Click here to load reader
Click here to load reader
Accepted Manuscript
Faster Imaging with a Portable Unilateral NMR Device
Asaf Liberman, Elad Bergman, Yifat Sarda, Uri Nevo
PII: S1090-7807(13)00084-0
DOI: http://dx.doi.org/10.1016/j.jmr.2013.03.009
Reference: YJMRE 5190
To appear in: Journal of Magnetic Resonance
Received Date: 24 December 2012
Revised Date: 18 March 2013
Please cite this article as: A. Liberman, E. Bergman, Y. Sarda, U. Nevo, Faster Imaging with a Portable Unilateral
NMR Device, Journal of Magnetic Resonance (2013), doi: http://dx.doi.org/10.1016/j.jmr.2013.03.009
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Faster Imaging with a Portable Unilateral NMR Device
Asaf Liberman, Elad Bergman, Yifat Sarda, and Uri Nevo∆
The Iby and Aladar Fleischman Faculty of Engineering, Tel Aviv University, Tel Aviv 69978, Israel ∆Correspondence to: [email protected]
March 2013
Keywords: Fast imaging; Unilateral NMR; Compressed sensing; Desktop NMR;
Portable NMR; Fast spin echo.
Abstract
Unilateral NMR devices are important tools in various applications such as non-
destructive testing and well logging, but are not applied routinely for imaging,
primarily because B0 inhomogeneity in these scanners leads to a relatively low signal
and requires use of the slow single point imaging scan scheme. Enabling high quality,
fast imaging could make this affordable and portable technology practical for various
imaging applications as well as for new applications that are not yet feasible with
MRI technology.
The goal of this work was to improve imaging times in a portable unilateral NMR
scanner. Both Compressed Sensing and Fast Spin Echo were modified and applied to
fit the unique characteristics of a unilateral device. Two printed phantoms, allowing
high resolution images, were scanned with both methods and compared to a
standard scan and to a low pass scan to evaluate performance. Both methods were
found to be feasible with a unilateral device, proving ways to accelerate single point
imaging in such scanners. This outcome encourages us to explore how to further
accelerate imaging times in unilateral NMR devices so that this technology might
become clinically applicable in the future.
Introduction
Imaging with Open NMR Scanners
Unilateral portable NMR [1-4] devices are used mainly for non-destructive testing
applications where imaging capabilities are not critical. Other applications of low-
NMR and open architecture NMR scanners include oil well logging, food analysis and
quality control [5-8], elastomer quality control [5, 9, 10], cultural heritage [11], and
skin and tendon profiling [12, 13].
Slice selective lateral imaging with open NMR scanners is challenging due to the
limitations of their geometry [14]. The open geometry generates an inhomogeneous
magnetic field with a strong, constant gradient defining non-flat and extremely
thin slices, restricting the acquired signal. A sensitive volume that is sufficiently flat
for imaging can be generated only at a defined distance from the magnet, severely
restricting the device's penetration depth. The constant gradient also leads to a
significant attenuation of the diffusion-weighted signal. Frequency encoding cannot
be utilized in the device under static field conditions, which introduce an
overwhelming read encoding gradient in the direction and cause rapid dephasing
of the signal [15].
2D lateral imaging can thus be done by applying a pure spin echo phase encoding
in both planar directions (also known as Single Point Imaging, SPI), as implemented
by Casanova et al. on the NMR-MOUSE [16]. This procedure limits imaging since scan
times with SPI scale as (for an sampling matrix).
To improve the Signal to Noise Ratio (SNR), Perlo et al. developed a CPMG-like
pulse sequence where a train of nominal pulses generates a train of echoes,
which are accumulated [17]. Introducing such a sequence in the presence of a strong
static magnetic gradient and grossly inhomogeneous and fields causes a
severe distortion in the phase encoding, so that one of the components of the echo
signal goes to zero after a transient period. The proposed solution is to obtain the
full complex signal in two experiments, by phase-cycling the RF refocusing pulse in
two orthogonal directions to acquire a different complex component in each step
[17].
This work demonstrates the application of two methods, Compressed Sensing
(CS) and Fast Spin Echo (FSE) for the purpose of accelerating imaging on a unilateral
low field NMR scanner. Both methods can overcome the basic restriction posed by
SPI: CS allows a subset of the k-space to be read while ensuring a high-quality
reconstruction; FSE enables more than one coefficient to be read following a single
excitation. By overcoming the limitations imposed by SPI, the minimum scan times
previously thought to be required with such a portable unilateral NMR device can be
reduced.
Theory
Compressed Sensing (CS)
Compressed Sensing is a novel acquisition and reconstruction scheme that uses
the sparsity of natural signals together with a non-linear algorithm to provide a high
quality reconstruction of a signal with significantly low sampling rates/percentages.
CS dictates an incoherent sub-sampling of the signal while demanding a
reconstruction based on norm error minimization between the initially sub-
sampled, zero filled image and the reconstructed image, both represented in a
sparse domain. The sub-sampling of the signal described allows for scan time
reduction with little compromise to the image quality.
As CS complements the scanning regime of MRI, combining the two has been
heavily researched in recent years. Lustig et al. have implemented CS in a clinical MRI
device to undersample the k-space, thus reducing scan times (or, interchangeably, to
improve scan resolution) [18]. Parasoglou et al. have combined CS with a single point
imaging scanning scheme to more rapidly image dynamic processes with short
[19].
Fast Spin Echo (FSE)
Fast Spin Echo is an imaging method engineered to acquire more than one k-line
following a single excitation [20]. FSE accelerates the basic spin echo imaging by
applying a CPMG sequence with an additional phase encoding gradient prior to each
echo. Each echo thus adds an additional k-line, leading to the acquisition of
significant segments of the k-space following a single excitation. In cases where is
long enough, full coverage of the k-space can be achieved in one shot (i.e., following
a single pulse) [21].
Materials and Methods
Hardware
NMR-scanner
The NMR-MOUSE scanner (ACT GmbH , Aachen, Germany) was used in this work.
This scanner has a permanent magnet combined with a mounted RF coil system and
gradient coils. The magnet is composed of two permanent rectangular blocks set on
an iron yoke. This setup provides a static magnetic field of about at a distance
of from the magnet (Larmor frequency of ), with a strong
gradient of along the z-direction, away from the face of the magnet. Combining
the strong gradient with hard RF pulses produces selective excitation of thin flat
slices.
An LVC-7700 amplifier (AE Techron Inc., Elkhart, IN, USA) was used to power the
gradient coils and generate pulsed gradients along the and directions. A surface
RF coil is positioned on top of the magnet and gradient coils, and was used to excite
and detect the NMR signal. A spectrometer (Magritek LTD, Wellington, New
Zealand) controls the operation of the gradient coils and the RF coil (input/output).
Phantoms
Two circular-shaped phantoms were used in this work: a phantom containing two
zero shapes and two curved lines and a phantom containing several rectangles and
several circles (hereinafter "phantom A" and "phantom B", respectively) (Fig. 1). The
phantoms contain fine features with width as low as , and a size less than
across, to be scanned with an -diameter coil. The phantoms were filled
with glycerol, a suitable material for testing in a unilateral NMR scanner due to its
short and times and high viscosity, which leads to low self-diffusion.
Fig. 1: Phantom diagrams. (a) Phantom A. Features are relatively scattered and
have different shapes and orientations to test the quality of reconstruction. (b)
Phantom B. Features are concentrated near the center of the FOV. Smallest feature
width is about .
Implementation of CS in a unilateral scanner
The Daubechies-6 wavelet is used as the sparsifying transform for the
reconstructions. The algorithms for the CS acquisition and reconstruction were
based on the work of Lustig et al. as described in [18], and using the supplementary
freeware codes attached (http://www.eecs.berkeley.edu/~mlustig/Software.html).
Since the sampled domain in MRI is the spatial frequency domain, most of the
information in scans of natural images will be concentrated in the center of the k-
space. Thus, by using a probability density function (PDF) that decreases gradually
according to a power of distance from the center with a predetermined full sampling
radius, a higher concentration of center sampling is prioritized, while simultaneously
assuring the incoherence required for the CS acquisition [18].
Implementation of an FSE-like sequence in a unilateral scanner
As opposed to the implementation of FSE in common MRI, the prototype FSE-like
sequence implemented in a unilateral NMR scanner in this work scans two k-space
coefficients following a single excitation (instead of several k-space lines as in
common MRI systems). (We limited the number of coefficients per excitation to
accumulate a sufficient signal per k-space coefficient, given the low signal
amplitude). Phase encoding in FSE is generated by addition of a gradient blip (See
Fig. 2 for a scheme of the pulse sequence). Time to echo (TE) is usually kept at a
minimum to maximize the generated number of echoes [17], nevertheless, to avoid
gradient overshoots and eddy currents, TE was extended in these experiments. It
should be noted here that attempting to extend the TE only for the application of the
blipped gradient pulses results in distortion of the echo train after the pulses, due to
the increased influence of stimulated echoes.
Within the echo train ( echoes), the first coefficient was encoded based on the
integration of the first echoes, and the second coefficient on the remaining
echoes (omitting from integrations the two echo periods used for
encoding of the second coefficient). The exact value of was calculated by
demanding equal energy for both parts of the FID, given the apparent (found in a
preliminary measurement).
Fig. 2: The Fast Spin Echo pulse sequence in a unilateral NMR. The sequence is
divided into four periods: the first encoding period, the first CPMG period, the
second encoding period (gradient blips), and the second CPMG period. The first
period is used to encode the phase of the initial coefficient using gradients employed
in the x and y directions (The encoding time to echo, TEe, is indicated). The second
period is used to accumulate data to be read as the first k-space coefficient with a
series of pulses (The detection time to echo, TEd is indicated). The third period is
used to shift phase to encode the next coefficient using gradient blips during two TE
periods. The fourth period is used to accumulate data to be read as the second k-
space coefficient with a series of pulses.
As noted, the CPMG-like sequence in the unilateral scanner nulls one of the
coefficients of the complex signal after a transient period, prompting the use of a 2-
step phase cycling scheme (repetition of the sequence in two orthogonal directions)
to read both components of the magnetization (real and imaginary) [17]. Since in FSE
two k-space coefficients are read, an extended 4-step phase cycling scheme is
employed to gather all four components of magnetization (see Table 1 and Fig. 3 for
the phase cycling scheme and extraction of the coefficients, respectively). By
alternating the post-blip orthogonal direction of the refocusing RF pulses, all four
combinations of the products of the pre-blip and post-blip phases (complex
components) are encoded. Using trigonometric relations between the encoded
products, the accumulated phases are calculated and the second k-space coefficient
is extracted.
Fig. 3: Recovery of complex coefficient values using the phase cycling scheme. The
real component of the first coefficient is obtained by applying pulses and the
imaginary component is obtained by applying pulses. In addition to this an
additional cycling is applied for the pulses after the gradient 'blip' ( and ).
Altogether the four combinations of acquired components result in reconstruction of
two complex k-space coefficients.
Experiments
Feasibility and performance of CS in a unilateral NMR scanner
To validate the real-time feasibility of CS scans with a low-SNR unilateral NMR and
to compare its performance to that of a basic undersampling method, both
phantoms were scanned with a 32 X 32 matrix (FOV of , spatial
resolution of ). Five CS scans were performed in real time, with sampling
percentages of 30%, 40%, 50%, 60%, and 70% and full sampling radii of 0.41, 0.48,
0.55, 0.6, and 0.67 respectively. CS scanning schemes for phantom A are shown in
Fig. 4l-p. Scans for phantom A were performed with parameters ,
with 350 echoes acquired, and
, which resulted in a total experimental time of 48
minutes per image. Scans for phantom B were performed with similar parameters,
except for and 500 echoes acquired, which resulted in a total
experimental time of 34 minutes per image. Times for CS scans can be calculated
accordingly, with the shortest CS scan for phantom B being the one with a sampling
percentage of 30%, which lasted 10 minutes. Full sampling radii were chosen based
on a preliminary experiment, which demonstrated an advantage for scans acquired
with a high (yet not maximal) full sampling radius.
Low Pass (LP), a basic undersampling scheme, was used for comparison. Each CS
scan was compared to an LP scan with an equal sampling percentage (and with equal
scan duration), similar to approaches in previous works [18]. All LP scans were
performed by extracting a square-shaped subset of the coefficients from the center
of a single fully scanned k-space. All scans were performed with two averaging
repetitions (each includes phase cycling, resulting in four excitations per experiment)
and compared via RMS error (RMSE) to a fully scanned k-space with 12 averaging
repetitions. RMSE is given by the following equation,
(1.1)
where is the reconstructed image, and is the gold standard reference
image.
Feasibility and performance of an FSE-like sequence in a unilateral NMR scanner
To validate the feasibility of the FSE-like sequence in a unilateral device, both
phantoms were scanned with a pixel matrix. The phantoms were scanned
twice: a standard full sampling scan with 24 averaging repetitions and an FSE scan
with 48 averaging repetitions (lasting the same amount of time and acquiring the
same number of total echoes as the standard scan). Scans for phantom A were
performed with parameters , to allow for blip gradients
with duration of to be inserted, with 200 echoes acquired,
and . Scans for
phantom B were performed with similar parameters except for and
450 echoes acquired. Images are presented in the results section.
The initial accuracy of the FSE method was verified by joining two k-spaces
scanned in opposite spatial directions resulting in the entire k-space being composed
of post-blip coefficients. The k-space was transformed into an image that was
validated relative to a standard image.
Results
Feasibility and performance of CS in a unilateral NMR scanner
Figure 4 presents the imaging results of phantom A. Images clearly show an
improvement in reconstruction with a higher sampling percentage. It is also evident
CS is superior to LP sampling, which blurs the image by cutting high spatial frequency
information (Fig. 4b-d), effectively causing a loss of sharpness. CS undersampling, on
the other hand, better preserves even small details and contrast, and reduces the
amount of visible noise in the imaged object (Fig. 4g-i).
Sampling 50% (or more) of the coefficients with CS (Fig. 4i-k) is sufficient to
recover most of the fine features of the 100% sampled image. On the other hand, LP
sampling of 50%-60% of the coefficients yields mostly recognizable features, but
with blurry edges and lower contrast. Sampling of 70% of the coefficients with both
methods produces good images. RMS error results are lower for CS sampling
throughout the experiment relative to the gold standard scan (Fig. 4a).
Fig. 4: Results of the experiment comparing CS and LP reconstructions of phantom
A with similar scan percentages. (a) 100% sampled image with 12 averaging
repetitions. (b-f) Low pass reconstructed images: sampling of 30%, 40%, 50%, 60%,
and 70% of the coefficients respectively. (g-k) Compressed sensing reconstructed
images: sampling of 30%, 40%, 50%, 60%, and 70% of the coefficients respectively.
RMS error relative to the high-averaged, 100% sampled image is indicated below
each image. (l-p) Scanning masks for CS scans: white pixels were scanned while black
pixels were skipped.
Figure 5 presents the phantom B imaging results. Both CS and LP lead to
reconstructed images that improve as the sampling percentage increases. CS
reconstructions are consistently less noisy, are sharper, and are more detailed (Fig.
5g-k).
Fine features in phantom B are already clear and sharp with only 30% of the
coefficients sampled with a CS acquisition (Fig. 5g), while LP sampling yields
identifiable features that are very blurry (Fig. 5b-d). It is important to note that
above 50% sampling (Fig. 5i-k) there is almost no improvement in feature sharpness
(although block artifacts are visible in lower sampling), indicating that there is no
need to sample more than 50% of the coefficients in such cases. A difference is
noticeable between the 50%-and-above scans with LP sampling (Fig. 5d-f), where the
boundaries are sharper, but the images are noisier.
RMS error results are lower for CS sampling throughout the experiment relative
to the gold standard scan (Fig. 5a).
Fig. 5: Results of the experiment comparing CS and LP reconstructions of phantom
B with a similar scan percentage. (a) 100% sampled image with 12 averaging
repetitions. (b-f) Low pass reconstructed images: sampling of 30%, 40%, 50%, 60%,
and 70% of the coefficients respectively. (g-k) Compressed sensing reconstructed
images: sampling of 30%, 40%, 50%, 60%, and 70% of the coefficients respectively.
RMS error relative to the high-averaged, 100% sampled image is indicated below
each image.
The fidelity provided by CS is further demonstrated by the one-dimensional cross
section profiles of two representative lines (lines 9 and 18) from the 30% and 50%
sampling reconstructions displayed in Figure 6. CS reconstructions (darker grey,
triangle markers, dash-dotted line) are indeed closer to the gold standard
reconstruction (black) relative to the LP reconstructions (lighter grey, circular
markers, dashed line). Edges in CS reconstructions are steeper and better preserved.
As expected, cross sections of images scanned with 50% of the coefficients are closer
to the gold standard (fig. 6b, d) image, relative to images scanned with 30% of the
coefficients (fig. 6a, c). (Note that most, but not all, one-dimensional profiles show
the superior quality of the CS over the LP images).
Fig. 6: Cross section results for the experiment comparing CS and LP phantom B
reconstructions with a similar scan percentage for two representative lines (the 9th
and 18th
). (a) 100% sampled image with 12 averaging repetitions. (b-c) Cross section
results for the 9th
line for GS-, CS-, and LP-reconstructed images with 30% and 50% of
the coefficients, respectively. The solid black line with rectangular markers
represents the GS; the dashed light grey line with circular markers represents the LP;
and the dark grey dash-dotted line with triangular markers represents the CS. (d-e)
Cross section results for the 18th
line for GS-, CS-, and LP-reconstructed images with
30% and 50% of the coefficients, respectively.
Feasibility and performance of FSE in a unilateral NMR scanner
Figure 7 presents the imaging results for phantoms A and B. An FSE scan with 48
averaging repetitions (Fig. 7b for phantom A, Fig. 7d for phantom B) is compared to a
standard scan with 24 averaging repetitions (Fig. 7a for phantom A, Fig. 7c for
phantom B) resulting in an equal scan time. The total number of echoes acquired is
equal for these scans. Both images present high quality and are very similar,
although the FSE image has a higher relative noise level. FSE is thus feasible in a
unilateral NMR.
Fig. 7: FSE experiment results with phantoms A and B. (a) Standard sampling
image with 24 averaging repetitions for phantom A. (b) FSE sampling image with 48
averaging repetitions for phantom A. (c) Standard sampling image with 24 averaging
repetitions for phantom B. (d) FSE sampling image with 48 averaging repetitions for
phantom B.
Discussion
This work demonstrates the feasibility of using techniques to accelerate imaging
performed using a unilateral NMR device. The low SNR caused by the
inhomogeneous magnetic field is a limiting factor in a unilateral device. The high
number of averaging repetitions needed for an acceptable SNR and the inability to
apply lateral frequency encoding gradients force long scan times. Our goal here was
not to generate scan times sufficiently short for clinical or biological use, since these
depend on hardware and will improve over time, but to demonstrate that significant
acceleration of pure phase-encoding imaging times is feasible and to develop specific
ways to do so.
Compressed Sensing
While the pure phase-encoding scan scheme is the main drawback in a scan
performed with a unilateral NMR device, this scheme actually is beneficial for CS as it
enables freedom in the creation of the scanning trajectory. On the other hand, the
limitation it poses may be overcome by using CS to scan fewer of the k-space
coefficients, thus significantly reducing scan times.
It is important to note that the limitations of a unilateral NMR device still apply,
regardless of the offered improvements. The above characteristics (small FOV, small
sampling matrix, low SNR) together with the currently limited field of view leave
little headroom for reducing the sampling percentage as each reduction might
potentially cause distortion. For example, attempting to transform these images to
the wavelet domain and then disposing of the weakest coefficients (thus executing a
basic method of compression) fails when a low percentage of the coefficients is kept.
Fast Spin Echo
An FSE-like pulse sequence was already suggested and applied by Casanova et al.
[22] in 2003. By applying two gradient pulses with opposite polarization before and
after the echo formation, an independent phase encoding was performed on each
echo, leaving a zero phase shift, prior to the next echo. The method presented here
offers several improvements over this pulse sequence: (a) The power dissipated by
the gradients is minimized since the blip gradients are only used to shift the phase
(from one k-space coefficient to an adjacent one), as opposed to dephasing and
rephasing it entirely. (b) The sensitivity is greatly increased due to the nominal
pulse train which maximizes the number of echoes read during the experiment. This
is possible due to the significantly shorter TE time, which is possible since it must
contain only a gradient blip, rather than two full gradients and an acquisition period.
(c) The number of echoes dedicated to the acquisition of each coefficient is
calculated based on the of the materials, thus avoiding a weighted
acquisition.
The implementation of FSE in a unilateral NMR scanner required the use of a 4-
step phase cycle to extract the components of both complex coefficients. With the
current application, extension of the FSE scan to coefficients encoded in each
CPMG train will demand longer phase cycling schemes ( cycles for coefficients).
This extended scheme can, in principle, be reduced back to a 2-step scheme by, for
example, estimating the complex magnetization components of a primary set of
echoes, and then using this data as a priori information for estimating the secondary
set of echoes (see fig. 3). Currently, small errors that arise in the factors of such
estimation due to low sensitivity lead to severe image distortion. However, a
reconstruction scheme that improves sensitivity might enable the use of this
shortened phase cycling scheme. FSE should be further optimized with a higher
amplitude gradient blip that will minimize the length of TE.
Conclusions
Imaging in a unilateral NMR device with implementation of CS was demonstrated.
Even sampling of as low as 50% of the coefficients produced clean, sharp images. A
Fast Spin Echo-like sequence was also implemented in a unilateral NMR,
demonstrating the feasibility of further accelerating imaging time. Since FSE and CS
reduce imaging times in entirely different ways, a combination of the two techniques
is also possible.
Unilateral NMR devices are not yet feasible for biomedical imaging applications.
Sensitivity requires improvement, penetration depth should be increased, and
imaging times should be further reduced. With the above reported methods, small
objects with details of the order of can be scanned in ~10 minutes for a
pixel image. A combination of hardware and software improvements may
lead to a further reduction in imaging times.
Acknowledgments
This study was supported by an IRG grant (MMDTIAN) of the Marie Curie Foundation
(EU). UN acknowledges support by the Colton family scholarship. We wish to thank
Mr. Ezra Shaked, as well as members of the ACT and Magritek companies, for
continuous technical support.
References
[1] L. Backhouse, M. Dias, J.P. Gorce, J. Hadgraft, P.J. McDonald, J.W. Wiechers,
GARField magnetic resonance profiling of the ingress of model skin-care product
ingredients into human skin in vitro, J. Pharm. Sci., 93 (2004) 2274-2283.
[2] B. Blumich, P. Blumler, G. Eidmann, A. Guthausen, R. Haken, U. Schmitz, K. Saito,
G. Zimmer, The NMR-mouse: construction, excitation, and applications, Magn.
Reson. Imaging, 16 (1998) 479-484.
[3] G. Eidmann, R. Savelsberg, P. Blumler, B. Blumich, The NMR MOUSE, a mobile
universal surface explorer, J. Magn. Reson., Ser. A, 122 (1996) 104-109.
[4] P.M. Glover, P.S. Aptaker, J.R. Bowler, E. Ciampi, P.J. McDonald, A novel high-
gradient permanent magnet for the profiling of planar films and coatings, J. Magn.
Reson., 139 (1999) 90-97.
[5] B. Blumich, S. Anferova, K. Kremer, S. Sharma, V. Herrmann, A. Segre, Unilateral
nuclear magnetic resonance for quality control - The NMR-mouse, Spectroscopy, 18
(2003) 18-34.
[6] O.V. Petrov, J. Hay, I.V. Mastikhin, B.J. Balcom, Fat and moisture content
determination with unilateral NMR, Food Res. Int., 41 (2008) 758-764.
[7] M. Prammer, NMR in well logging and hydrocarbon exploration, Appl. Magn.
Reson., 25 (2004) 637-649.
[8] E. Veliyulin, I.V. Mastikhin, A.E. Marble, B.J. Balcom, Rapid determination of the
fat content in packaged dairy products by unilateral NMR, J. Sci. Food Agric., 88
(2008) 2563-2567.
[9] G. Zimmer, A. Guthausen, B. Blumich, Characterization of cross-link density in
technical elastomers by the NMR-MOUSE, Solid State Nucl. Magn. Reson., 12 (1998)
183-190.
[10] H. Kühn, M. Klein, A. Wiesmath, D.E. Demco, B. Blümich, J. Kelm, P.W. Gold, The
NMR-MOUSE®: quality control of elastomers, Magn. Reson. Imaging, 19 (2001) 497-
499.
[11] E. Del Federico, S.A. Centeno, C. Kehlet, P. Currier, D. Stockman, A. Jerschow,
Unilateral NMR applied to the conservation of works of art, Anal. Bioanal. Chem.,
396 (2010) 213-220.
[12] M.V. Landeghem, E. Danieli, J. Perlo, B. Blumich, F. Casanova, Low-gradient
single-sided NMR sensor for one-shot profiling of human skin, J. Magn. Reson., 215
(2012) 74-84.
[13] O. Miltner, A. Schwaiger, C. Schmidt, A. Bucker, C. Kolker, C.H. Siebert, K.W.
Zilkens, F.U. Niethard, B. Blumich, Portable NMR-MOUSE (R): a new method and its
evaluation of the Achilles tendon, Zeitschrift Fur Orthopadie Und Ihre Grenzgebiete,
141 (2003) 148-152.
[14] F. Balibanu, K. Hailu, R. Eymael, D.E. Demco, B. Blumich, Nuclear magnetic
resonance in inhomogeneous magnetic fields, J. Magn. Reson., 145 (2000) 246-258.
[15] P.J. Prado, B. Blumich, U. Schmitz, One-dimensional imaging with a palm-size
probe, J. Magn. Reson., 144 (2000) 200-206.
[16] F. Casanova, B. Blumich, Two-dimensional imaging with a single-sided NMR
probe, J. Magn. Reson., 163 (2003) 38-45.
[17] J. Perlo, F. Casanova, B. Blumich, 3D imaging with a single-sided sensor: an open
tomograph, J. Magn. Reson., 166 (2004) 228-235.
[18] M. Lustig, D. Donoho, J.M. Pauly, Sparse MRI: The application of compressed
sensing for rapid MR imaging, Magn. Reson. Med., 58 (2007) 1182-1195.
[19] P. Parasoglou, D. Malioutov, A.J. Sederman, J. Rasburn, H. Powell, L.F. Gladden,
A. Blake, M.L. Johns, Quantitative single point imaging with compressed sensing, J.
Magn. Reson., 201 (2009) 72-80.
[20] J. Hennig, A. Nauerth, H. Friedburg, RARE imaging: a fast imaging method for
clinical MR, Magn. Reson. Med., 3 (1986) 823-833.
[21] Z.P. Liang, P.C. Lauterbur, I.E.i. Medicine, B. Society, Principles of magnetic
resonance imaging: a signal processing perspective, SPIE Optical Engineering Press,
2000.
[22] F. Casanova, J. Perlo, B. Blumich, K. Kremer, Multi-echo imaging in highly
inhomogeneous magnetic fields, J. Magn. Reson., 166 (2004) 76-81.
Tables:
Table 1: Phase cycling scheme for FSE pulse sequence. Using a 4-step phase cycle
four complex components of two coefficients are read.
Phase \ Cycle No. 1st 2nd 3th 4th
phase x x x x
1st CPMG phase x x y y
2nd
CPMG phase x y x y
Figure Captions:
Fig. 1: Phantom diagrams. (a) Phantom A. Features are relatively scattered and
have different shapes and orientations to test the quality of reconstruction. (b)
Phantom B. Features are concentrated near the center of the FOV. Smallest feature
width is about 0.3mm.
Fig. 2: The Fast Spin Echo pulse sequence in a unilateral NMR. The sequence is
divided into four periods: the first encoding period, the first CPMG period, the
second encoding period (gradient blips), and the second CPMG period. The first
period is used to encode the phase of the initial coefficient using gradients employed
in the x and y directions (The encoding time to echo, TEe, is indicated). The second
period is used to accumulate data to be read as the first k-space coefficient with a
series of pulses (The detection time to echo, TEd is indicated). The third period is
used to shift phase to encode the next coefficient using gradient blips during two TE
periods. The fourth period is used to accumulate data to be read as the second k-
space coefficient with a series of pulses.
Fig. 3: Recovery of complex coefficient values using the phase cycling scheme. The
real component of the first coefficient is obtained by applying pulses and the
imaginary component is obtained by applying pulses. In addition to this an
additional cycling is applied for the pulses after the gradient "blip" ( and ).
Altogether the four combinations of acquired components result in reconstruction of
two complex k-space coefficients.
Fig. 4: Results of the experiment comparing CS and LP reconstructions of phantom
A with similar scan percentages. (a) 100% sampled image with 12 averaging
repetitions. (b-f) Low pass reconstructed images: sampling of 30%, 40%, 50%, 60%,
and 70% of the coefficients respectively. (g-k) Compressed sensing reconstructed
images: sampling of 30%, 40%, 50%, 60%, and 70% of the coefficients respectively.
RMS error relative to the high-averaged, 100% sampled image is indicated below
each image. (l-p) Scanning masks for CS scans: white pixels were scanned while black
pixels were skipped.
Fig. 5: Results of the experiment comparing CS and LP reconstructions of phantom
B with a similar scan percentage. (a) 100% sampled image with 12 averaging
repetitions. (b-f) Low pass reconstructed images: sampling of 30%, 40%, 50%, 60%,
and 70% of the coefficients respectively. (g-k) Compressed sensing reconstructed
images: sampling of 30%, 40%, 50%, 60%, and 70% of the coefficients respectively.
RMS error relative to the high-averaged, 100% sampled image is indicated below
each image.
Fig. 6: Cross section results for the experiment comparing CS and LP phantom B
reconstructions with a similar scan percentage for two representative lines (the 9th
and 18th
). (a) 100% sampled image with 12 averaging repetitions. (b-c) Cross section
results for the 9th
line for GS-, CS-, and LP-reconstructed images with 30% and 50% of
the coefficients, respectively. The solid black line with rectangular markers
represents the GS; the dashed light grey line with circular markers represents the LP;
and the dark grey dash-dotted line with triangular markers represents the CS. (d-e)
Cross section results for the 18th
line for GS-, CS-, and LP-reconstructed images with
30% and 50% of the coefficients, respectively.
Fig. 7: FSE experiment results with phantoms A and B. (a) Standard sampling
image with 24 averaging repetitions for phantom A. (b) FSE sampling image with 48
averaging repetitions for phantom A. (c) Standard sampling image with 24 averaging
repetitions for phantom B. (d) FSE sampling image with 48 averaging repetitions for
phantom B.
Tables:
Table 1: Phase cycling scheme for FSE pulse sequence. Using a 4-step phase cycle
four complex components of two coefficients are read.
Phase \ Cycle No. 1st
2nd
3th
4th
phase x x x x
1st
CPMG phase x x y y
2nd CPMG phase x y x y
Article highlights
• Two techniques were used to enable faster imaging in a Unilateral NMR device.
• Compressed Sensing was used to sub-sample the k-space.
• Fast spin echo was used to acquire two coefficients following each excitation.
• Reduction in scan times caused relatively minor deterioration of image quality.
• Further hardware and software improvements will facilitate imaging applications.