Professional Documents
Culture Documents
Compressed Sensing in Quantitative Determination of GAG Concentration in Cartilage by Microscopic MRI
Compressed Sensing in Quantitative Determination of GAG Concentration in Cartilage by Microscopic MRI
Purpose: To evaluate the potentials of compressed sensing (SZ) where the fibers are parallel to the tissue surface,
(CS) in MRI quantification of glycosaminoglycan (GAG) con- the transitional zone (TZ) where the fibers are randomly
centration in articular cartilage at microscopic resolution. oriented, and the radial zone (RZ) where the fibers are per-
Methods: T1-weighted 2D experiments of cartilage were fully pendicular to the tissue surface (2,3). Both collagen and
sampled in k-space with five inversion times at 17.6 mm reso- GAG contents in articular cartilage are responsible for the
lution. These fully sampled k-space data were re-processed, load-bearing property of cartilage (4–6). The reduction of
by undersampling at various 1D and 2D CS undersampling GAG will result in a poor mechanical response and can be
factors (UFs). The undersampled data were reconstructed indi- regarded as an early sign of the tissue degradation, which
vidually into 2D images using nonlinear reconstruction, which
eventually leads to arthritis (7–13).
were used to calculate 2D maps of T1 and GAG concentration.
The values of T1 and GAG in cartilage were evaluated at differ-
The relaxation parameters in MRI have been used to
ent UFs (up to 16, which used 6.25% of the data). K-space detect tissue degradation (14–22). In contrast to the
sampling pattern and zonal variations were also investigated. anisotropic and depth-dependent distributions of T2 and
Results: Using 2D variable density sampling pattern, the T1 T1r, T1 in healthy cartilage is mostly uniform across the
images at UFs up to eight preserved major visual information tissue depth and isotropic with respect to the specimen
and produced negligible artifacts. The GAG concentration orientation in the magnetic field (23–27). In the presence
remained accurate for different sub-tissue zones at various of gadolinium (Gd) contrast agents (28–30), T1 can
UFs. The variation of the mean GAG concentration through the become sensitive to the GAG concentration in cartilage
whole tissue depth was 1.20%, compared to the fully sampled
((6),(31–33)). Quantitative mapping of the GAG concen-
results. The maximum variation was 2.24% in the deep zone
tration can be achieved by acquiring two T1 images: T1
of cartilage. Using 1D variable density sampling pattern, the
quantitative T1 mapping and GAG concentration at UFs up to before Gd administration (T1b) and T1 after Gd adminis-
4 showed negligible variations. tration (T1a) (6,34).
Conclusion: This study demonstrates that CS could be benefi- Because quantitative T1 measurements commonly use
cial in microscopic MRI (mMRI) studies of cartilage by acquiring the inversion recovery sequence, acquisition time can be
less data, without losing significant accuracy in the quantification long (29,35). Any method to accelerate quantitative T1
of GAG concentration. Magn Reson Med 000:000–000, 2017. experiments is, therefore, highly desirable. In recent
C
V 2017 International Society for Magnetic Resonance in years, compressed sensing (CS) has emerged as a new
Medicine. framework that can accelerate image acquisition (36–40).
Key words: compressed sensing; MRI; cartilage; anisotropy; The basic CS theory relies on the inherent sparsity and
GAG; T1 compressibility of MR data, which allows images to be
recovered from randomly undersampled k-space data
using a nonlinear reconstruction algorithm to overcome
undersampling-induced artifacts (36,39). The application
of CS to accelerate T1 mapping in the quantification of
INTRODUCTION GAG concentration in cartilage has not yet been investi-
gated. This study aimed to examine the potentials of CS
Articular cartilage is a thin layer of load-bearing tissue in T1 mapping of cartilage by re-processing the fully
that covers the bones in synovial joints. The main extra- sampled T1 data from microscopic MRI (mMRI), at both
cellular components of cartilage are water, collagen 1D and 2D variable density patterns and at different
fibers, and negatively charged glycosaminoglycans (GAG) undersampling factors (UFs). The quantification of GAG
(1). The orientation of the collagen fibers varies along in cartilage was used as the criteria for the feasibility
the thickness of cartilage, which is commonly subdi- investigation (6,41).
vided into three histological zones: the superficial zone
1
Center for In Vivo Microscopy, Department of Radiology, Duke University, METHODS
Durham, North Carolina 27710, USA.
2
Department of Physics and Center for Biomedical Research, Oakland Specimen preparation
University, Rochester, Michigan 48309, USA.
Grant sponsor: NIH; Grant numbers: (PI: Xia) R01 AR052353; R01 AR069047. Humeral heads were harvested shortly after the sacrifice
*Correspondence to: Yang Xia, PhD, Department of Physics, Oakland University, of mature and healthy canines that were used for an
244 Meadow Brook Road, Rochester, MI 48309, USA. E-mail: xia@oakland.edu unrelated research, which were approved by the Institu-
Received 3 July 2017; revised 6 September 2017; accepted 26 September tional Animal Care and Use Committee (IACUC). The
2017
imaging specimen was about 3.5 2.5 6 mm in size and
DOI 10.1002/mrm.26973
Published online 00 Month 2017 in Wiley Online Library (wileyonlinelibrary. contained the full-thickness cartilage still attached to the
com). underlying bone (42). The specimens were soaked in
C 2017 International Society for Magnetic Resonance in Medicine
V 1
2 Wang et al.
physiological saline with 1% protease inhibitor (Sigma, f ðxÞ ¼ jjFx yjj22 þ l1 jjCxjj1 þ l2 TV ðxÞ; [2]
St. Louis, MO). The specimens were never frozen.
where x is the image and y is its corresponding k-space,
mMRI protocols F is the FFT, C is the sparse transform, l1 and l2 are
weighting factors, and TV is the total variation. In this
All experiments were performed at room temperature on study, l1 equals 0.006 for the sparse solution, and l2
a Bruker AVANCE II 300 NMR spectrometer (Billerica, equals 0.0012 for the data consistency. Various CS
MA), equipped with a 7T/89 mm vertical-bore magnet undersampling factors (UF ¼ 1, 2, 4, 8, 16, where 1
and microimaging accessory. The 2D spin-echo imaging stands for the fully sampled data and 16 stands for using
experiments were carried out with an acquisition matrix 1/16 of the fully sampled data) were used to assess the
of 256 128 (that was post-reconstructed into a 256 accuracy for quantitative GAG concentration in the
256 matrix) and a single-slice thickness of 1 mm. The cartilage.
field of view was 0.45 m 0.45 cm, resulting in the 2D
in-plane pixel size of 17.6 mm. The repetition time was 2 GAG quantification by dGEMRIC method
s without Gd immersion and 0.8 s with Gd immersion
Quantitatively, the GAG concentration can be calculated
(30). from the T1 images by a set of three equations based on
Quantitative 2D T1 imaging experiments were carried
the Donnan equilibrium theory, which have been docu-
out at 55 with respect to the B0, and followed the previ- mented extensively in the literature (28,41,45). The fully
ously established protocols (23,30,34). Briefly, T1 con- sampled GAG data was considered the “ground truth”,
trast used an inversion recovery magnetization-prepared because they have been correlated with a number of mul-
sequence, with five inversion points (0, 0.4, 1.1, 2.2, tidisciplinary imaging and non-imaging techniques (41).
4.0 s) before Gd solution soaking, and with five inversion The error in the GAG concentration by CS method was
points (0, 0.1, 0.3, 0.5, 1.0 s) after immersing in the Gd calculated based on the GAG concentration map from
solution. The scan time for a T1 mapping without and the fully sampled data:
with Gd immersion was 8 h and 2 h, respectively
(because of the long delays in the inversion recovery). T1 ½GAGerror i ¼ 100 ðj½GAGcs i ½GAGfull i j=½GAGfull i Þ; [3]
mapping of cartilage tissue was calculated by a single-
component fit on a pixel-by-pixel basis using MATLAB where [GAG]full is the GAG concentration calculated
(The MathWorks, Natick, MA). from the fully sampled data set, [GAG]cs is the GAG con-
centration calculated from various undersampled data
Compressed sensing sampling sets, and i stands for the different histological zones: SZ,
TZ, and RZ.
The varied density k-space sampling pattern (SPk) gener-
ated by a probability density function (PDF) for CS was RESULTS
determined by parameters pa and pb using the equation
(43,44): Optimize CS sampling patterns
FIG. 1. The reconstructed 2D T1-weighted images using three different 2D sampling patterns (a–d) at an undersampling factor of 4, at
two different inversion times, (i–p, the inversion time of 0.0 s; q–x: the inversion time of 1.1s, which had lower signal-to-noise ratio). The
fully sampled images were also shown as the ground truth in (i and q). The point spread function images for different sampling patterns
were shown in (e–h). The quality of the reconstructed T1-weighted images was sensitive to the sampling patterns. The optimized sam-
pling pattern was illustrated in (b) with pa ¼ 1.8, and pb ¼ 3.6. SZ, superficial zone; TZ, transitional zone; URZ upper radial zone; LRZ
Lower radial zone. “10X” in the figure label means that the display scale (the up limit) has been reduced to 1/10 to show more clearly
the differences between the ground truth and CS reconstructed images.
4 Wang et al.
FIG. 2. The optimized k-space sampling patterns at different undersampling factors (UF), quantitative T1b maps, and the corresponding
error maps from the fully sampled k-space data (a–c, 1D k-space sampling patterns; d–f, 2D k-space sampling patterns). The recon-
structed T1b maps are found to be visually comparable with the references at UF up to 4 (1D patterns) and 8 (2D patterns), with major
information qualitatively preserved and negligible artifacts, whereas the reconstructed relaxation maps have worse quality at UF of
8 and 16 (1D patterns) and 16 (2D patterns). T1b, T1 mapping before Gd administration.
data (UF ¼ 1), as well as the undersampled data with UF of some extent, with exhibition of spatial blurring (the T1a
2, 4, 8, and 16. Using the 1D k-space SPk, the qualities of and their error maps can be found in Supporting Fig. S1,
the reconstructed T1b images were visually comparable which have similar features).
with the ground truth at UF up to 4, whereas the qualities
of the constructed images became visibly inferior at UFs of
T1 profiles and mean GAG concentration
8 and 16. It is interesting to note that a much higher error
was found within the bone region (white arrows) when Quantitative depth-dependent profiles of T1a, T1b (Fig. 3a–
compared to the limited error in the cartilage area (yellow 3b), GAG concentration (Fig. 3c–3d), and mean GAG con-
arrows). Using the 2D k-space SPk, the reconstructed T1b centration values at different UF (Fig. 3e–3f) are illustrated.
images were found to be visually comparable with the Several conclusions can be reached from this set of data.
references (the fully sampled results) at UF up to 8, with First, T1b values are always higher than T1a values through
major information qualitatively preserved and negligible the whole tissue depth, regardless of 1D or 2D UFs. T1a
artifacts. At 2D UF of 16, the image quality diminished to profiles showed strong depth-dependent properties
Compressed Sensing in mMRI of Articular Cartilage 5
FIG. 3. Quantitative T1b and T1a profiles (a, b), GAG concentration profiles (c, d), and the mean GAG concentration values (e, f) at differ-
ent undersampling factors (1, 2, 4, 8, 16) from articular surface (0 mm) to cartilage-bone interface (640 mm). The left half and the right
half of the figure used 2D and 1D undersampling patterns, respectively. T1b, T1a, and the GAG concentration profiles using compressed
sensing were very consistent with the fully sampled data using 2D undersampling pattern, whereas the profiles varied significantly at 1D
undersampling factors of 8 and 16 (arrows). Little variation of the mean GAG concentration value was found even at a 2D undersampling factor
of 16 (e), whereas bigger variations were noticeable at 1D undersampling factors of 8 and 16.
throughout the entire cartilage region. In contrast, this factors (1, 2, 4, 8, 16). Because the spatial resolution of carti-
depth-dependent appearance is much weaker before Gd lage usually is much coarser in clinical MRI, the mMRI carti-
administration. These observations were consistent with lage data was divided to four sub-tissue structural zones: SZ,
our previous findings (30,34). Second, the GAG profiles TZ, upper RZ (URZ), and lower RZ (LRZ) to investigate the
also showed a strong depth-dependence profile: lower at GAG concentration variations in these zones at different UFs
the surface zone and monotonically increased to the deep and both 1D and 2D patterns. The variation of T1a, T1b, and
zone of the tissue. The T1b, T1a, and the GAG concentra- GAG concentration at each sub tissue zones was found to be
tion profiles by different CS reconstructions were consis- small even at 2D UF of 8 or 16, at 1D UF of 4. The maximum
tent with the fully sampled data even at 2D UF of 16, variation was found at the LRZ with 2.24% difference from
while the profiles varied significantly at 1D UF of 8 and 16 the ground truth at 2D UF of 16, whereas the maximum varia-
(arrows). Finally, little variation of the mean values in the tion was found at the LRZ with 14.18% at 1D UF of 16. A
GAG concentration was found at the 2D undersampling detailed comparison of the GAG concentrations was summa-
SPk of 16 (Fig. 3e), whereas the variations were larger at rized in Table 1 and Supporting Figure S2.
the 1D compressed sensing SPk of 8 and 16 (Fig. 3f).
DISCUSSION
Quantitative T1 and GAG concentration in sub zones
It is rare that one has access to the original k-space data from
Figure 4 showed the zonal changes of T1b, T1a, and GAG con- quantitative 2D T1 experiments of cartilage at 17.6 mm reso-
centration in articular cartilage at various undersampling lution, and also know the statistical correlation between
6 Wang et al.
FIG. 4. Minimum variations in the T1b (a–b), T1a (c–d) and GAG concentration (e–f) in the three histological zones of articular cartilage at vari-
ous 1D (right) and 2D (left) undersampling factors (1, 2, 4, 8, 16). Little variation of T1a, T1b, and GAG concentration at any sub-tissue zones
was found even at a 2D undersampling of 16. SZ, superficial zone; TZ, transitional zone; URZ, upper radial zone; LRZ, lower radial zone.
these mMRI GAG data and the biochemical GAG quantifica- the quantitative T1 studies of cartilage to reduce the acquisi-
tion based on the same specimen (6,41). This study demon- tion time. An undersampling factor of 16 (i.e., using only
strates that, at high spatial resolution, CS can be applied to 6.25% of the data) could be achieved when a 2D sampling
Table 1
GAG Concentrations at Different Zones With Different CS UF
Sampling UF 1 2 4 8 16
Patterns k-Space Use 100% 50% 25% 12.5% 6.25%
2D GAG (mg/ml) SZ 30.90 6 1.90 30.86 6 1.91 30.92 6 1.85 31.19 6 1.87 30.63 6 1.45
(% error) (0.0%) (0.13%) (0.06%) (0.93%) (0.87%)
GAG (mg/ml) TZ 56.44 6 2.94 56.65 6 2.83 56.64 6 2.76 55.89 6 2.79 56.84 6 2.80
(% error) (0.0%) (0.37%) (0.37%) (0.97%) (0.71%)
GAG (mg/ml) URZ 79.93 6 2.31 80.59 6 2.25 80.71 6 2.58 81.11 6 2.44 80.47 6 2.26
(% error) (0.0%) (0.83%) (0.97%) (1.47%) (0.68%)
GAG (mg/ml) LRZ 108.63 6 4.00 109.56 6 5.42 110.07 6 4.48 109.05 6 5.74 111.06 6 5.21
(% error) (0.0%) (0.86%) (1.33%) (0.39%) (2.24%)
1D GAG (mg/ml) SZ 30.90 6 1.90 30.92 6 1.95 31.08 6 1.88 31.73 6 1.84 32.42 6 1.41
(% error) (0.0%) (0.06%) (0.58%) (2.69%) (4.92%)
GAG (mg/ml) TZ 56.44 6 2.94 56.73 6 2.98 56.17 6 2.65 54.48 6 2.56 52.06 6 3.13
(% error) (0.0%) (0.51%) (0.48%) (3.47%) (7.76%)
GAG (mg/ml) URZ 79.93 6 2.31 80.27 6 2.54 79.42 6 2.51 81.94 6 2.14 77.21 6 2.54
(% error) (0.0%) (0.43%) (0.64%) (2.51%) (3.41%)
GAG (mg/ml) LRZ 108.63 6 4.00 109.77 6 5.13 109.99 6 4.16 101.37 6 5.98 124.03 6 5.54
(% error) (0.0%) (1.05%) (1.25%) (6.68%) (14.18%)
UF, undersampling factors; SZ, superficial zone; TZ, transitional zone; URZ, upper radial zone; LRZ, lower radial zone.
Compressed Sensing in mMRI of Articular Cartilage 7
pattern is used, without losing significant accuracy in the of bone and joint (46), because the lower resolution and
GAG quantification in cartilage, based on the zonal analysis. the higher partial volume effect may cause larger errors
for T1 or GAG concentration quantification in the RZ of
Effect of sampling pattern cartilage (close to the bone and cartilage interface).
Although equidistant k-space undersampling and recon-
Difference between undersampling factor and reduction
struction by zero-filling results in coherent aliasing, ran-
of data acquisition time
dom k-space undersampling exhibits incoherent artifacts
that behave much like additive random noise. Based on It should be pointed out that the reduction of the data
Eq. [1], 2D variable density random undersampling in acquisition time at different undersampling factors
Cartesian imaging has been proposed, which was used in depends on several experimental features in imaging,
this study. As shown in Figure 1, variable density sam- including the patterns of k-space trajectory and the
pling combines with denser sampling near the center of dimensions of the imaging experiments. In a typical 2D
k-space, matching the energy distribution in k-space imaging using the Cartesian coordinate, the overall
(concentrated close to the center of k-space and rapidly experimental time is limited by the repetition time in
decaying toward the periphery). The different combina- MRI. Because k-space sampling at the read direction is
tions of pa and pb were further tested using point spread carried out quickly, there is little advantage to undersam-
function (Fig. 1). For example, the SPk (UF ¼ 4) was opti- ple the read dimension. The reduction of the scan time
mized with pa ¼ 1.8 and pb ¼ 3.6, and the reconstructed can be achieved by omitting the collection of individual
T1 images preserved major information with few arti- k-space lines in the phase encoding direction. We show
facts. This optimized SPk are likely to differ for different in this study that a 1D varied density sampling pattern
studies, tissues, MRI parameters, and k-space features, can be used in the 2D imaging to achieve a factor of 4
which call for the extra caution in CS MRI experiments. time saving without introducing noticeable error (Fig. 2).
Furthermore, because 2D sparsity is fully exploited using The use of 2D variable density pattern in 2D Cartesian
2D variable density SPk, the images have a sparser repre- imaging would not be beneficial, because most phase
sentation, therefore can achieve to a higher undersam- encoding directions cannot be omitted because of the
pling factor (UF of 16) than the 1D variable density SPk remaining acquisition points (36). For any 3D imaging,
(UF up to 4), without introducing major deviation to the much significant time saving can be achieved when a 2D
quantification T1 and GAG values. variable density sampling pattern is applied to the two
phase-encoding directions (47). The design of novel k-
Effect of Gd administration space trajectory pattern can facilitate further reduction
Paramagnetic Gd ions can reduce the MRI relaxation on the data acquisition time. In addition, 3D T1 imaging
times and enhance the MRI image contrast. Whether or are time-consuming, which warrants the evaluation of
not Gd administration would affect the CS reconstruc- the GAG concentration using 2D undersampling patterns.
tion has not been investigated thoroughly. In this study, In conclusion, to the best of our knowledge, this is the
the sequences of the 2D T1-weighted images and their first study that demonstrates the feasibility of imple-
calculated T1 mappings maps (prior and post Gd admin- menting CS in mMRI quantification of GAG. We reveal
istration) at various undersampling factors were calcu- the challenges of using CS for quantitative imaging of
lated. No apparent reduction in image quality was found cartilage, especially in the deep RZ of cartilage and the
in T1-weighted images, in comparison with the ground interface between cartilage and bone. We show that 1D
truth for different 2D undersampling factors and each and 2D sampling patterns can achieve different time sav-
sub regions of the cartilage, once proper k-space sam- ings. The calculated GAG concentration did not exhibit
pling patterns were used. major deviation in quantification, even at high under-
sampling factors and at different sub tissue zones (SZ,
Effect of undersampling factor TZ, and RZ). This significant undersampling could
potentially be translated into major reduction in data
Although both T1a and T1b maps exhibited qualitatively
acquisition time, which would be extremely beneficial to
good quality even at 2D UF of 16 (using only 6.25% k-
any ex vivo study of cartilage by MRI. The time saved
space data), the images began to blur at high undersam-
can be used to increase the sample size, to map the topo-
pling factors, which may be caused by the significant
graphical variation of the tissue parameters over a joint
reduction of high frequency components in k-space,
surface, and to acquire better quality data.
hence, making it more difficult to recover the fine infor-
mation of the image. Compared to the relatively robust
ACKNOWLEDGMENTS
reconstruction in the cartilage area, the bone area
showed much larger errors (Figs. 2 and 3). This can be Y.X. is grateful to the National Institutes of Health for
attributed to the lower signal-to-noise ratio in the bone grants (R01 AR052353; R01 AR069047). The Center for
area of the T1-weighted images, where all five inversion In Vivo Microscopy is supported through NIH (award
times were used in the exponential fitting in the calcula- P41 EB015897 to Dr. G Allan Johnson). The authors
tion of the T1 maps. In addition, the bone region has thank Drs. Cliff Les and Hani Sabbah (Henry Ford Hospi-
more random structure and intensity than the highly tal, Detroit) for providing the canine specimens, and Ms.
structured cartilage, hence, requiring more caution in the Carol Searight (Department of Physics, Oakland Univer-
CS reconstruction. It may become severe in clinical MRI sity) for editorial comments on the manuscript.
8 Wang et al.