Page 1
Optimized efficient liver T1ρ mapping using limited spin lock times
This article has been downloaded from IOPscience. Please scroll down to see the full text article.
2012 Phys. Med. Biol. 57 1631
(http://iopscience.iop.org/0031-9155/57/6/1631)
Download details:
IP Address: 137.189.150.110
The article was downloaded on 08/03/2012 at 01:08
Please note that terms and conditions apply.
View the table of contents for this issue, or go to the journal homepage for more
HomeSearchCollectionsJournalsAboutContact usMy IOPscience
Page 2
IOP PUBLISHING
PHYSICS IN MEDICINE AND BIOLOGY
Phys. Med. Biol. 57 (2012) 1631–1640
doi:10.1088/0031-9155/57/6/1631
Optimized efficient liver T1ρmapping using limited
spin lock times
Jing Yuan1, Feng Zhao1, James F Griffith1, Queenie Chan2
and Yi-Xiang J Wang1
1Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong,
Shatin, New Territories, Hong Kong, People’s Republic of China
2Philips Healthcare, Philips Electronics Hong Kong Limited, Hong Kong, People’s Republic of
China
E-mail: jyuan@cuhk.edu.hk
Received 27 June 2011, in final form 26 January 2012
Published 7 March 2012
Online at stacks.iop.org/PMB/57/1631
Abstract
T1ρrelaxation has recently been found to be sensitive to liver fibrosis and
has potential to be used for early detection of liver fibrosis and grading.
Liver T1ρimaging and accurate mapping are challenging because of the long
scan time, respiration motion and high specific absorption rate. Reduction and
optimization of spin lock times (TSLs) are an efficient way to reduce scan
time and radiofrequency energy deposition of T1ρimaging, but maintain the
near-optimal precision of T1ρmapping. This work analyzes the precision in
T1ρestimation with limited, in particular two, spin lock times, and explores
the feasibility of using two specific operator-selected TSLs for efficient and
accurate liver T1ρmapping. Two optimized TSLs were derived by theoretical
analysis and numerical simulations first, and tested experimentally by in vivo
rat liver T1ρimaging at 3 T. The simulation showed that the TSLs of 1 and
50 ms gave optimal T1ρestimation in a range of 10–100 ms. In the experiment,
no significant statistical difference was found between the T1ρmaps generated
using the optimized two-TSL combination and the maps generated using the
six TSLs of [1, 10, 20, 30, 40, 50] ms according to one-way ANOVA analysis
(p = 0.1364 for liver and p = 0.8708 for muscle).
(Some figures may appear in colour only in the online journal)
Introduction
MRI offers a variety of mechanisms to produce different contrasts in anatomical and
functional images. In addition to T1, T2and T2
in the rotating frame) is another relaxation parameter that represents the time constant of
transversemagnetizationdecayduringtheapplicationofanon-resonantcontinuouswave(CW)
∗, T1ρrelaxation (the spin–lattice relaxation
0031-9155/12/061631+10$33.00 © 2012 Institute of Physics and Engineering in MedicinePrinted in the UK & the USA 1631
Page 3
1632J Yuan et al
radiofrequency (RF) pulse, i.e.spinlock (SL) pulse, aligned withthe net magnetization vector.
With T1ρrelaxation, there is potential to investigate interactions between motionally restricted
water molecules and surrounding macromolecules due to the high sensitivity of T1ρto low
frequency motional and static processes. T1ρrelaxation has been proposed for various clinical
applications. For example, T1ρrelaxation can be used as a sensitive biomarker to detect early
stages of intervertebral disk degeneration (Blumenkrantz et al 2006, 2010, Johannessen et al
2006, Nguyen et al 2008) and cartilage degeneration in osteoarthritis (Duvvuri et al 1997,
Mlynarik et al 1999, Regatte et al 2003, Li et al 2009). T1ρrelaxation variation has also been
investigated as a predictor of Alzheimer’s disease progression (Borthakur et al 2006, 2008,
Haris et al 2009). Recently, Wang et al found that T1ρrelaxation is sensitive to liver fibrosis
and has potential tobeused forliver fibrosisdiagnosis and assessment(Wangetal2011,Sirlin
2011). Different from T1ρimaging of intervertebral disk, cartilage and brain that are almost
stationary, in vivo liver T1ρimaging is subject to respiration motion that may compromise
the accuracy and precision of T1ρmapping. Therefore, a short scan time for in vivo liver
T1ρimaging is essential.
Accurate and precise T1ρmapping is challenging under the scan time constraint because
multiple SL times (TSLs) are required for T1ρ-weighted magnetization preparation and a
long delay time is usually necessary in the SL pulse sequence for longitudinal magnetization
restoration. SL pulses are also sensitive to magnetic field heterogeneities. SL pulses with
different TSLs may introduce different spatially distributed banding-like artifacts in images
thatmaycompromisetheaccuracyofT1ρmapping.Inpractice,theuseofhighSLfrequenciesis
helpfulinalleviatingtheseartifacts,however,remarkablyincreasesthetotalspecificabsorption
rate (SAR) and RF energy deposition in patients.
The reduction of TSL numbers is an apparent strategy to enhance T1ρimaging efficiency.
However, the reduced TSLs have to be optimized to minimize the mapping uncertainty.
T1ρ relaxation is conventionally described as an exponential decay with respect to TSL,
similar to T1and T2. Many studies have been conducted on the choice of optimal selection of
observation time (e.g. inversion recovery time for T1measurement) to produce the smallest
estimation error for T1and T2mapping (Becker et al 1980, Weiss et al 1980, Sezginer et al
1991, Bain 1990, Taitelbaum et al 1994, Jones et al 1996, Zhang et al 1998, Fleysher et al
2008). Although T1ρmapping with limited TSLs could be inspired from these studies, there
are still some limitations and restrictions in the previous studies. The optimal observation
times proposed for T1or T2mapping are usually illustrated either in complicated mathematical
equations (Weiss et al 1980, Weiss and Ferretti 1985a, 1985b, Becker et al 1980, Zhang
et al 1998) or in implicit criteria (Bain 1990). The scan time efficiency of the proposed
optimal observation time is not usually maximized. More importantly, most studies are
completely based on theoretical derivation or numerical simulation without experiment
verifications. In addition, the distinctive characteristics of T1ρimaging, such as high SAR
and sensitivity to field heterogeneity, make accurate T1ρmapping under the critical time
constraint even more challenging than T1and T2mapping. To the best of our knowledge,
no optimized efficient acquisition strategies have been proposed for T1ρ imaging and
mapping.
In this study, we first simplified the theoretical analysis based on a two-parameter
exponential T1ρrelaxation model so that only two optimized TSLs were required to minimize
the estimation variance and maximize the time efficiency. These two optimized TSLs were
determined by mathematical analysis and then tested by Monte Carlo numerical simulation.
Finally,theT1ρmappingprecisionwiththeoptimizedtwoTSLswasconfirmedexperimentally
by in vivo rat liver T1ρimaging.
Page 4
Efficient T1ρmapping with limited TSLs1633
Figure 1. The schematic of a rotary echo SL pulse imbedded bFFE SL pulse sequence.
Methods
Theoretical analysis
T1ρrelaxation can be normally described as a three-parameter exponential decay function with
SL time TSL, as shown in equation (1):
Si= A · e−TSLi/T1ρ+ B,
where A, B and T1ρrelaxation are the three parameters to fit. Siis the acquired image intensity
attheSLtimeofTSLi.Giventhattheflipangleofthepreparationpulse(90◦
restoration pulse (90◦
after the restoration pulse shows purely exponential-decayed components (Charagundla et al
2003), so the image intensity can be simplified as a two-parameter exponential function of
S0and T1ρ:
Si= S0· e−TSLi/T1ρ,
where S0isthebaseline image intensityacquired atTSLof zero.The standard deviation?σiT1ρ
?∂T1ρ
Thenoiselevelδiisusuallydeterminedbythereceiverhardware,imagingprotocolandsample
properties, and independent of SL duration, so can be rewritten as a general term of δ. By
substituting equation (2) into equation (3), it is derived that
?∂T1ρ
The term S0/δ is defined as the equilibrium signal-to-noise ratio (SNR). When a series of N
SL times (i = 1, 2,..., N) is used, the aggregate standard deviation (σT1ρ) is dependent on
many factors, such as the individual standard deviation (σiT1ρ), the exponential decay model,
as well as the least-squares fitting (Zhang et al 1998, Weiss and Ferretti 1985a, 1985b, Weiss
et al 1980, Becker et al 1980).
Here, the standard deviation analysis can be significantly simplified since only two TSLs
are specified for optimal T1ρestimation. The first specific SL time TSL1(i = 1) is selected as
zero or a very small duration relative to the true T1ρvalue (T1ρt). As such, S1is approximated
as S0(S1≈ S0). Meanwhile, S1is normally much larger than the noise, i.e. S1? δ. The second
SLtimeTSL2(i=2)shouldbeselectedsufficientlyfarawayfromTSL1toavoidapronounced
deviation for large true T1ρvalues (T1ρt? TSL1and T1ρt? TSL2). Meanwhile, TSL2is also
required to satisfy S2? δ to avoid an overwhelming noise level. It is derived that
σ2
S2
0
∂s2
S2
1
∂s2
(1)
xinfigure1)andthe
−xin figure 1) are well calibrated to 90◦, the longitudinal magnetization
(2)
?
or variance?σ2
iT1ρ
?due to the presence of noise (δi) for TSLican be written as
σ2
i
∂Si
iT1ρ= δ2
?2
.
(3)
σ2
iT1ρ= δ2
∂Si
?2
=δ2
S2
0
?∂T1ρ
∂si
?2
=
1
SNR2·
TSL4
i· (logsi)4
i
s2
where
si=Si
S0.
(4)
2T1ρ=δ2
?∂T1ρ
?2
≈δ2
?∂T1ρ
?2
=
1
SNR2·
TSL4
2
s2
2· (logs2)4=e2·TSL2/T1ρ
SNR2
·
T4
1ρ
TSL2
2
.
(5)
Page 5
1634 J Yuan et al
By neglecting the covariance factor between S1and S2, σ2T1ρis approximately equal to the
aggregate σT1ρ. As seen from equation (5), T1ρestimation variance is inversely proportional to
the square of SNR. It can also be proved from equation (5) that the minimum of σ2
at a TSL2that is equal to the true T1ρvalue to be fit.
2T1ρis found
Numerical simulation
DeviationsofT1ρestimationweretestedbyMonteCarlosimulationonMATLAB(MathWorks,
Natick, MA, USA) for different true T1ρvalues by using different combinations of TSLs. In
Monte Carlo simulations, signals were generated according to equation (2) with the given true
T1ρvalues and the selected TSLs. Rician noise was added to each signal at the mean level of
δ = 0.01S0(SNR = 100). These noise-imposed signals were then least-squares fitted using
equation (2) to obtain the estimated T1ρvalues. Here the T1ρdeviation was reported as the
absolute difference between the true T1ρvalue and the best fitted T1ρvalue in the presence
of Rician noise. Each simulation was repeated 60000 times, sufficiently large to obtain the
reliable and smooth estimated T1ρdistribution. Note that no assumption of S1≈ S0was made
in Monte Carlo simulations.
T1ρimaging
Five male Sprague-Dawley rats with weight of 200–250 g were scanned under appropriate
animal ethical approval. T1ρimaging was performed on a 3 T clinical scanner (Philips Medical
Systems, Best, The Netherlands) using a T1ρprepared multi-slice 2D balanced fast field echo
(bFFE) (or balanced steady-state free precession, SSFP) sequence (Witschey et al 2008) with
a four-channel human wrist array coil. A schematic for the SL bFFE sequence is shown in
figure 1. A spectrally selective RF pulse was used for fat suppression followed by the
T1ρ preparation module in which a B1-insensitive rotary echo SL pulse was embedded
(Charagundla et al 2003). In the bFFE module, a half-alpha pulse and four startup pulses
were used first to approach the steady state but with T1ρ-weighted preparation maintained
(Witschey et al 2008). Normal phase alternating bFFE readout was used for acquisition
with a centric acquisition order to maintain the T1ρ-weighting of images during the transient
status. Eighty k-space lines were acquired for each shot acquisition. A delay time (TD) of
5000 ms after each shot acquisition was set to restore equilibrium magnetization prior to the
next T1ρpreparation. Primary imaging parameters included TR/TE = 5.0/2.5 ms, FA = 30◦,
FOV = 80 mm, pixel size = 0.3 mm, slice thickness = 2 mm, slice number = 6, NSA (number
of signal average) = 4 and receiver bandwidth BW = 128 kHz. A SL time series of 1, 10,
20, 30, 40 and 50 ms was applied and the SL frequency was 500 Hz. A sensitivity encoding
(SENSE) factor of 1.5 was used for scan and constant level appearance (CLEAR) was applied
for coil sensitivity correction during reconstruction. The total scan time was around 20 min
for each rat scan.
All images were processed using a home-made MATLAB program to generate T1ρmaps
according to equation (2). T1ρmaps generated from all SL times were compared to those
generated from the selected SL times in the full series.
Results
Theoretical analysis and numerical simulation
Figure 2(a) shows a plot of the estimated T1ρstandard deviation distribution for the true
T1ρ range from 15 to 140 ms using different TSL2durations according to equation (5).
Page 6
Efficient T1ρmapping with limited TSLs1635
(a)
(b)
Figure 2. (a) The theoretical standard deviations of T1ρestimation for the true T1ρrange from
15 to 140 ms using different TSL2durations; (b) a zoomed plot of (a) showing the T1ρstandard
deviation within a narrower true T1ρfrom 20 to 80 ms.
(a)(b)
Figure 3. Monte Carlo simulation results: (a) comparison of the T1ρestimation deviations when
using different TSL combinations. (b) The T1ρestimation deviations using different combinations
of two TSLs.
Figure 2(b) shows a zoomed plot of figure 1(a) within a narrower range. The deviation is
generally smaller than 4.5 ms by using a TSL2within the range of 20–80 ms for the true
T1ρvalues within the same range. The optimal TSL2was found to be 50 ms through a 1 ms
step-wise search within the true T1ρrange of 10 to 100 ms.
ThehistogramsofbestfittedT1ρvaluesintheMonteCarlosimulationwereapproximately
normally distributed in the presence of Rician noise. Figure 3 shows simulation results. In
figure 3(a), the TSL combination of [1, 10, 20, 30, 40, 50] ms gave the smallest deviation
sum (the area under the curve) in the true T1ρrange of 10 to 100 ms. The TSLs of [1, 20,
40, 60, 80, 100] ms gave smaller deviations for true T1ρvalues larger than 60 ms but much
larger deviations for small true T1ρvalues. For two TSLs of [1, 50] ms, every individual
T1ρdeviation was smaller than 4 ms. The deviation was just slightly larger (<1 ms) than the
corresponding deviation with the TSLs of [1, 10, 20, 30, 40, 50] ms. It is interesting to find
that T1ρdeviation for [1, 50] ms was even smaller than that for [1, 20, 50] ms. Figure 3(b)
shows the T1ρdeviations using different combinations of two TSLs. [1, 50] ms provided a
Page 7
1636 J Yuan et al
Figure 4. T1ρ-weighted rat liver images acquired at different TSLs. The mean image intensity ratio
of signal (S) to noise (N) ROI is used for SNR estimation.
good compromise in T1ρestimation for wide true T1ρranges, and had the smallest deviation
sum in total, consistent with the theoretical analysis. Agreeing with equation (3), simulation
also showed that T1ρdeviation is inversely proportional to the SNR for true T1ρvalues larger
than 20 ms by using different two-TSL combinations.
Experiments
Figure 4 shows T1ρ-weighted in vivo rat liver images acquired at different TSLs. The SNR was
estimated as the mean intensity ratio of the signal region-of-interest (ROI) (S in figure 4) to
the noise ROI (N in figure 4). The signal ROI was selected as a square region in the rat liver
that had relatively homogenous intensities. The noise ROI was selected to be a position-fixed
background square region that was relatively close to the rat liver but without motion induced
artifact to account for the effect of coil sensitivity correction on noise calibration by CLEAR.
In general, the liver SNRs were sufficiently high (SNR > 40) for T1ρmapping.
T1ρmaps generated from the different combinations of TSLs are shown in the first row of
figure5.TheT1ρmapgeneratedfromthefullseriesofsixTSLswasusedasabaselinereference
for comparison. The second row shows the corresponding coefficient of determination (R2)
maps.TheoverallR2>0.9indicatesthatexcellentfittingwasachieved.NotethatR2wasalways
1 for fitting with only two TSLs of 1 and 50 ms. The third row shows the absolute T1ρmap
differences from the baseline reference. The T1ρmaps generated from TSL combination of
[1, 20, 50] and [1, 50] ms show very small discrepancies compared to the reference T1ρmap.
However, no apparent improvement of T1ρestimation is achieved with a TSL combination
of [1, 20, 50] ms compared to [1, 50] ms. In comparison, the T1ρmaps generated from TSL
combinations of [1, 10, 20] and [10, 20, 30, 40] ms were considerably different from the
baseline reference.
Page 8
Efficient T1ρmapping with limited TSLs 1637
Figure 5. Rat liver T1ρmapping comparison. The top row shows the T1ρmaps generated from the
different TSL combinations, of which the very left one is a baseline reference for comparison. The
middlerowshowsmapsofthecoefficientofdetermination(R2)forthecorrespondingT1ρmapping.
The bottom row shows the T1ρmap differences compared to the baseline reference.
(a)(b)
Figure 6. Box plots of pixel T1ρvalues in the liver ROI1 (a) and muscle ROI2 (b) as shown in
figure 5. In the liver ROI1, the difference between the medians of the first three boxes is smaller
than 1 ms. The p value of a one-way ANOVA analysis is 0.1364 (>0.05), indicating no significant
difference. In the muscle ROI2, an even smaller difference in the T1ρmaps is found (ANOVA p =
0.8708). Pixels are drawn as outliers (+) in the box plot if they are larger than q3 + 1.5x(q3 – q1)
or smaller than q1 – 1.5x(q3 – q1), where q1 and q3 are the 25th and 75th percentiles, respectively.
To quantitatively analyze the T1ρmap difference for different TSL combinations, two
representative ROIs were drawn (as shown in figure 5) on liver (ROI1) and muscle (ROI2).
T1ρvalues in these two ROIs were compared by the box plots shown in figure 6. In the liver
ROI (figure 6(a)), the difference between the medians of the first three boxes is smaller than
1 ms. Their notches also overlapped with each other, indicating that their true medians did
Page 9
1638 J Yuan et al
not differ with 95% confidence. The T1ρdistributions within 25th (the lower edge of the
box) and 75th percentiles (the upper edge of the box) are all smaller than 2 ms. However,
the median and the T1ρdistribution of the fourth and fifth boxes were significantly different
from the reference (the first box on the left). A one-way ANOVA was applied to analyze the
statistical difference between the first three boxes. The p value was 0.1364 (>0.05), showing
no statistically significant difference of T1ρestimation. In the muscle ROI (figure 6(b)), an
even smaller T1ρestimation difference (p = 0.8708) was found. This may attribute to the
smaller respiration motion effect, the absence of major blood vessels and the smaller tissue
heterogeneity in muscles.
For the other four rats, the T1ρmaps obtained by using all TSLs were not significantly
different (averaged p = 0.1684 in livers and p = 0.8508 in muscles) from those obtained by
using the TSLs of 1 and 50 ms.
Discussion
In principle, more TSLs are considered to be more likely to achieve precise T1ρestimation.
However, when TSLs are all much shorter than the true T1ρvalue, large estimation deviation
can be introduced. On the other hand, TSLs that are too long may result in images with low
SNRs. For T1ρestimation with two TSLs, having an approximate value of S0by using a very
short TSL should be important. In both simulation and experiments, it was interesting to find
that the use of the combination of [1, 20, 50] ms did not improve the precision in T1ρmapping
as compared to TSL combination of [1, 50] ms. However, this finding may be only valid for
the situation of high SNRs larger than 100.
The simplified derivation presented in equation (5), although consistent with the full
theoretical frame described in literatures (Weiss et al 1980, Zhang et al 1998), has limited
validity for only two particularly selected TSLs. For a specific application, a pre-knowledge
of true T1ρvalues of interest is quite useful. In this study, the TSL combination of 1 and 50 ms,
although considered optimal for the true T1ρrange of 10 to 100 ms, is particularly suitable for
liver T1ρmapping because the normal and the fibrosed rat livers have T1ρvalues within 40–
65 ms (Wang et al 2011), and normal human livers have T1ρvalues within 40–50 ms (Singh
et al 2011). This TSL combination should also be robust for T1ρmapping in other soft tissues
such as brain and cartilage as their T1ρvalues are usually smaller than 100 ms.
T1ρ deviation using the optimized two TSLs is inversely proportional to the SNR
(equation (5)) in theory. Although this conclusion was derived by assuming S1 ≈ S0, it
was verified by the Monte Carlo simulation, where no such assumption was made, that this
conclusion could still be valid even for a low SNR. This indicates that the noise imposed
on TSL2dominates the T1ρdeviation. It is worth noting that this inverse proportionality
is independent of the noise type. As the SNR is also proportional to the square root of
the NSA, a relevant interesting question would be whether it is useful to measure signal
intensities at two TSLs with higher SNR or at more TSLs with lower SNR during the same
measurement time. According to equation (5), T1ρdeviation should be inversely proportional
tothesquarerootofNSA.Referringtofigure3(a),T1ρdeviationusingTSLsof[1,50]mswith
doubled NSA only at TSL2of 50 ms may be comparable with that using TSLs of [1, 10, 20,
30, 40 50] ms.
To ensure the precision of T1ρmapping with limited TSLs, a high SNR for T1ρ-weighted
images is of critical importance. For this reason, RF coils with high sensitivity and pulse
sequences with high SNR efficiency are preferable. As the SNR is also dependent on many
other factors such as pixel size, NSA, TE and TR, all these factors should be optimized
according to the specific application.
Page 10
Efficient T1ρmapping with limited TSLs1639
The susceptibility of T1ρimaging to B0and B1field inhomogeneity related artifacts
may compromise the robustness and reliability of using a limited number of TSLs for
T1ρmapping. Unlike the case for multiple-TSL acquisition, the presence of artifacts by using
two TSLs would result in considerable T1ρdeviation that cannot be compensated through
least-squares fitting due to the lack of extra data points. Therefore, the minimization of field
inhomogeneity related artifacts is critically important. Good volume shimming should be
helpful to reduce B0inhomogeneity, particularly for SSFP sequence. A rotary echo pulse
(Charagundla et al 2003) was used in this study to minimize the banding artifacts caused
by B1field inhomogeneity. To further reduce the B0inhomoegentiy induced artifacts, a SL
pulse that is insensitive to both B0and B1field inhomogeneity could be applied (Witschey
et al 2007). In a pilot study of human liver T1ρimaging that we are conducting, it is shown
that good breath hold of subjects is also of importance to reduce field inhomogeneity related
artifacts.
By the use of the optimized two TSLs, the scan time was reduced approximately by a
factorof3withoutcompromisingtheT1ρmappingprecision,comparedtoanacquisitionusing
a full series of six TSLs. For T1ρmapping for moving organs such as liver, the reduction of
TSLsmayreduceinter-scanbulkmotiondisplacementbutnotintra-scanmotionartifacts;good
breath hold of subjects should be of essence to reduce respiration artifact for the translation to
human scan. In terms of RF energy deposition, SL RF SAR level is proportional to the square
of the flip angle of SL pulses within the unit pulse duration (Witschey et al 2007, Collins
et al 1998), equivalent to SL frequency. Therefore, the average SAR was not reduced by the
reduction of TSLs. However, the total RF energy deposition by the SL pulses is approximately
reduced by a factor of 3. This method is also compatible with other means such as keyhole
(Wheaton et al 2003), parallel imaging (Pakin et al 2006), and off-resonance SL imaging
(Santyr et al 1994) for further reduction of SAR. Currently, we are testing the human liver
T1ρimaging using the rotary echo SL pulse with the SL frequency of 500 Hz; SAR deposition
is generally under the allowable food and drug administration limits of 0.4 W kg−1averaged
over the whole body, as indicated on the MRI console, confirming its safety for translation to
human T1ρimaging.
In conclusion, optimized T1ρmapping with high precision using two optimized specific
TSLs was proposed for liver T1ρmapping. This technique is able to significantly reduce scan
time and total RF energy deposition for T1ρimaging, in particular at high field strength.
Acknowledgements
This work is supported by Hong Kong ITF grant ITS/021/10 and Research Grants Council
grant CUHK475911 and SEG_CUHK02.
References
Bain A D 1990 The choice of parameters in an NMR experiment. Application to the inversion-recovery T1method
J. Magn. Reson. 89 153–60
Becker E D, Ferretti J A, Gupta R K and Weiss G H 1980 The choice of optimal parameters for measurement of spin-
lattice relaxation-times: II Comparison of saturation recovery, inversion recovery, and fast inversion recovery
experiments J. Magn. Reson. 37 381–94
Blumenkrantz G, Li X, Han E T, Newitt D C, Crane J C, Link T M and Majumdar S 2006 A feasibility study of in
vivo T1ρimaging of the intervertebral disc Magn. Reson. Imaging 24 1001–7
BlumenkrantzG,ZuoJ,LiX,KornakJ,LinkT MandMajumdarS2010Invivo3.0-teslamagneticresonanceT1ρand
T2relaxation mapping in subjects with intervertebral disc degeneration and clinical symptoms Magn. Reson.
Med. 63 1193–200
Page 11
1640J Yuan et al
Borthakur A, Gur T, Wheaton A J, Corbo M, Trojanowski J Q, Lee V M and Reddy R 2006 In vivo measurement of
plaque burden in a mouse model of Alzheimer’s disease J. Magn. Reson. Imaging 24 1011–7
Borthakur A, Sochor M, Davatzikos C, Trojanowski J Q and Clark C M 2008 T1ρMRI of Alzheimer’s disease
Neuroimage 41 1199–205
Charagundla S R, Borthakur A, Leigh J S and Reddy R 2003 Artifacts in T(1rho)-weighted imaging: correction with
a self-compensating spin-locking pulse J. Magn. Reson. 162 113–21
Collins C M, Li S and Smith M B 1998 SAR and B1 field distributions in a heterogeneous human head model within
a birdcage coil. Specific energy absorption rate Magn. Reson. Med. 40 847–56
Duvvuri U, Reddy R, Patel S D, Kaufman J H, Kneeland J B and Leigh J S 1997 T1ρ-relaxation in articular cartilage:
effects of enzymatic degradation Magn. Reson. Med. 38 863–7
Fleysher R, Fleysher L and Gonen O 2008 The optimal MR acquisition strategy for exponential decay constants
estimation Magn. Reson. Imaging 26 433–5
Haris M, McArdle E, Fenty M, Singh A, Davatzikos C, Trojanowski J Q, Melhem E R, Clark C M and Borthakur A
2009 Early marker for Alzheimer’s disease: hippocampus T1ρ(T(1ρ)) estimation J. Magn. Reson. Imaging
29 1008–12
Johannessen W, Auerbach J D, Wheaton A J, Kurji A, Borthakur A, Reddy R and Elliott D M 2006 Assessment
of human disc degeneration and proteoglycan content using T1ρ-weighted magnetic resonance imaging Spine
(Phila Pa 1976) 31 1253–7
JonesJ A,HodgkinsonP,BarkerA LandHoreP J1996Optimalsamplingstrategiesforthemeasurementofspin–spin
relaxation times J. Magn. Reson. Ser. B 113 25–34
Li X, Pai A, Blumenkrantz G, Carballido-Gamio J, Link T, Ma B, Ries M and Majumdar S 2009 Spatial distribution
andrelationshipofT1ρandT2relaxationtimesinkneecartilagewithosteoarthritisMagn.Reson.Med. 61 1310–8
Mlynarik V, Trattnig S, Huber M, Zembsch A and Imhof H 1999 The role of relaxation times in monitoring
proteoglycan depletion in articular cartilage J. Magn. Reson. Imaging 10 497–502
NguyenA M,JohannessenW,YoderJ H,WheatonA J,VresilovicE J,BorthakurAandElliottD M2008Noninvasive
quantification of human nucleus pulposus pressure with use of T1ρ-weighted magnetic resonance imaging
J. Bone Joint Surg. Am. 90 796–802
Pakin S K, Xu J, Schweitzer M E and Regatte R R 2006 Rapid 3D-T1ρmapping of the knee joint at 3.0T with parallel
imaging Magn. Reson. Med. 56 563–71
Regatte R R, Akella S V, Borthakur A, Kneeland J B and Reddy R 2003 In vivo proton MR three-dimensional T1ρ
mapping of human articular cartilage: initial experience Radiology 229 269–74
Santyr G E, Fairbanks E J, Kelcz F and Sorenson J A 1994 Off-resonance spin locking for MR imaging Magn. Reson.
Med. 32 43–51
Sezginer A, Kleinberg R L, Fukuhara M and Latour L L 1991 Very rapid simultaneous measurement of nuclear
magnetic resonance spin-lattice relaxation time and spin–spin relaxation time J. Magn. Reson. 92 504–27
Singh A, Haris M, Cai K, Witschey W, Hariharan H and Reddy R 2011 Fluid suppressed T1ρmapping of human liver
on clinical scanners Proc. 19th Scientific Meeting, Int. Society for Magnetic Resonance in Medicine p 4502
Sirlin C B 2011 Science to practice: can T1ρimaging be used to diagnose and assess the severity of hepatic fibrosis?
Radiology 259 619–20
Taitelbaum H, Ferretti J A, Spencer R G S and Weiss G H 1994 Optimization of 2-stage measurements of T-1
J. Magn. Reson. Ser. A 109 166–71
Wang Y X, Yuan J, Chu E S, Go M Y, Huang H, Ahuja A T, Sung J J and Yu J 2011 T1ρMR imaging is sensitive to
evaluate liver fibrosis: an experimental study in a rat biliary duct ligation model Radiology 259 712–9
Weiss G H and Ferretti J A 1985a The choice of optimal parameters for measurement of spin-lattice relaxation times:
III Mathematical preliminaries for nonideal pulses J. Magn. Reson. 61 490–8
Weiss G H and Ferretti J A 1985b The choice of optimal parameters for measurement of spin-lattice relaxation times:
IV Effects of nonideal pulses J. Magn. Reson. 61 499–515
WeissG H,GuptaR K,FerrettiJ AandBeckerE D1980Choiceofoptimalparametersformeasurementofspin-lattice
relaxation times: I Mathematical formulation J. Magn. Reson. 37 369–79
Wheaton A J, Borthakur A and Reddy R 2003 Application of the keyhole technique to T1ρrelaxation mapping
J. Magn. Reson. Imaging 18 745–9
Witschey W R, Borthakur A, Elliott M A, Fenty M, Sochor M A, Wang C and Reddy R 2008 T1ρ-prepared balanced
gradient echo for rapid 3D T1ρMRI J. Magn. Reson. Imaging 28 744–54
Witschey W R II, Borthakur A, Elliott M A, Mellon E, Niyogi S, Wallman D J, Wang C and Reddy R 2007 Artifacts
in T1ρ-weighted imaging: compensation for B(1) and B(0) field imperfections J. Magn. Reson. 186 75–85
Zhang Y, Yeung H N, O’Donnell M and Carson P L 1998 Determination of sample time for T1measurement J. Magn.
Reson. Imaging 8 675–81
Download full-text