient echo imaging. Magn Reson Imaging 2008;26:347–359.
40. Hernando D, Kellman P, Haldar JP, Liang ZP. Robust water/fat sepa- ration in the presence of large field inhomogeneities using a graph cut algorithm. Magn Reson Med 2010;63:79–90.
41. Yu H, Reeder SB, Shimakawa A, Brittain JH, Pelc NJ. Field map esti- mation with a region growing scheme for iterative 3-point water-fat decomposition. Magn Reson Med 2005;54:1032–1039.
42. Ren J, Dimitrov I, Sherry AD, Malloy CR. Composition of adipose tis- sue and marrow fat in humans by 1H NMR at 7 Tesla. J Lipid Res 2008;49:2055–2062.
43. Tsao J, Jiang Y. Hierarchical IDEAL: fast, robust, and multiresolution separation of multiple chemical species from multiple echo times.
Magn Reson Med 2013;70:155–159.
44. Yu H, Shimakawa A, Hines CD, McKenzie CA, Hamilton G, Sirlin CB, Brittain JH, Reeder SB. Combination of complex-based and magnitude-based multiecho water-fat separation for accurate quantifi- cation of fat-fraction. Magn Reson Med 2011;66:199–206.
45. Hernando D, Hines CD, Yu H, Reeder SB. Addressing phase errors in fat-water imaging using a mixed magnitude/complex fitting method.
Magn Reson Med 2012;67:638–644.
46. Jordan BF, Magat J, Colliez F, Ozel E, Fruytier AC, Marchand V, Mignion L, Gallez B. Application of MOBILE (mapping of oxygen by
imaging lipids relaxation enhancement) to study variations in tumor oxygenation. Adv Exp Med Biol 2013;789:281–288.
47. Colliez F, Neveu MA, Magat J, Cao Pham TT, Gallez B, Jordan BF.
Qualification of a noninvasive magnetic resonance imaging bio- marker to assess tumor oxygenation. Clin Cancer Res 2014;20:5403–
5411.
48. Safronova MM, Colliez F, Magat J, Joudiou N, Jordan BF, Raftopoulos C, Gallez B, Duprez T. Mapping of global R1 and R2* values versus lipids R1 values as potential markers of hypoxia in human glial tumors: a feasibility study. Magn Reson Imaging 2016;34:105–113.
49. Wankhade UD, Shen M, Yadav H, Thakali KM. Novel browning agents, mechanisms, and therapeutic potentials of brown adipose tis- sue. Biomed Res Int 2016;2016:2365609.
50. Trayhurn P, Alomar SY. Oxygen deprivation and the cellular response to hypoxia in adipocytes - perspectives on white and brown adipose tissues in obesity. Front Endocrinol (Lausanne) 2015;6:19.
51. Sbarbati A, Cavallini I, Marzola P, Nicolato E, Osculati F. Contrast- enhanced MRI of brown adipose tissue after pharmacological stimula- tion. Magn Reson Med 2006;55:715–718.
52. Branca RT, He T, Zhang L, Floyd CS, Freeman M, White C, Burant A.
Detection of brown adipose tissue and thermogenic activity in mice
by hyperpolarized xenon MRI. Proc Natl Acad Sci USA 2014;111:
18001–18006.
53. Khanna A, Branca RT. Detecting brown adipose tissue activity with BOLD MRI in mice. Magn Reson Med 2012;68:1285–1290.
54. Hu HH, Smith DL Jr, Nayak KS, Goran MI, Nagy TR. Identification of brown adipose tissue in mice with fat-water IDEAL-MRI. J Magn Reson Imaging 2010;31:1195–1202.
55. Casteilla L, Penicaud L, Cousin B, Calise D. Choosing an adipose tis- sue depot for sampling: factors in selection and depot specificity.
Methods Mol Biol 2008;456:23–38.
SUPPORTING INFORMATION
Additional Supporting Information may be found in the online version of this article.
Table S1.Experimental mean R1values for fat and water, measured with the chemical shift T1method, are presented with standard deviation (n53).
Percentage error is calculated for the experimental values in comparison to the reference value obtained with the small receiver bandwidth method as
% error5100*abs(experimental R12reference R1)/reference R1.
