ORIGINAL ARTICLE

Compressed Sensing SEMAC: 8-fold Accelerated High

Resolution Metal Artifact Reduction MRI of Cobalt-Chromium
Knee Arthroplasty Implants
Jan Fritz, MD,* Shivani Ahlawat, MD,* Shadpour Demehri, MD,* Gaurav K. Thawait, MD,*
Esther Raithel, PhD,† Wesley D. Gilson, PhD,‡ and Mathias Nittka, PhD†

3 pulse sequences showed peripheral structures similarly well. The implant

Objective: The aim of this study was to prospectively test the hypothesis that a
artifact size was 46% to 51% larger on high-BW images when compared with
compressed sensing–based slice encoding for metal artifact correction (SEMAC)
both types of SEMAC images (P < 0.0001). Signal-to-noise ratios and CNRs
turbo spin echo (TSE) pulse sequence prototype facilitates high-resolution metal
of fat tissue, tendon tissue, muscle tissue, and fluid were statistically similar
artifact reduction magnetic resonance imaging (MRI) of cobalt-chromium knee
on intermediate-weighted MR images of all 3 pulse sequence types. On short
arthroplasty implants within acquisition times of less than 5 minutes, thereby
tau inversion recovery images, the SNRs of tendon tissue and the CNRs of fat
yielding better image quality than high-bandwidth (BW) TSE of similar
and fluid, fluid and muscle, as well as fluid and tendon were significantly higher
length and similar image quality than lengthier SEMAC standard of reference
on SEMAC and compressed sensing SEMAC images (P < 0.005, respectively).
pulse sequences.
Conclusions: We accept the hypothesis that prospective compressed sensing ac-
Materials and Methods: This prospective study was approved by our institu-
celeration of SEMAC is feasible for high-quality metal artifact reduction MRI of
tional review board. Twenty asymptomatic subjects (12 men, 8 women; mean
cobalt-chromium knee arthroplasty implants in less than 5 minutes and yields
age, 56 years; age range, 44–82 years) with total knee arthroplasty implants
better quality than high-BW TSE and similarly high quality than lengthier
underwent MRI of the knee using a commercially available, clinical 1.5 T MRI
SEMAC pulse sequences.
system. Two compressed sensing–accelerated SEMAC prototype pulse sequences
with 8-fold undersampling and acquisition times of approximately 5 minutes each Key Words: compressed sensing, SEMAC, MARS, arthroplasty, knee,
were compared with commercially available high-BW and SEMAC pulse se- cobalt-chromium, metal, artifact, reduction, MRI
quences with acquisition times of approximately 5 minutes and 11 minutes, respec- (Invest Radiol 2016;51: 666–676)
tively. For each pulse sequence type, sagittal intermediate-weighted (TR,
3750–4120 milliseconds; TE, 26–28 milliseconds; voxel size, 0.5
0.5

3 mm3) and short tau inversion recovery (TR, 4010 milliseconds; TE, 5.2–7.5
milliseconds; voxel size, 0.8
0.8
4 mm3) were acquired. Outcome variables
M etal artifact reduction magnetic resonance imaging (MRI) of
the knee allows for comprehensive imaging evaluation of pain
and dysfunction after arthroplasty including periprosthetic bone re-
included image quality, display of the bone-implant interfaces and pertinent knee sorption and osteolysis, synovitis, infection, periprosthetic fractures,
structures, artifact size, signal-to-noise ratio (SNR), and contrast-to-noise ratio arthrofibrosis, component malalignment, extensor mechanism in-
(CNR). Statistical analysis included Friedman, repeated measures analysis jury, and instability.1–4
of variances, and Cohen weighted k tests. Bonferroni-corrected P values of Conventional turbo spin echo (TSE) sequences with modified
0.005 and less were considered statistically significant. radiofrequency pulses and high receiver bandwidth (BW) in excess of
Results: Image quality, bone-implant interfaces, anatomic structures, artifact 500 Hz per pixel result in a meaningful decrease in metal implant–
size, SNR, and CNR parameters were statistically similar between the com- induced image artifacts including geometric distortions, signal pileups,
pressed sensing–accelerated SEMAC prototype and SEMAC commercial pulse and signal voids.5 However, specifically with cobalt-chromium knee
sequences. There was mild blur on images of both SEMAC sequences when implants, substantial artifacts remain along the bone-implant interfaces
compared with high-BW images (P < 0.001), which however did not impair and in the immediate vicinity of the implants, which may obscure im-
the assessment of knee structures. Metal artifact reduction and visibility of central portant MRI findings.6
knee structures and bone-implant interfaces were good to very good and signifi- Multispectral MRI techniques such as slice encoding for metal ar-
cantly better on both types of SEMAC than on high-BW images (P < 0.004). All tifact correction (SEMAC) in combination with view-angle tilting pro-
vide superior metal artifact reduction capabilities through the additional
Received for publication June 7, 2016; and accepted for publication, after revision, reduction of through-plane displacement of signal.7–10 Slice encoding
July 9, 2016. for metal artifact correction improves visualization of bone-implant-
From the *Russell H. Morgan Department of Radiology and Radiological Science, tissue interfaces and periprosthetic structures in the immediate vicinity
Johns Hopkins University School of Medicine, Baltimore, MD; †Siemens
Healthcare GmbH, Erlangen, Germany; and ‡Siemens Healthcare USA, Inc,
of implants and increases the detection rate of abnormal findings.8
Baltimore, MD. A practical limitation of high-resolution SEMAC MRI of the knee
Conflicts of interest and sources of funding: Jan Fritz received institutional research can be prolonged length of time for data acquisition, which often ranges
funds and speaker's honorarium from Siemens Healthcare USA and is a scientific around 10 minutes for 1 pulse sequence acquisition,7 even if conven-
advisor of Siemens Healthcare USA and Alexion Pharmaceuticals, Inc. Shivani
Ahlawat declares no potential conflicts of interest. Shadpour Demehri received re-
tional acceleration techniques such as partial Fourier or parallel imaging
search funds and support from Vital images, Inc. and Toshiba America Medical are applied.11 The compressed sensing approach offers a novel, alterna-
System and Carestream Health, Inc. and General Electric. Gaurav K. Thawait de- tive acceleration strategy for MRI that utilizes incoherent k-space
clares no potential conflicts of interest. Esther Raithel is an employee of Siemens undersampling in combination with iterative image reconstruction
Healthcare GmbH. Wesley D. Gilson is an employee of Siemens Healthcare
USA. Matthias Nittka is an employee of Siemens Healthcare GmbH. This study
methods, thereby exploiting sparse information in image contents.12–14
was supported by Siemens Healthcare USA. Slice encoding for metal artifact correction is especially suited due to a
Correspondence to: Jan Fritz, MD, Russell H. Morgan Department of Radiology and high degree of sparsity in the additional phase encoding dimension
Radiological Science, Johns Hopkins University School of Medicine, 601 N where only distorted signal is located, which is usually only a tiny frac-
Caroline St, JHOC 3140A, Baltimore, MD 21287. E-mail: [email protected].
Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.
tion of the covered volume.
ISSN: 0020-9996/16/5110–0666 Therefore, the purpose of this prospective study was to test the
DOI: 10.1097/RLI.0000000000000317 hypothesis that an 8-fold, undersampled compressed sensing–based

666 www.investigativeradiology.com Investigative Radiology • Volume 51, Number 10, October 2016

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Investigative Radiology • Volume 51, Number 10, October 2016 Compressed Sensing SEMAC TSE of the Knee

SEMAC pulse sequence prototype facilitates high-resolution metal arti- transformation of the additional SEMAC dimension results in a set of
fact reduction MRI of cobalt-chromium knee arthroplasty implants subimages for each slice, which may be referred to as uncombined par-
within acquisition times of less than 5 minutes, thereby yielding better titions. The set of uncombined partitions consists of a central partition
image quality than high-BW TSE of similar length and similar image that contains the nondistorted image content, whereas the remaining
quality than lengthier SEMAC standard of reference pulse sequences. partitions only contain distorted signal. During image postprocessing,
the uncombined partitions are resorted and summed up to form a distor-
MATERIALS AND METHODS tion corrected final image. Ideally, the number of SEMAC encoding
steps, for example, the size of the additional phase encoding dimension,
Study Design is chosen such that the majority of the actual slice distortion artifact
This prospective study was approved by our institutional review is covered.
board and complied both with the Declaration of Helsinki and the The compressed sensing reconstruction was performed using a
Health Insurance Portability and Accountability Act. Written informed nonlinear iterative SENSE-type algorithm with the L1 norm of the
consent was obtained from all subjects. Wavelet-transformed image as regularization term.18 The rationale for
Inclusion criteria were the presence of asymptomatic cobalt- the compressed sensing–based approach is based on the nonrandom na-
chromium total knee arthroplasty implants, adult age, agreement to ture and sparsity properties of MR images in general and specifically of
cooperate, and written informed consent. Exclusion criteria were con- uncombined SEMAC images. The uncombined MR images b f that are
traindications to MRI as well as acute pain and dysfunction. No subjects used for the SEMAC reconstruction were found by minimizing the fol-
were excluded. The MRI examinations were obtained for research lowing cost function:
purposes only.
Outcome variables included image quality parameters, visibility Nc
b
f ¼ arg min 1 ∑ jjync −Fu ðSnc ⊙f Þjj þ λjjW f jj1
2
of bone-implant interfaces, visibility of specific anatomic structures, f 2 nc¼1 2
size of the implant-generated image artifact, as well as apparent
signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR).
With Nc the number of receive channels, yn the nth channel of the ac-
Subjects quired raw data, Fu the undersampled Fourier transform, Sn the nth coil
sensitivity, λ a regularization parameter, and w the redundant Haar
The study population consists of 20 adult asymptomatic living
wavelet transform. The data fidelity term ensures the consistency be-
human subjects with cobalt-chromium total knee arthroplasty implants
tween the raw data and the estimated image. The regularization term
(12 men, 8 women; mean age, 56 years; age range, 44–82 years; mean
promotes sparsity by penalizing the L1 norm of the image of the current
body mass index, 28 kg/m2; body mass index range, 21–38 kg/m2).
iteration in the Wavelet domain. Lambda steers the tradeoff between
MRI Technique smoothness and the removal of aliasing and noise.
The non–fat-suppressed intermediate-weighted accelerated
All MRI studies were performed on a commercially avail-
CS-SEMAC data set was reconstructed with 30 iterations and a regular-
able, clinical wide-bore 1.5 T MRI system (MAGNETOM Aera,
ization parameter of 0.005, and the STIR CS-SEMAC data set was re-
NUMARIS/4 Syngo MR D13A; Siemens Healthcare, Erlangen,
constructed with 20 iterations and a regularization parameter of 0.008.
Germany) with 48 radiofrequency channels, maximum gradient
field amplitude of 45 mT/m, and a slew rate of 200 T/m/s. Depending
on the knee size, either a commercially available 15-channel transmit/ Image Evaluation
receive knee coil (QED, Mayfield Village, OH) or an 18-channel A total of 60 pairs of corresponding intermediate-weighted and
receive-only matrix surface coil (Siemens Healthcare) in combination STIR high-BW (n = 20), SEMAC (n = 20) and CS-SEMAC (n = 20)
with dedicated spine array surface coil embedded in the scanner table data set were evaluated. The data sets were randomized, and all image
(Siemens Healthcare) was used. The 15-channel transmit/receive knee annotations were removed by editing the DICOM headers of the indi-
coil was used in 15 patients, and the 18-channel receive-only matrix vidual data sets (DicomBrowser software, version 1.7.0b5, http://nrg.
surface coil was used in 5 patients. wustl.edu/software/dicom-browser). All evaluations were performed
The MRI protocol consisted of pairs of sagittal intermediate- with commercially available picture archiving and communications sys-
weighted and short tau inversion recovery (STIR) TSE pulse sequences tem software (PACS, Ultravisual, Merge Healthcare, Chicago, IL) at ap-
utilizing different metal artifact reduction techniques (Table 1). An proximately 1 lux on a 5-megapixel, 30-inch, diagnostic-quality liquid
8-fold accelerated slice-encoding metal artifact correction SEMAC crystal display monitor, which was calibrated to DICOM standards.
pulse sequence prototype (CS-SEMAC)15 was compared against com- Two fellowship-trained, full-time musculoskeletal radiologists,
mercially available optimized high-BW (high-BW TSE) and conven- each with 10 years of experience in musculoskeletal MRI, graded im-
tional SEMAC sequences.1,8 All STIR pulse sequences used high-BW age quality, bone-implant interfaces, and pertinent anatomic structures
radiofrequency inversion pulses, which matched the BW of the excita- of the knee. Ratings were performed with equidistant 5-point Likert
tion pulse. The pulse sequences were acquired in random order. scales, where a rating of 1 denoted either “very bad” or “nondiagnostic”
Based on the theories of compressed sensing and sparsity-driven with resultant complete obscuration of anatomic details, 2 denoted
MRI,10,12,16–26 the CS-SEMAC pulse sequence prototype used an inco- “poor” or “severe” with substantial obscuration of anatomic details, 3
herent k-space undersampling pattern,27 which was based on a Poisson denoted “adequate” or “moderate” with impaired depiction of small
disk variable density pattern, elliptical scanning, and a fully sampled areas of anatomic detail, 4 denoted “good” or “mild” with minimal im-
k-space center. This approach resulted in a reduction in image data ac- pairment but preservation of all anatomic detail, and 5 denoted “very
quisition time of approximately 60% when compared with the commer- good” or “none” with unimpaired depiction of all anatomic details.16,28
cially available conventional SEMAC pulse sequence. Each pulse Standardized hanging protocols were used. Observers were free to use
sequence was acquired once only. their preferred window and level settings, magnification, and scrolling
The image reconstruction performed in this work was a combi- mode. Data sets were reviewed over a period of 8 weeks.
nation of an initial compressed sensing reconstruction,18 followed by The bone-implant interfaces were grouped into patellar compo-
a SEMAC reconstruction.7 Slice encoding for metal artifact correction nent, tibial component (medial plateau, lateral plateau, and keel), and
is based on the principle of applying additional 3-dimensional phase femoral component (flange, central medial and lateral femoral condyle,
encoding steps to spatially encode through-plane distortions. Fourier and posterior medial and lateral femoral condyle).

© 2016 Wolters Kluwer Health, Inc. All rights reserved. www.investigativeradiology.com 667

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Fritz et al Investigative Radiology • Volume 51, Number 10, October 2016

TABLE 1. MRI Pulse Sequence Protocol

Parameters High-BW TSE Compressed Sensing SEMAC SEMAC

Orientation Sagittal Sagittal Sagittal Sagittal Sagittal Sagittal
Weighting Intermediate STIR Intermediate STIR Intermediate STIR
Repetition time, ms 4120 4010 3750 4010 3750 4010
Echo time, ms 26 5.2 28 7.5 28 7.5
Fat suppression No STIR No STIR No STIR
Inversion time, ms — 160 — 160 — 160
No. SEMAC steps — — 19 19 19 19
Echo train length 19 15 11 19 11 19
Receiver BW, Hz/pixel 601 601 601 601 601 601
Flip angle, degree 150 150 150 150 150 150
Field of view, mm 170
170 200
200 170
170 200
200 170
170 200
200
Matrix 320
256 320
256 320
256 256
205 320
256 256
205
In-plane resolution, mm 0.5
0.5 0.6
0.6 0.5
0.5 0.8
0.8 0.5
0.5 0.8
0.8
Slice thickness/gap, mm 3/0 4/0 3/0 4/0 3/0 4/0
No. excitations 3 3 1 1 1 1
No. concatenations 2 2 1 2 1 2
No. slices 34 26 34 26 34 26
Acceleration factor 1 1 8 8 3 3
In-plane frequency Anterior to posterior Anterior to posterior Anterior to posterior Anterior to posterior Anterior to posterior Anterior to posterior
encoding direction
Acquisition time 4 min 26 s 4 min 58 s 4 min 28 s 4 min 50 s 11 min 05 s 11 min 23 s
MRI indicates magnetic resonance imaging; BW, bandwidth; TSE, turbo spin echo; SEMAC, slice encoding for metal artifact correction; STIR, short tau
inversion recovery.

Image quality parameters included the degree of motion artifact, to the measurements of the conventional SEMAC images and given as rel-
metal artifact, failure of fat suppression on STIR images, edge blurring, ative percentage. Area measurements were performed by 1 observer with
and image noise, as well as the quality of tissue differentiation. 5 years of experience in MRI. The measurements were carried out 3 times
Anatomic structures assessed were grouped into anterior com- with 2-week intervals in between.
partment (extensor tendons, suprapatellar recess, retroquadriceps fat To determine the number of SEMAC encoding steps required to
pad, Hoffa's fat pad), central compartment (posterior cruciate ligament correct the through-plane displacement of signal caused by the knee
if retained, posterior capsule, joint fluid), and posterior compartment implants, a fellowship-trained, full-time musculoskeletal radiologist
(muscles and popliteal vessels). with 10 years of experience evaluated the uncombined, individual
Artifact size was quantified by manual segmentation of the implant intermediate-weighted SEMAC and CS-SEMAC partitions for the
area (femoral component and tibial component including polyethylene presence of displaced signal starting from the center image to the me-
liner) without discernible anatomic information on intermediate-weighted dial and lateral periphery. The central slice of a data set was used for this
and STIR images, including the implant as well as signal void (low-signal analysis. The partition number (n) of the last image that contained sig-
intensity) and signal pileup (high-signal intensity). Area measurements nal was documented. The evaluation was repeated twice, and in the case
were obtained on a central slice as well as slices through the center of of discordance, a third consensus read was performed. The SEMAC
the medial and lateral femoral condyle. A PACS area segmentation tool encoding step number was calculated as 2
n + 1.
(Ultravisual) was used. Measurements were performed on the same slices Apparent SNRs were calculated through manual signal intensity
of corresponding high-BW, SEMAC, and CS-SEMAC images. The sum measurements of intra-articular fat in Hoffa's fat pad, joint fluid in the
of the artifact areas on the 3 slices was calculated. Values were normalized suprapatellar recess, patella tendon at the mid portion, and muscle

TABLE 2. Image Quality Assessments

Pulse Sequence Type

Parameter High-BW TSE (I) Compressed Sensing SEMAC (II) SEMAC (III) Friedman Test P Multiple Comparison Analysis
Motion artifacts 5 (4–5)* 5 (4–5)* 4 (4–5) 0.002 *Different from (III)
Fat suppression 5 (3–5) 5 (4–5) 5 (4–5) 0.289 Not applicable
Metal artifact 3 (2–4) 4 (3–5)* 4 (3–5)* <0.001 *Different from (I)
Edge blurring 5 (4–5)* 4 (2–4) 4 (2–5) 0.001 *Different from (II) (III)
Image noise 4 (3–5) 4 (3–5) 4 (3–5) 0.541 Not applicable
*Statistically different value (P < 0.005) in comparison to the values in the columns denoted by the numbers in parentheses.
BW indicates bandwidth; TSE, turbo spin echo; SEMAC, slice encoding for metal artifact correction.

668 www.investigativeradiology.com © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Investigative Radiology • Volume 51, Number 10, October 2016 Compressed Sensing SEMAC TSE of the Knee

TABLE 3. Visibility of Anatomic Structures

Pulse Sequence Type

Compressed Sensing
Parameter High-BW TSE (I) SEMAC (II) SEMAC (III) Friedman Test P Multiple Comparison Analysis
Anterior compartment
Extensor tendons 4 (2–5) 4 (3–5) 4 (3–5) 0.360 Not applicable
Suprapatellar recess 4 (2–5) 4 (3–5) 4 (3–5) 0.491 Not applicable
Retroquadriceps fat pad 4 (2–5) 4 (2–5) 4 (3–5) 0.887 Not applicable
Hoffa's fat pad 4 (2–5) 4 (2–5) 4 (3–5) 0.353 Not applicable
Central compartment
Posterior cruciate ligament 3 (2–4) 4 (4–5)* 5 (4–5)* 0.004 *Different from (I)
Posterior joint capsule 3 (1–4) 4 (3–5)* 4 (3–5)* <0.001 *Different from (I)
Joint fluid 3 (1–5) 4 (3–5)* 4 (2–5)* <0.001 *Different from (I)
Posterior compartment
Muscle 4 (4–5) 4 (4–5) 4 (4–5) 0.019 Not applicable
Popliteal vessels 4 (4–5) 4 (4–5) 4 (4–5) 0.019 Not applicable
*Statistically different value (P < 0.005) in comparison to the values in the columns denoted by the numbers in parentheses.
BW indicates bandwidth; TSE, turbo spin echo; SEMAC, slice encoding for metal artifact correction.

FIGURE 1. A 64-year-old man with right cobalt-chromium knee arthroplasty implants. Corresponding sagittal, intermediated-weighted and short tau
inversion recovery (STIR) high-bandwidth (BW) turbo spin echo (TSE), slice-encoding metal artifact correction (SEMAC), and compressed
sensing–accelerated SEMAC MR images. Both SEMAC pulse sequences used 19 encoding steps, which afforded a more powerful metal artifact reduction
than high-BW technique. The use of compressed sensing sampling resulted in an 8-fold acceleration of SEMAC acquisition and a scan time reduction of
approximately 60% when compared with conventional SEMAC and similar acquisition time than high-BW TSE. Observer ratings indicated good visibility
of the extensor mechanism (gray arrows) on high-BW TSE, SEMAC, and compressed sensing SEMAC MR images. Articular structures, such as joint fluid
(asterisks) and posterior capsule (white arrows) were significantly better seen on SEMAC and compressed sensing SEMAC MR images than on high-BW
images.

© 2016 Wolters Kluwer Health, Inc. All rights reserved. www.investigativeradiology.com 669

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Fritz et al Investigative Radiology • Volume 51, Number 10, October 2016

TABLE 4. Visibility of Bone-Implant Interfaces

Pulse Sequence Type

Compressed Sensing
Parameter High-BW TSE (I) SEMAC (II) SEMAC (III) Friedman Test P Multiple Comparison Analysis
Patellar component 4 (4–5) 5 (4–5) 4 (4–5) 0.071 Not applicable
Femoral component
Flange 3 (2–4) 4 (4–5)* 4 (3–5)* 0.003 *Different from (I)
Central medial femoral condyle 3 (1–4) 4 (2–5)* 3.5 (2–5)* 0.002 *Different from (I) (III)
*Different from (I) (II)
Central lateral femoral condyle 3 (2–4)* 4 (2–5)* 3.5 (1–5) 0.004 *Different from (II) (III)
*Different from (I) (III)
Posterior medial femoral condyle 3 (1–4) 4 (2–4)* 4 (2–5)* 0.004 *Different from (I)
Posterior lateral femoral condyle 3 (1–4) 4 (2–4)* 3.5 (2–5)* 0.003 *Different from (I)
Tibial component
Medial tibia plateau 3 (2–5) 4 (3–5)* 4 (3–5)* 0.002 *Different from (I)
Lateral tibia plateau 3 (2–5) 4 (3–5)* 4 (3–5)* 0.002 *Different from (I)
Keel 3 (2–5) 4 (4–5)* 4 (4–5)* <0.002 *Different from (I)
*Statistically different value (P < 0.005) in comparison to the values in the columns denoted by the numbers in parentheses.
BW indicates bandwidth; TSE, turbo spin echo; SEMAC, slice encoding for metal artifact correction.

FIGURE 2. A 72-year-old woman with left cobalt-chromium knee arthroplasty implants. Corresponding sagittal, intermediated-weighted and short tau
inversion recovery (STIR) high-bandwidth (BW) turbo spin echo (TSE), slice-encoding metal artifact correction (SEMAC), and compressed sensing
SEMAC MR images at the level of the lateral femoral condyle. Both SEMAC pulse sequences were acquired with 19 encoding steps. There was better
visibility of the femoral (white arrows) and tibial (gray arrows) bone-implant interface on conventional and compressed sensing accelerated SEMAC MR
images, when compared with high-BW MR images. The asterisks demonstrate small areas of ripple artifacts.

670 www.investigativeradiology.com © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Investigative Radiology • Volume 51, Number 10, October 2016 Compressed Sensing SEMAC TSE of the Knee

tissue in gastrocnemius heads at the level of the proximal tibia on non– test power of 0.90 with a Bonferroni-corrected α error of 0.005. Variables
fat-suppressed and fat-suppressed MRI data sets. Signal intensity values are given as the median with minimum and maximum in parentheses or
and the first standard deviation of room air were obtained by using a ratios with percentages. Interreader agreement or ordinal data were assessed
PACS region of interest measurement tool (Ultravisual). Signal-to- using Cohen weighted k test and graded according to the recommendations
noise ratios were defined as SIROI/SDbackground, where SIROI is the mean of Landis and Koch.29 To simplify data presentation, the results of both
signal intensity of the region of interest of the respective tissue and readers were combined. The Friedman test was used to test for ordinal
SDbackground is the standard deviation of the background signal. The re- scale differences between sequences. In the case of a positive Friedman
gions of interest were 4 to 5 mm2 for background signal and 2 mm2 for test, a multiple comparison analysis was performed to identify signifi-
intra-articular fat, fluid, tendon, and muscle. The background noise was cant differences. Repeated measures analysis of variances was used to
measured with 6 regions of interest of 4 to 5 mm2 that were positioned in test for differences of artifact size, SNR, and CNR values between se-
room air at the level of the anterosuperior, anterior, anteroinferior, pos- quences. Equality of variances estimates were used for correction. In
terosuperior, posterior, and posteroinferior close to the skin surface of the case of a significant difference between measurements, pairwise
the knee. Care was taken to avoid room air measurements in areas, which comparison between sequences was performed. Measurement reliability
potentially contained artifacts, such as aliasing. All measurements were was assessed using intraclass correlation coefficient and classified into
averaged over 3 different slices. Measurements were performed by 1 ob- first class (0.80–1.00), second class (0.50–0.80), third class (0.20–0.50),
server with 5 years of experience in MRI over a period of 2 weeks. Mea- and fourth class (0.00–0.20). Owing to multiple comparisons, P values
surement averages were used to calculate CNR as |SNR1 − SNR2|. of 0.005 and less were considered statistically significant.

Statistical and Quantitative Assessments

Statistical analyses were performed using JMP Pro 12.1 software RESULTS
(SAS Institute, Cary, NC). To detect a difference of 1 of visual ratings, The results of the observer analysis of image quality are given in
an a priori power calculation based on repeated measures analysis of Table 2. The interobserver agreements were moderate to good (k,
variances derived a minimum sample size of 20 subjects to achieve a 0.409–0.678). Motion artifacts were mildly present on SEMAC images

FIGURE 3. A 68-year-old man with left cobalt-chromium knee arthroplasty implants. Corresponding sagittal, intermediated-weighted high-bandwidth
(BW) turbo spin echo (TSE), slice-encoding metal artifact correction (SEMAC), and compressed sensing SEMAC MR images at central level. Both SEMAC
sequences were acquired with 19 encoding steps. The white arrows demonstrate the difference in size of the implant induced metal artifact. In addition,
image distortions can be seen on the high-BW image. Manual segmentation (lower row) quantified the area of the metal artifact, corroborating the
visibly smaller artifact size on conventional and compressed sensing SEMAC MR images, when compared with high-BW MR images. There was no
meaningful difference between the measured areas on the conventional and compressed sensing accelerated SEMAC sequences. Figure 3 can be
viewed online in color at www.investigativeradiology.com.

© 2016 Wolters Kluwer Health, Inc. All rights reserved. www.investigativeradiology.com 671

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Fritz et al Investigative Radiology • Volume 51, Number 10, October 2016

and, on average, absent on high-BW and CS-SEMAC images (P < 0.001). SEMAC and CS-SEMAC images when compared with high-BW
Image noise was rated as mild or absent, and fat suppression was rated images (P ≤ 0.002–0.002). There were few significant but small dif-
very good on high-BW, SEMAC, and CS-SEMAC images (P > 0.001). ferences between SEMAC and CS-SEMAC (Table 4). There were
The degree of metal artifacts was moderate on high-BW and mild on occasional, mild bin-combination artifacts around the tip of pegs
CS-SEMAC and SEMAC images (P < 0.001). There was mild edge and screws in SEMAC and CS-SEMAC images (Fig. 2).
blurring on both CS-SEMAC and SEMAC images, whereas there was The measurement reliability of the implant artifact size (Fig. 3) was
none on high-BW images (P < 0.002). There were no statistical differ- first class for intermediate-weighted images (intraclass correlation coeffi-
ences between SEMAC and CS-SEMAC TSE. cient, 0.912; 95% confidence interval, 0.8596–0.985) and for STIR images
The results of the evaluation of the visibility of anatomic struc- (intraclass correlation coefficient, 0.869; 95% confidence interval, 0.8173–
tures of the knee are given in Table 3. The interobserver agreements 0.967). There were no significant differences in the size of the implant
were moderate to very good (k, 0.456–0.836). The visibility of struc- artifact between intermediate-weighted SEMAC and CS-SEMAC images
tures of the anterior and posterior compartments was good without sig- (P = 0.128) or between STIR SEMAC and CS-SEMAC images (P =
nificant difference between high-BW, SEMAC, and CS-SEMAC 0.325). The implant artifact size was on average 51.5% (range, 20.1%–
images (Fig. 1). The structures of the central compartment including 73.1%) larger on intermediate-weighted high-BW images and 46.3%
7 cases with a retained posterior cruciate ligament were significantly (range, 25.7%–72.8%) on STIR images, when compared with SEMAC
better seen on SEMAC and CS-SEMAC images when compared with images (P < 0.0001) and CS-SEMAC images (P < 0.0001, respectively).
high-BW (P ≤ 0.001–0.004) (Fig. 1). There were no statistical differ- Based on the analysis of uncombined, intermediate-weighted
ences between SEMAC and CS-SEMAC TSE. SEMAC partitions, 17 SEMAC encoding steps15–19 are required on av-
The results of the bone-implant interface evaluation are given erage to compensate for the entirety of the through-plane displacement
in Table 4. The interobserver agreements were moderate (k, 0.412– of signal (Fig. 4). There was no statistical difference between SEMAC
0.583). The patellar component–bone interface showed good to very and CS-SEMAC.
good visibility on high-BW, SEMAC, and CS-SEMAC images with- The results of the SNR analysis are given in Figure 5. For the
out statistical differences. The femoral component–implant interface intermediate-weighted images, there was no significant difference in
was significantly better seen on SEMAC and CS-SEMAC images SNRs between the 3 pulse sequences for fat, fluid, tendon, and mus-
when compared with high-BW images (P = 0.002–0.004). Similarly, cle tissue. For STIR images, the SNRs of tendon tissue and fluid
the tibial component–bone interface was significantly better seen on were significantly higher on SEMAC and CS-SEMAC images

FIGURE 4. Conventional and compressed sensing SEMAC MR images showing several of the 19 individual encoding step images from centrally (low
numbers) to peripherally (high numbers). The sequence of images shows how the amount of displaced signal decreases in the periphery with the least
amount of signal being present on the most peripheral ninth images. The combined SEMAC and compressed sensing image were created with a sum of
square algorithm based on the signal included in the individually encoded SEMAC and compressed sensing MR images. There was no visually
detectable difference between the conventional and compressed sensing accelerated SEMAC sequences.

672 www.investigativeradiology.com © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Investigative Radiology • Volume 51, Number 10, October 2016 Compressed Sensing SEMAC TSE of the Knee

FIGURE 5. Box-and-whisker plots of signal-to-noise ratios (SNR) of intermediated-weighted and STIR high-bandwidth turbo spin echo (high-BW TSE),
slice-encoding metal artifact correction turbo spin echo (SEMAC TSE), and compressed sensing SEMAC (CS-SEMAC TSE) MR images of intra-articular fat
(Fat), joint fluid (Fluid), patella tendon (Tendon), and muscle tissue (Muscle). The asterisk indicates the significant difference when compared with all other
sequences. There was no significant SNR difference on intermediated-weighted MR images. The SNR of fluid and tendons was significantly lower on
high-BW STIR images, when compared with SEMAC and CS-SEMAC STIR TSE.

(P < 0.005, for both) compared with high-BW images, whereas there implementation of compressed sensing–based acceleration of SEMAC
was no significant difference of SNRs for fat and muscle. Further, MRI can markedly reduce the data acquisition time and maintain high
there were no significant differences for any tissue types between image quality.
SEMAC and CS-SEMAC STIR images. Slice encoding for metal artifact correction is a naturally time-
The results of the CNR analysis are given in Figure 6. There were consuming technique because it involves additional 3-dimensional
no significant differences between tissue types on intermediate- phase encoding for each image slice. The larger the coverage needed
weighted high-BW, SEMAC, and CS-SEMAC images. On STIR im- to correct the through-plane signal displacement due to the metallic im-
ages, the CNRs of fat and fluid, fluid and muscle, and fluid and tendon plant, the higher the number of SEMAC encoding steps needed. Each
were significantly higher for SEMAC and CS-SEMAC images increase in encoding steps results in a linear increase in acquisition time.
(P < 0.005, respectively) compared with high-BW images, whereas Titanium- and zirconium-based implants have a low magnetic suscepti-
there were no significant differences between sequences for the CNR bility and require less encoding steps, whereas cobalt-chromium alloys
of fat and muscle tissue. There were no significant CNR differences be- have a substantially higher magnetic susceptibility and require a higher
tween SEMAC and CS-SEMAC STIR images for any tissue pairs. number of encoding steps. Based on our clinical experience, we de-
signed this prototype with 19 SEMAC encoding steps to maximize
DISCUSSION the correction of through-plane displaced signal that is typically associ-
We demonstrate the feasibility of prospective compressed ated with cobalt-chromium knee arthroplasty implants at 1.5 T. Our
sensing–based acceleration of SEMAC MRI with incoherent k-space analysis of the uncombined SEMAC partitions showed that, depending
undersampling and iterative reconstruction for metal artifact reduction on the cobalt-chromium knee arthroplasty implant type, the vast major-
MRI of the knee using a commercially available clinical MRI system ity or all of the displaced signal was corrected for, as evidenced by little
and coils. Compressed sensing–based undersampling of SEMAC data to no signal being present on the most peripheral encoded images. This
afforded an effective acceleration of approximately 2.4 when compared resulted in excellent metal artifact reduction on SEMAC MR images,
with the current publicly available, conventional SEMAC pulse se- which were clearly superior to the high-BW images.
quence with parallel imaging acceleration, which reduced the acquisi- The acquisition times of multispectral pulse sequences for high-
tion time from 11 minutes to less than 5 minutes. The compressed resolution MRI of the knee often range between 8 and 12 minutes,1,7
sensing–accelerated SEMAC prototype achieved significantly better which in regards to today's health care environment and patient compli-
metal artifact reduction, better visibility of the bone-implant interfaces ance poses a practical limitation. Previously used acceleration strategies
and central periprosthetic structures, and similar visibility of more pe- included linear parallel imaging acceleration, shortening repetition
ripheral structures than high-BW MRI. Furthermore, the compressed times, limiting the number of encoding steps, increasing slice thickness,
sensing–accelerated SEMAC prototype achieved similar results to con- increasing echo train length, and lowering spatial resolution,8,9,11
ventional SEMAC MRI without perceptible loss of image quality or which, however, may result in increased image noise and blur as
metal reduction capacity. Our initial data demonstrate that the clinical well as decreased metal artifact reduction and anatomic detail.

© 2016 Wolters Kluwer Health, Inc. All rights reserved. www.investigativeradiology.com 673

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Fritz et al Investigative Radiology • Volume 51, Number 10, October 2016

FIGURE 6. Box-and-whisker plots of contrast-to-noise ratios (CNRs) of intermediated-weighted and STIR high-bandwidth turbo spin echo (high-BW TSE),
slice-encoding metal artifact correction turbo spin echo (SEMAC TSE), and compressed sensing SEMAC (CS-SEMAC TSE) MR images of intra-articular fat
and joint fluid (Fat-Fluid), fat and muscle tissue (Fat-Muscle), joint fluid and muscle tissue (Fluid-Muscle), joint fluid and patella tendon (Fluid-Tendon). The
asterisk indicates the significant difference when compared with all other sequences. There was no significant CNR difference on
intermediated-weighted MR images. The CNRs of intra-articular fat and joint fluid, joint fluid and muscle tissue, and joint fluid and patella tendon were
significantly lower on high-BW STIR images, when compared with SEMAC and CS-SEMAC STIR TSE.

Compressed sensing offers a different strategy for the acceleration of reduction of in-plane signal displacement and preservation of
data acquisition, which based on the results of our study may be used MR signal.
without degrading quality or reducing the metal suppression capacity In addition to the potential economic importance of shorter ac-
of SEMAC. quisition times of SEMAC MRI, this speed up may also decrease the
Our technique of compressed sensing–based k-space sampling number of data sets with patient motion, which is suggested by the ob-
resulted in an 8-fold total and 2.4 relative acceleration, which is an im- servations of overall mild but significantly less motion artifacts on
provement when compared with a parallel imaging–based acceleration 5 minutes high-BW and CS-SEMAC MRI data sets.
factor of 2 to 3 and prospective compressed sensing–based acceleration The combination of STIR and SEMAC technique can be chal-
factor of approximately 1.2 from previous studies.1,8,14 The compressed lenging because of the considerable loss of signal from fat-bound
sensing–based 8-fold acceleration corresponded to an approximately protons, the time-consuming incorporation of the inversion time,
60% reduction of conventional SEMAC data acquisition equating to a and the need for concatenation of the image volume. The resulting
total acquisition time of approximately 5 minutes for intermediate- increase in acquisition time and relative decrease in signal required
weighted and STIR contrast. In comparison to the optimized high-BW compensation through increasing the echo train length and decreas-
pulse sequence with a similar acquisition time of 5 minutes, the time ing the spatial resolution. Both adaptations resulted in some loss of
gains of compressed sensing acquisition allowed for the time-neutral image sharpness, which, however, may be of limited clinical impor-
use of the 19 SEMAC encoding steps and high degree of metal artifact tance as STIR images mainly serve to assess signal, rather than morphol-
reduction with the CS-SEMAC pulse sequence prototype. ogy, which is provided by the high-resolution intermediate-weighted
Furthermore, the compressed sensing–based gains in acquisi- data sets.
tion time of SEMAC also afforded the preservation of sequence param- Implant loosening is an important indicator of arthroplasty im-
eters that are important for high image quality MRI of the knee. Long plant failure and may be caused by mechanical effects and infection.
repetition times around 4000 milliseconds were used in combination Bone resorption and osteolysis are the principal imaging markers of im-
with echo times of intermediate length to maximize signal and achieve plant loosening that typically occur along the bone-implant interface.
intermediate weighting including increasing the signal of fluid and As such, accurate visualization of the bone-implant interfaces is para-
maximizing contrast resolution of musculoskeletal tissue.1 The SNR mount but perhaps most challenging to achieve with MRI in the presence
CS-SEMAC was sufficiently high to support small voxel sizes of of cobalt-chromium implants. The results of our study demonstrate the
0.5
0.5
3 mm3 intermediate-weighted and 0.8
0.8
4 mm3 overall good visualization of the femoral and tibial bone-implant inter-
STIR SEMAC MR images for visualizing small anatomical detail. faces with both SEMAC techniques, which was clearly superior to
Echo train length was purposely kept as low as possible to limit image high-BW MRI and most likely afforded by the high number of SEMAC
blur. While a high receiver BW is important to reduce in-plane sig- encoding steps. Importantly, there was no meaningful statistical differ-
nal displacement, we chose 600 Hz/pixel to balance sufficient ence between SEMAC and CS-SEMAC MR images. The patellar

674 www.investigativeradiology.com © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Investigative Radiology • Volume 51, Number 10, October 2016 Compressed Sensing SEMAC TSE of the Knee

implants were made of polyethylene in all patients, and consequently ACKNOWLEDGMENTS

high-BW, SEMAC, and CS-SEMAC MR images showed similarly good The authors thank Christoph Forman, Jens Wetzl, and Michael
to very good visualization of the patellar implant-bone interface. O. Zenge for their work on the image reconstruction framework.
There were mild bin-combination artifacts or ripple artifacts
around the tips of the pegs and screws of the femoral and tibial compo-
nents in SEMAC and CS-SEMAC images alike, which is a known ef- REFERENCES
fect of gradient selection-based multispectral imaging techniques.30 1. Fritz J, Lurie B, Potter HG. MR imaging of knee arthroplasty implants. Radio-
Simulations and phantom experiments suggested that overlapping of graphics. 2015;35:1483–1501.
SEMAC slice can facilitate a near-complete reduction of this artifact. 2. Kapadia BH, Berg RA, Daley JA, et al. Periprosthetic joint infection. Lancet.
The time gains afforded through compressed sensing may be in part 2016;387:386–394.
invested in this time-consuming strategy.31 3. Heyse TJ, Chong le R, Davis J, et al. MRI analysis for rotation of total knee com-
ponents. Knee. 2012;19:571–575.
Similar to the results of the bone-implant analysis, SEMAC and
4. Plodkowski AJ, Hayter CL, Miller TT, et al. Lamellated hyperintense synovitis:
CS-SEMAC were in similar fashion significantly better than high-BW potential MR imaging sign of an infected knee arthroplasty. Radiology. 2013;
for the assessment of central joint structures, which are in close vicinity 266:256–260.
to the femoral and tibial components. Importantly, despite the mild de- 5. Fritz J, Lurie B, Miller TT, et al. MR imaging of hip arthroplasty implants. Radio-
gree of blurring on SEMAC and CS-SEMAC images, there were no sig- graphics. 2014;34:E106–E132.
nificant differences of the observer assessments of peripherally located 6. Talbot BS, Weinberg EP. MR imaging with metal-suppression sequences for eval-
anterior and posterior structures such as the extensor mechanism. Given uation of total joint arthroplasty. Radiographics. 2016;36:209–225.
the similar acquisition times of the high-BW and CS-SEMAC pulse 7. Ai T, Padua A, Goerner F, et al. SEMAC-VAT and MSVAT-SPACE sequence strat-
egies for metal artifact reduction in 1.5 T magnetic resonance imaging. Invest
sequences, CS-SEMAC images may be able to partially or completely Radiol. 2012;47:267–276.
replace high-BW images. A patient study will be required to test 8. Sutter R, Hodek R, Fucentese SF, et al. Total knee arthroplasty MRI featuring
this hypothesis. slice-encoding for metal artifact correction: reduction of artifacts for STIR
The iterative reconstruction process of the undersampled k-space and proton density-weighted sequences. AJR Am J Roentgenol. 2013;201:
1315–1324.
CS-SEMAC data was performed inline in the background of the MRI
9. Agten CA, Del Grande F, Fucentese SF, et al. Unicompartmental knee arthroplasty
systems while running the pulse sequences of the research protocol or MRI: impact of slice-encoding for metal artefact correction MRI on image quality,
after the research protocol had finished in case CS-SEMAC was ac- findings and therapy decision. Eur Radiol. 2015;25:2184–2193.
quired last. The average time needed for image reconstruction is depen- 10. Lu W, Pauly KB, Gold GE, et al. SEMAC: slice encoding for metal artifact correc-
dent on the matrix size and processing power of the work station as well tion in MRI. Magn Reson Med. 2009;62:66–76.
as number of coil channels and iterations. Based on our initial empirical 11. Hargreaves BA, Chen W, Lu W, et al. Accelerated slice encoding for metal artifact
experience, we selected 30 iterations and a regularization parameter of correction. J Magn Reson Imaging. 2010;31:987–996.
0.005 for the non–fat-suppressed intermediate-weighted accelerated 12. Lustig M, Donoho D, Pauly JM. Sparse MRI: the application of compressed sens-
ing for rapid MR imaging. Magn Reson Med. 2007;58:1182–1195.
CS-SEMAC data sets and 20 iterations and a regularization parameter
13. Yang AC, Kretzler M, Sudarski S, et al. Sparse reconstruction techniques in mag-
of 0.008 for the STIR CS-SEMAC data sets for reconstruction, which netic resonance imaging: methods, applications, and challenges to clinical adop-
resulted in reconstruction times between 4 and 6 minutes per data set. tion. Invest Radiol. 2016;51:349–364.
Shortening of the reconstruction time can be achieved by using more 14. Worters PW, Sung K, Stevens KJ, et al. Compressed-sensing multispectral
powerful processing units and determining the minimum number of it- imaging of the postoperative spine. J Magn Reson Imaging. 2013;37(1):243–248.
erations and optimal regularization for each CS-SEMAC sequence 15. Nittka MOR, Rybak LD, Block KT, et al. Highly Accelerated SEMAC Metal
type, which will require additional studies. Implant Imaging Using Joint Compressed Sensing and Parallel Imaging.
Proceedings of the 21th Annual Meeting of ISMRM, Salt Lake City, UT,
Our study had limitations. For this first study, only asymptomatic 2013. p 2558.
subjects were included, which may have resulted in less frequent motion 16. Fritz J, Raithel E, Thawait GK, et al. Six-fold acceleration of high-spatial
artifacts than in a patient population. Although the data sets were ran- resolution 3D SPACE MRI of the knee through incoherent k-Space
domized and blinded, observers may have been able to differentiate undersampling and iterative reconstruction-first experience. Invest Radiol.
2016;51:400–409.
high-BW from SEMAC and CS-SEMAC data sets due to the signifi-
17. Bhave S, Lingala SG, Newell JD Jr, et al. Blind compressed sensing enables
cantly larger metal artifacts. To carry out the segmentation of the metal 3-dimensional dynamic free breathing magnetic resonance imaging of lung vol-
artifact on the same image slices, the 3 corresponding data sets were pre- umes and diaphragm motion. Invest Radiol. 2016;51:387–399.
sented together, which may have introduced a bias. Accurate SNR and 18. Zijlstra F, Viergever MA, Seevinck PR. Evaluation of variable density and data-
CNR measurements on images acquired with parallel imaging and non- driven k-space undersampling for compressed sensing magnetic resonance imag-
linear image reconstructions may be limited as the noise distribution can ing. Invest Radiol. 2016;51:410–419.
vary across an image. The difference method,32 which requires the sub- 19. Fushimi Y, Fujimoto K, Okada T, et al. Compressed sensing 3-dimensional time-
of-flight magnetic resonance angiography for cerebral aneurysms: optimization
traction of 2 identical data sets, was practically not feasible due to the and evaluation. Invest Radiol. 2016;51:228–235.
length of the research protocol. We therefore attempted to compensate 20. Chandarana H, Feng L, Ream J, et al. Respiratory motion-resolved compressed
for nonuniform noise distributions by averaging measurements across sensing reconstruction of free-breathing radial acquisition for dynamic liver mag-
different image regions and slices. The qualitative analysis included intact netic resonance imaging. Invest Radiol. 2015;50:749–756.
structures, joint fluid, and bone-implant interfaces, which provides sug- 21. Chandarana H, Block TK, Ream J, et al. Estimating liver perfusion from
gestive evidence of sequence performance; however, a patient study is re- free-breathing continuously acquired dynamic gadolinium-ethoxybenzyl-
diethylenetriamine pentaacetic acid-enhanced acquisition with compressed sens-
quired to determine the diagnostic performance. ing reconstruction. Invest Radiol. 2015;50:88–94.
In conclusion, we accept the hypothesis that 8-fold acceleration of 22. Sharma SD, Fong CL, Tzung BS, et al. Clinical image quality assessment of ac-
SEMAC through the implementation of compressed sensing–based inco- celerated magnetic resonance neuroimaging using compressed sensing. Invest
herent k-space undersampling and iterative reconstruction is clinically Radiol. 2013;48:638–645.
feasible for high-quality MRI of cobalt-chromium knee arthroplasty im- 23. Chandarana H, Feng L, Block TK, et al. Free-breathing contrast-enhanced
multiphase MRI of the liver using a combination of compressed sensing, par-
plants and that a compressed sensing–accelerated SEMAC prototype allel imaging, and golden-angle radial sampling. Invest Radiol. 2013;48:
with sampling times of less than 5 minutes can yield better quality than 10–16.
high-BW of similar length and similarly high quality than lengthier 24. Clark Z, Johnson KM, Wu Y, et al. Accelerated time-resolved contrast-enhanced
SEMAC pulse sequences. magnetic resonance angiography of dural arteriovenous fistulas using highly

© 2016 Wolters Kluwer Health, Inc. All rights reserved. www.investigativeradiology.com 675

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Fritz et al Investigative Radiology • Volume 51, Number 10, October 2016

constrained reconstruction of sparse cerebrovascular data sets. Invest Radiol. 28. Fritz J, Fritz B, Thawait GG, et al. Three-dimensional CAIPIRINHA SPACE TSE
2016;51:365–371. for 5-minute high-resolution MRI of the knee. Invest Radiol. 2016. [Epub ahead
25. Yamamoto T, Fujimoto K, Okada T, et al. Time-of-flight magnetic resonance an- of print].
giography with sparse undersampling and iterative reconstruction: comparison 29. Landis JR, Koch GG. The measurement of observer agreement for categorical
with conventional parallel imaging for accelerated imaging. Invest Radiol. 2016; data. Biometrics. 1977;33:159–174.
51:372–378. 30. Koch KM, Brau AC, Chen W, et al. Imaging near metal with a MAVRIC-SEMAC
26. Haubenreisser H, Henzler T, Budjan J, et al. Right ventricular imaging in hybrid. Magn Reson Med. 2011;65:71–82.
25 seconds: evaluating the use of sparse sampling CINE with iterative reconstruc- 31. den Harder JC, van Yperen GH, Blume UA, et al. Ripple artifact reduction using
tion for volumetric analysis of the right ventricle. Invest Radiol. 2016;51: slice overlap in slice encoding for metal artifact correction. Magn Reson Med.
379–386. 2015;73:318–324.
27. Stalder AF, Schmidt M, Quick HH, et al. Highly undersampled contrast-enhanced 32. Reeder SB, Wintersperger BJ, Dietrich O, et al. Practical approaches to the evalua-
MRA with iterative reconstruction: integration in a clinical setting. Magn Reson tion of signal-to-noise ratio performance with parallel imaging: application with car-
Med. 2015;74:1652–1660. diac imaging and a 32-channel cardiac coil. Magn Reson Med. 2005;54:748–754.

676 www.investigativeradiology.com © 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.