diagnostics

Review
MR-Imaging in Osteoarthritis: Current Standard of Practice and
Future Outlook
Jonathan Ehmig 1 , Günther Engel 1 , Joachim Lotz 1 , Wolfgang Lehmann 2 , Shahed Taheri 2 ,
Arndt F. Schilling 2 , Ali Seif Amir Hosseini 1 and Babak Panahi 1, *

1 Institute of Diagnostic and Interventional Radiology, University Medical Center Göttingen,

37075 Göttingen, Germany; jonathan.ehmig@med.uni-goettingen.de (J.E.);
guenther.engel@med.uni-goettingen.de (G.E.)
2 Clinic of Trauma, Orthopedics and Reconstructive Surgery, Georg-August-University of Göttingen,
37075 Göttingen, Germany
* Correspondence: babak.panahi@med.uni-goettingen.de

Abstract: Osteoarthritis (OA) is a common degenerative joint disease that affects millions of people
worldwide. Magnetic resonance imaging (MRI) has emerged as a powerful tool for the evaluation
and monitoring of OA due to its ability to visualize soft tissues and bone with high resolution.
This review aims to provide an overview of the current state of MRI in OA, with a special focus
on the knee, including protocol recommendations for clinical and research settings. Furthermore,
new developments in the field of musculoskeletal MRI are highlighted in this review. These include
compositional MRI techniques, such as T2 mapping and T1rho imaging, which can provide additional
important information about the biochemical composition of cartilage and other joint tissues. In
addition, this review discusses semiquantitative joint assessment based on MRI findings, which is a
widely used method for evaluating OA severity and progression in the knee. We analyze the most
common scoring methods and discuss potential benefits. Techniques to reduce acquisition times and
the potential impact of deep learning in MR imaging for OA are also discussed, as these technological
advances may impact clinical routine in the future.

Keywords: osteoarthritis; magnetic resonance imaging (MRI); joint disease; degenerative disease;
Citation: Ehmig, J.; Engel, G.; Lotz, J.;
Lehmann, W.; Taheri, S.; Schilling,
bone imaging; semiquantitative joint assessment; OA severity; OA progression; T2 mapping; Kellgren
A.F.; Seif Amir Hosseini, A.; Panahi, and Lawrence grading
B. MR-Imaging in Osteoarthritis:
Current Standard of Practice and
Future Outlook. Diagnostics 2023, 13,
2586. https://doi.org/10.3390/ 1. Introduction
diagnostics13152586 Osteoarthritis (OA) is the most common disease of the joint worldwide and is char-
Academic Editor: Anja Müller-Lutz acterized by a multifactorial pathogenesis resulting in pain and loss of joint function. OA
is considered a leading cause of disability and is associated with a high socioeconomic
Received: 28 June 2023
burden [1]. In 2020, 7% of the global population was affected by OA, ranking the disease
Revised: 30 July 2023
15th regarding years lived in disability worldwide [2]. Risk factors can be divided into
Accepted: 1 August 2023
two subgroups that interact to determine an individual’s risk for OA. Patient-specific risk
Published: 3 August 2023
factors include age, gender, obesity, genetics, and diet. Joint-specific risk factors include
abnormal loading of the joint, trauma, and malalignment [3]. In addition, OA and the
resulting reduced mobility can be considered a risk factor in itself, as several studies have
Copyright: © 2023 by the authors. linked it to an increased mortality from dementia and cardiovascular disease [4]. The
Licensee MDPI, Basel, Switzerland. associated economic burden resulting from treatment costs and occupational disability is
This article is an open access article estimated between 1 and 2.5% of the GNP (gross national product) in Western countries [5].
distributed under the terms and This systematic review aims to provide an overview of the value of MRI in the di-
conditions of the Creative Commons agnosis of osteoarthritis. We will consider the current standard of clinical care as well
Attribution (CC BY) license (https:// as recent developments in the field. Furthermore, we will provide recommendations for
creativecommons.org/licenses/by/ structured joint assessment and analyze currently available scoring systems. Finally, we
4.0/).

Diagnostics 2023, 13, 2586. https://doi.org/10.3390/diagnostics13152586 https://www.mdpi.com/journal/diagnostics

Diagnostics 2023, 13, 2586 2 of 19

will discuss future developments in MRI and radiology itself, especially considering how
artificial intelligence might reshape the landscape of MRI diagnostics.
The most frequently affected site of OA manifestation is the knee joint [6], thereby
accounting for a predominant focus of scientific investigations. Consequently, this review is
focused on OA of the knee, even though OA can impact various articulations in the human
body [7,8]. However, most of the statements may be applicable to other joints as well.

2. Osteoarthritis—A Whole Joint Disease

OA has long been considered a degenerative condition primarily affecting articular
cartilage. However, advances in imaging and histopathological research have led to a
paradigm shift, recognizing OA as a complex joint disorder involving not only cartilage
but also menisci, ligaments, synovia, subchondral bone, and periarticular muscle tissue [9].
It is widely known that ligamental tears or laxity as well as meniscal damage are
associated with the development of OA [10,11]. Moreover, synovial inflammation is
reported in many cases [12]. Recent studies have highlighted the critical role of subchondral
bone in OA development and progression [13–15]. Chen et al. proposed a model describing
subchondral bone loss in early OA, leading to the formation of sclerotic, less mineralized
bone with altered mechanical properties and eventual disruption of the osteochondral
junction in late-stage OA [13,16]. Interestingly, these changes have been observed in non-
arthritic joint compartments, indicating the potential pre-eminence of subchondral bone
involvement before cartilage damage occurs [13].
Furthermore, bone remodeling is significantly influenced by sex hormones [17]. In
menopausal women, a study conducted by Zoli et al. revealed a notable association between
osteoporosis and erosive OA [18]. Similarly, the Women’s Health Initiative conducted an
extensive study, which identified a connection between self-reported OA and hysterectomy
and unilateral oophorectomy. Interestingly, hormone replacement therapy in this study
appeared to exhibit a protective effect [19]. Substantiating these findings, a study conducted
by Jung et al. also supported the potential benefits of hormone replacement therapy [20].
Moreover, OA is frequently accompanied by impairment of periarticular muscle,
which significantly contributes to functional limitations. Two major mechanisms have
been identified: muscle fiber atrophy, favoring fast-twitch Type 2 fibers, and arthrogenic
muscle inhibition, which refers to reduced excitability due to alterations in joint sensory
receptors. It is noteworthy that atrophy in periarticular muscles is suspected to arise from
chronic inflammatory processes rather than merely disuse due to pain. Histopathologic
examinations found an increased amount of intramyocellular lipids probably as result of
mitochondrial degeneration and fibrotic tissue between muscle fibers [9,21–23].

3. Diagnosing Osteoarthritis
Diagnosis of OA is primarily based on a thorough clinical examination of the joint,
and imaging has always been important in detecting joint damage. Radiography has
so far played an important role in the diagnostic process even though it is limited to
the assessment of osseous structures. Additionally, patients with symptomatic OA show
radiographic changes in only about half of the cases [24]. Earlier stages and potentially
reversible changes of the joint can be detected by magnetic resonance imaging (MRI)
which allows to assess soft tissues such as cartilage, synovia, menisci, and the surrounding
muscles and ligaments [25]. However, so far MRI only plays a minor role in the primary
diagnosis of OA in clinical routine, even though its sensitivity to detect structural changes
in the joint has been confirmed in research settings [26].

3.1. Radiography
Radiography, which is still the most commonly used imaging technique for OA, is
usually acquired in two planes, i.e., the lateral and anterior-posterior view. It is widely
available and inexpensive. In addition, weight-bearing images can be obtained [27]. De-
pending on the clinical facility and the clinical patient history, additional views, such as the
 Diagnostics 2023, 13, x FOR PEER REVIEW 3 o

3.1. Radiography
Radiography, which is still the most commonly used imaging technique for OA
Diagnostics 2023, 13, 2586 3 of 19
usually acquired in two planes, i.e., the lateral and anterior-posterior view. It is wid
available and inexpensive. In addition, weight-bearing images can be obtained [27].
pending on the clinical facility and the clinical patient history, additional views, such
patella view or the
the Rosenberg
patella view view,
or thecan be obtained
Rosenberg to can
view, evaluate specifictoregions
be obtained evaluate ofspecific
the joint.regions of
Introduced by Kellgren and Lawrence
joint. Introduced in 1957,
by Kellgren andthe gradinginof
Lawrence OA the
1957, is still conducted
grading of OA is onstill
a conduc
four-grade scaleon(Figure 1). Grade
a four-grade scale0(Figure
indicates
1). the absence
Grade of OA-specific
0 indicates the absence changes in the change
of OA-specific
joint, Grade I isthe
defined as doubtful
joint, Grade OA-changes,
I is defined as doubtfuland Grades II and
OA-changes, and III referIItoand
Grades minimal
III refer to mini
and moderate
and moderate changes, changes,and
respectively, respectively, and can be distinguished
can be distinguished by the presence by the presence or abse
or absence
of subchondral of subchondral
sclerosis. sclerosis.
Finally, Grade Finally, Gradesevere
IV considers IV considers
stages severe
of OA stages of OA
associated associated w
with
joint deformity and severely reduced joint
joint deformity and severely reduced joint space width (JSW) [28].space width (JSW) [28].

Figure 1. OA Stages (Kellgren and Lawrence, 1957)—(a) Grade 0: Physiological joint. (b) Grad
Figure 1. OA Stages (Kellgren and Lawrence, 1957)—(a) Grade 0: Physiological joint. (b) Grade I:
Subtle JSN in the medial compartment with osteophytic lipping. (c) Grade II: Definite JSN in
Subtle JSN in the medial compartment with osteophytic lipping. (c) Grade II: Definite JSN in the
medial compartment. (d) Grade III: Definite JSN in the medial compartment and sclerosis of
medial compartment. (d) Grade
subchondral bone.III:(e)Definite JSN
Grade IV: inwith
JSN the amedial compartment
bone-on-bone and sclerosis
phenomenon of the of the me
and deformity
subchondral bone.tibial
(e) Grade
plateauIV:
asJSN
wellwith a bone-on-bone
as the medial femoralphenomenon
condyle. and deformity of the medial
tibial plateau as well as the medial femoral condyle.
OA grading on plain radiographs is based on the assessment of osseous tissues w
OA grading on plainthickness
cartilage radiographs is based
can only on the assessment
be evaluated indirectly of
asosseous
a measuretissues while
of JSW [27]. Howe
cartilage thickness
MRIcan onlyhave
studies be evaluated
shown that indirectly asnarrowing
joint space a measure(JSN)
of JSW [27].
is not However,
solely dependent on
MRI studies have of shown
cartilagethat joint space
thickness narrowing
but can (JSN)
rather be is not solely
considered dependentofonmeniscal
as a composite loss dama
of cartilage thickness
meniscalbut can rather
extrusion, andbecartilage
considered as a[29].
damage composite of meniscal damage,
meniscal extrusion, and cartilage damage [29].

3.2. MRI in Musculoskeletal Imaging

MRI is an established imaging technique available in most clinical institutions. Most
available scanners have preset protocols. For the imaging of cartilage, 1.5 T and 3 T scanners
do not differ in sensitivity for detecting pathologies. As imaging at higher field strengths
results in a higher signal-to-noise ratio (SNR), spatial resolution, accuracy, and specificity
are increased at 3 T [30]. Furthermore, acquisition time can be reduced at 3 T. However, it
should be noted that if orthopedic hardware is implanted close to or in the imaged region,
Diagnostics 2023, 13, 2586 4 of 19

higher field strengths can lead to an increase in susceptibility artifacts caused by magnetic
field inhomogeneities [31].
In addition to standard high-field MRI (HF-MRI) systems which typically operate at
field strengths of 1.5 T and 3.0 T, low-field MRI (LF-MRI) systems have recently gained
new attention. LF-MRI systems are available in two main configurations: standard large-
bore machines and dedicated extremity scanners. Dedicated extremity scanners have
demonstrated several advantages, including reduced noise and high patient comfort,
making them an attractive option for focused joint examinations. In addition, these scanners
have a more economical profile and offer a degree of portability, facilitating their use in
different clinical settings [32–34].
Historically, LF-MRI has faced challenges in competing with HF-MRI regarding image
resolution and contrast, limiting its diagnostic utility. However, innovative imaging pro-
tocols that exploit the unique characteristics of low-field strengths, such as shortened T1
times and longer T2 and T2* times, have significantly improved image quality. Low SNR
can be addressed by applying multiple averaging which effectively increases the overall
quality of LF-MRI images at the expense of longer acquisition times [33].
Despite these advances, few studies have directly compared contemporary LF-MRI
systems to HF-MRI counterparts in the musculoskeletal domain. Early evidence suggests
that LF-MRI performs comparably to HF-MRI in the examination of the ankle, foot, and
knee. The results for shoulder imaging have been somewhat inconsistent, with certain
studies reporting more management-changing results with HF-MRI. LF-MRI may be par-
ticularly suitable for acute injuries, but its sensitivity for smaller, chronic abnormalities
may be limited due to lower resolution. A notable advantage of LF-MRI is its reduced
susceptibility to artefacts from orthopedic hardware [33–35].
To ensure that patients receive accurate and high-quality diagnostic images for effec-
tive musculoskeletal management, decisions regarding the best-suited imaging modality
should be made by experienced personnel. As more studies are warranted to comprehen-
sively compare LF-MRI and HF-MRI, ongoing advancements in LF-MRI may shape the
future of musculoskeletal imaging, contributing to improved outcomes. This review focuses
on HF-MRI, which is more widely available due to the limitations of earlier generations of
LF-MRI [32].
In addition, image quality depends on the choice of the receiver coil. Lutterbey
et al. demonstrated that even in a high-field-strength scanner the image quality can be
impeded by a wrong choice of coil [36]. The use of a dedicated surface coil is recommended.
Furthermore, multichannel coils allow for parallel imaging, which can be used to obtain
better image quality or shorter acquisition time [37].

3.3. MR Acquisition Protocols—The Current Standard of Clinical Care

Despite the paradigm shift towards a more holistic view of OA, the current state of
imaging protocols remains tailored to evaluate the internal structures of the knee, rather
than the entire joint.
The current guidelines for MRI of the knee as published by the European Society of
Skeletal Radiology (ESSR) suggest the acquisition of fat-saturated proton density-weighted
(PDw FS) images in three standard planes as well as a T1-weighted (T1w) image in a sagittal
or coronal orientation.
After correct placement of the patient and the receiver coil (Figure 2), the workflow
starts with the acquisition of three-plane localizers. Axial images are planned parallel to
the knee joint line, while sagittal planes are planned parallel to the medial facet of the
lateral condyle. Coronal planes are arranged parallel to the posterior facets of the femoral
condyles. A contrast-enhanced T1w FS image may be added as an option if inflammation,
such as synovitis or osteitis, or malignancy is suspected. The ESSR (Table 1) recommends
a slice thickness of 3 mm for all images. The field of view (FOV) for the PDw images is
recommended to be 16 cm with a matrix of 288 × 384. T1w-images should be acquired at
an FOV of 18 cm with a matrix of 358 × 512 [38,39].
 the knee joint line, while sagittal planes are planned parallel to the medial facet of
lateral condyle. Coronal planes are arranged parallel to the posterior facets of the femo
condyles. A contrast-enhanced T1w FS image may be added as an option if inflammati
such as synovitis or osteitis, or malignancy is suspected. The ESSR (Table 1) recomme
Diagnostics 2023, 13, 2586 a slice thickness of 3 mm for all images. The field of view (FOV) for the 5 ofPDw
19 image
recommended to be 16 cm with a matrix of 288 × 384. T1w-images should be acquired
an FOV of 18 cm with a matrix of 358 × 512 [38,39].

Figure 2. PatientsFigure 2. Patients

enter the scannerenter
feet the
firstscanner feet first
in a supine in a supine
position. position.
The coil Thearound
is placed coil is placed around the k
the knee
and the joints are immobilized with adequate padding.
and the joints are immobilized with adequate padding.
Table 1. Standard knee protocol as recommended by the ESSR.
Table 1. Standard knee protocol as recommended by the ESSR.
Sequence FOV Slice Thickness TR TE Matrix
Sequence Sag PD FOV
FS Slice Thickness
160 3 TR TE 3570Matrix
39 288 × 384
Sag PD FS Cor PD FS
160 3 160 3 3570 39 3570288 ×
39384 288 × 384
Cor PD FS Ax PD FS
160 3 160 3 3570 39 3570288 ×
39384 288 × 384
Ax PD FS Cor/Sag160
T1 3 180 3 3570 39 470 288 ×
13384 358 × 512
Optional CE T1 FS 180 3 470 13 358 × 512
Cor/Sag T1 180 3 470 13 358 × 512
Optional CE T1 FS The180 3 of three planes is very
acquisition 470important13to ensure358 × compartment
each 512 can
adequately evaluated. Finally, articular cartilage is best assessed in an orthogonal ima
For example,
The acquisition of three while
planesevaluation of retropatellar
is very important to ensurecartilage is limited on coronal
each compartment can be images d
to slice thickness and partial volume effects, it is easily conducted
adequately evaluated. Finally, articular cartilage is best assessed in an orthogonal image. on axial or sagittal
For example, while evaluation of retropatellar cartilage is limited on coronal images due tointernal c
ages. PDw-images allow detailed assessment of intra-articular cartilage and
tilage
slice thickness and composition.
partial However,
volume effects, it issome
easilyinstitutions
conductedprefer intermediate-weighted
on axial or sagittal images. images
selecting a slightly longer echo time (TE) to benefit
PDw-images allow detailed assessment of intra-articular cartilage and internal from the advantages of T2-weigh
cartilage
composition. However, some institutions prefer intermediate-weighted images by selecting
a slightly longer echo time (TE) to benefit from the advantages of T2-weighted (T2w)
sequences. These are less susceptible to magic angle artifacts and improve delineation
between cartilage and synovial fluid. The overall signal on intermediate-weighted images is
increased compared to T2w images [40,41]. Additional T1w images are useful for evaluation
of the general anatomy, bone marrow lesions (BML), and subchondral sclerosis, as well as
the search for loose bodies, that may cause locking of the joint [42]. As the T1w-images are
the only non-fat-suppressed images in a standard acquisition protocol, they are particularly
useful for evaluation of bone marrow and muscles that are physiologically separated by
fatty tissues.
Diagnostics 2023, 13, 2586 6 of 19

3.4. Fat Suppression

Fat suppression can be useful to increase contrast at the osteochondral junction. Tech-
niques include fat saturation (FS), inversion recovery (IR) imaging, and in- and opposed-
phase imaging.
FS uses a fat-specific 90◦ pulse immediately before the imaging sequence, tipping
the protons in fatty tissues out of plane and suppressing their signal. FS relies on a
homogeneous magnetic field and is very sensitive to inhomogeneities such as those caused
by metallic implants. Fat suppression also has limitations owing to the signal from water
in adipose tissue, and from a minor portion of fatty acids that have a resonance frequency
equal to that of water. Fat saturation pulses are administered for approximately 10 ms,
which can lead to a substantial increase in acquisition time. At low field strength, the
difference in resonance frequencies between water and fat is reduced, resulting in incorrect
tissue discrimination. Since radiofrequency pulses are tissue-specific, they can be used
universally even after administration of a contrast agent [43].
IR-imaging achieves fat suppression by inverting spins along the z-axis and imaging
at a specific time point where the protons of fatty tissues have no longitudinal magneti-
zation at all. This technique is fast and less prone to magnetic field inhomogeneities [42].
However, the overall SNR is reduced and tissues with similar T1 compared to fat are
equally suppressed.
In- and opposed-phase imaging is based on the differing precession frequencies
of fat and water protons. Immediately after excitation, protons are in-phase and start
dephasing until they reach a 180◦ dephasation. Echo time can be adapted in order to either
acquire in-phase images with signal from both fat and water or opposed-phase images that
show a subtraction of the two [43]. The Dixon technique is based on acquisition of both
types of images and results in a pure water or pure fat image achieved by addition and
subtraction of in- and opposed-phase images. Although sensitive to small amounts of fat,
this technique does not suppress the signal from pure fat and is sensitive to magnetic field
inhomogeneites [44,45].
Most commercially available MRI scanners today have field strengths of either 1.5 or
3.0 T, placing them in the high-field range. This allows sufficient accuracy of fat-specific
radiofrequency pulses and makes FS the most commonly used technique for MRI of the
joint with preserved efficacy on contrast-enhanced scans. Given its sensitivity to magnetic
field inhomogeneities, IR should be considered as an alternative in the presence of metallic
implants. Moreover, IR can be chosen as an effective fat-suppression technique in low-field
systems. The Dixon technique can be used for the detection of small amounts of fat as
well as for fat quantification, which may be helpful in case of tumors or neuromuscular
disorders. In this case, as explained above, two images must be acquired, which doubles
the acquisition time.

4. Magnetic Resonance Imaging—Common Findings

While radiography visualizes relatively late and mostly irreversible stages of OA,
MRI has the potential to detect pre-osseous deformities. A common MRI finding in OA
is articular cartilage damage. Bone marrow lesions (BML) in the subchondral bone and
subarticular cysts reflect extensive remodeling of the subchondral bone [13,46]. BML
appear as PDw-hyperintense lesions and can be found in early OA. A study by Muratovic
et al. found that BML that also appear on T1w-images are associated with late OA [47].
Subarticular bone loss and marginal osteophytes may also be seen. In the surrounding
tissues, damage to the cruciate and collateral ligaments as well as the menisci may be found.
Periarticular cysts, synovitis, joint effusion, and loose intraarticular bodies are considered
diagnostic [12,48]. Figure 3 shows a standard MRI protocol for the knee comparing a
physiological joint to a severe OA of the medial compartment.
 as PDw-hyperintense lesions and can be found in early OA. A study by Muratovic et al.
found that BML that also appear on T1w-images are associated with late OA [47]. Subar-
ticular bone loss and marginal osteophytes may also be seen. In the surrounding tissues,
damage to the cruciate and collateral ligaments as well as the menisci may be found. Peri-
Diagnostics 2023, 13, 2586
articular cysts, synovitis, joint effusion, and loose intraarticular bodies are considered di-
7 of 19
agnostic [12,48]. Figure 3 shows a standard MRI protocol for the knee comparing a phys-
iological joint to a severe OA of the medial compartment.

Figure 3. Standard protocol for knee evaluation. Healthy knee joint (a–d) vs. severe OA of the medial
Figure 3. Standard protocol for knee evaluation. Healthy knee joint (a–d) vs. severe OA of the
compartment (e–h) with BML (arrowheads), meniscal extrusion (star), and small osteophytes (ar-
medialThe
rows). compartment (e–h) with
articular cartilage BMLirregular
shows (arrowheads), meniscal
thickness extrusion
with some (star), andlesions
full thickness small osteophytes
(g).
(arrows). The articular cartilage shows irregular thickness with some full thickness lesions (g).
OA has been shown to progress in different patterns, which were characterized by
OA has been shown to progress in different patterns, which were characterized by
Deveza et al. as clinical and structural OA-phenotypes [49]. Two structural phenotypes
Deveza et al. as clinical and structural OA-phenotypes [49]. Two structural phenotypes
were identified based on the extent of osteophyte formation: the hyper- and the atrophic
were identified based on the extent of osteophyte formation: the hyper- and the atrophic
type. The inflammatory phenotype is characterized by joint effusion and synovitis, and
type. The inflammatory phenotype is characterized by joint effusion and synovitis, and
the subchondral type is defined by BML. The meniscal phenotype shows meniscal dam-
the subchondral type is defined by BML. The meniscal phenotype shows meniscal damage
age
and and extrusion
extrusion on MRI
on MRI [50].[50]. Future
Future treatment
treatment options
options maymay
needneed to take
to take thesethese pheno-
phenotypes
types into account and become more individualized. Therefore, careful patient selection
into account and become more individualized. Therefore, careful patient selection may be
may be required
required for designing
for designing of trials.
of clinical clinical trials.

5.
5. Additional
Additional MRI-Techniques
MRI-Techniques
Three-Dimensional Image Acquisition
5.1. Three‐Dimensional
The acquisition
acquisition of of3D
3Ddata
datasets
setswith
withisotropic
isotropicvoxels enables
voxels enables thethe
radiologist to perform
radiologist to per-
multiplanar
form reformation
multiplanar (MPR)(MPR)
reformation of the images, allowing
of the images, the reconstruction
allowing of any desired
the reconstruction of any
angulation.
desired Volumetric
angulation. analysisanalysis
Volumetric may be may
performed in orderintoorder
be performed assesstocartilage defects de-
assess cartilage for
individualized
fects orthopedicorthopedic
for individualized treatment [51]. Sequences
treatment [51]. used for thisused
Sequences purpose include
for this Double-
purpose in-
Echo-Steady-State
clude (DESS), SPoiled
Double-Echo-Steady-State GRadient
(DESS), EchoGRadient
SPoiled (SPGR), andEcho Sampling
(SPGR),Perfection with
and Sampling
Application optimized
Perfection with Contrasts
Application using different
optimized Contrasts flipusing
angledifferent
Evolutions flip(SPACE)-sequences.
angle Evolutions
SPGR is capable
(SPACE)-sequences. of acquiring high-resolution 3D images with nearly isotropic voxels.
The technique is based on a “spoiler” pulse that dephases protons
SPGR is capable of acquiring high-resolution 3D images with nearly isotropic to reduce transverse
voxels.
magnetization
The technique is and create
based onthe impression
a “spoiler” of T1w
pulse that and PDw images.
dephases protons The disadvantage
to reduce transverseof
this technique is
magnetization andthecreate
long acquisition time of
the impression of about
T1w and 10–12
PDwminimages.
due to The
a lowdisadvantage
signal-to-noise
of
ratiotechnique
this (SNR). Inisaddition, being a gradient-echo
the long acquisition time of about sequence,
10–12 min it is sensitive
due to a lowtosignal-to-noise
susceptibility
artifacts [42,52].
ratio (SNR). In addition, being a gradient-echo sequence, it is sensitive to susceptibility
DESS
artifacts acquires two signals after reaching a steady state between longitudinal and
[42,52].
transversal magnetization (Figure 4). The first is a post excitation signal after a free induc-
tion decay, while the second is acquired after a refocusing pulse. Both signals are then
combined by a sum of squares. Thus, the contrast in DESS images is determined by the
T1/T2 ratio [53]. DESS has the potential to calculate estimated T2 maps, and therefore may
provide an opportunity to combine compositional and quantitative assessment in a single
acquisition [54].
 DESS acquires two signals after reaching a steady state between longitudinal and
transversal magnetization (Figure 4). The first is a post excitation signal after a free induc-
tion decay, while the second is acquired after a refocusing pulse. Both signals are then
combined by a sum of squares. Thus, the contrast in DESS images is determined by the
T1/T2 ratio [53]. DESS has the potential to calculate estimated T2 maps, and therefore may
Diagnostics 2023, 13, 2586 8 of 19
provide an opportunity to combine compositional and quantitative assessment in a single
acquisition [54].

Figure 4. Coronal DESS image of the right knee.

Figure 4. Coronal DESS image of the right knee.

Three-dimensional-SPACEimages
Three-dimensional-SPACE imageshave
haveaahigher
higherSNR
SNRthan
thanother
other3D
3Dtechniques
techniquesand
and
deliver a high T2w contrast [55]. K-space under-sampling and novel parallel imaging
deliver a high T2w contrast [55]. K-space under-sampling and novel parallel imaging appli- ap-
plications have recently allowed reducing acquisition time to just under 10 min while
cations have recently allowed reducing acquisition time to just under 10 min while maintain-
maintaining
ing diagnosticdiagnostic performance
performance and imageand image
quality quality to
compared compared to a 2D
a 2D standard standard
protocol pro-
[56,57].
tocol [56,57].
5.2. UTE-/ZTE-Imaging
5.2. UTE‐/ZTE‐Imaging
A common challenge in joint imaging is the assessment of tissues with very short T2
times.AEspecially in OA, the
common challenge in assessment
joint imaging of isshort T2 tissues, of
the assessment such as menisci,
tissues with veryligaments,
short T2
calcified cartilage, in
times. Especially andOA,subchondral bone of
the assessment is of paramount
short importance.
T2 tissues, Recently,
such as menisci, Roemer
ligaments,
et al. demonstrated
calcified cartilage, andan inferior assessment
subchondral bone isof of osteophytes
paramountonimportance.
MRI compared to CTRoemer
Recently, [58].
et al. demonstrated an inferior assessment of osteophytes on MRI compared to CT [58].an
Subchondral bone as the anchorage for articular cartilage is suspected to play
important role in the
Subchondral pathogenesis
bone of OA.for
as the anchorage The osteochondral
articular cartilagejunction consists
is suspected of the
to play ansub-
im-
chondral boneinplate
portant role and a calcifiedoflayer
the pathogenesis OA.ofThecartilage. Conventional
osteochondral junctionimaging typically
consists of the uses
sub-
echo timesbone
chondral in the millisecond
plate range,layer
and a calcified making it impossible
of cartilage. to discriminate
Conventional these
imaging structures.
typically uses
Ultra-short-echo
echo times in the time (UTE) imaging
millisecond and zero-echo
range, making timeto
it impossible (ZTE) imagingthese
discriminate are techniques
structures.
used to acquire signal
Ultra-short-echo time from
(UTE)these tissues,
imaging andthat has mostly
zero-echo timedecayed on conventional
(ZTE) imaging MRI.
are techniques
On inversion recovery UTE/ZTE-images, the osteochondral junction
used to acquire signal from these tissues, that has mostly decayed on conventional MRI. can be seen as a
band of high signal intensity, that may appear disrupted in OA.
On inversion recovery UTE/ZTE-images, the osteochondral junction can be seen as a band As bone has one of the
lowest T2 values in the human body, ZTE imaging can be used to accurately
of high signal intensity, that may appear disrupted in OA. As bone has one of the lowest depict osseous
tissues [59].inThe
T2 values theimages
humanhave body,a CT-like appearance
ZTE imaging can beand
usedmight have the potential
to accurately to further
depict osseous tis-
replace bone CT with comparable resolution of 0.8–1.2 mm [60]. Furthermore,
sues [59]. The images have a CT-like appearance and might have the potential to further UTE imaging
allows
replacefor T2*,CT
bone T1, andcomparable
with T1rho mapping and can
resolution hence mm
of 0.8–1.2 be used
[60].toFurthermore,
evaluate cartilage and
UTE imag-
monitor therapy efficacy [61,62].
ing allows for T2*, T1, and T1rho mapping and can hence be used to evaluate cartilage
and monitor therapy efficacy [61,62].
6. Functional Assessment on Real-Time MRI
Real-time MRI sequences allow the visualization of joint movement and function.
With this technology, joint motion can be observed in real-time, providing valuable in-
sights into joint disorders and abnormalities [63]. Research into automated bone-tracking
could provide biomechanical data and potentially replace fluoroscopic or ultrasonographic
examinations [64].

6.1. Quantitative MRI

Quantitative assessment of joint structures depends on the acquisition of 3D data
sets. These techniques allow volumetric analysis of articular cartilage and measurements
Diagnostics 2023, 13, 2586 9 of 19

of cartilage thickness as well as the area of denuded bone among many others. A novel
approach investigated by Bowes et al. is the calculation of a so-called “b-score”, i.e., a single
parameter reflecting the change in bone shape using statistical shape modeling [65]. Carti-
lage volumetrics requires cartilage segmentation. This can be achieved in a time-consuming
manual process. However, implementation of DL-based segmentation algorithms might
facilitate this significantly in the future. Novel algorithms are based on Convolutional
Neural Networks (CNN) that have been tested on both SPGR and DESS data sets [66].
Recent studies by Desai et al. showed that the four best performing CNNs were equally
good at segmenting cartilage and menisci compared to manual segmentation [67].

6.2. Compositional MRI

The imaging techniques mentioned above focus mainly on a morphologic assessment
of cartilage acquiring parameters such as cartilage thickness, volume, etc. However, several
techniques have been developed in recent years to detect changes in the cartilage matrix
that occur before changes in cartilage thickness and volume are detectable. Available
techniques include T2 mapping, T1rho mapping, dGEMRIC, DWI (diffusion-weighted
imaging), Sodium imaging, and gagCEST. Both sodium imaging and gagCEST benefit from
ultra-high field strength scanners (>3 Tesla) and are hence in our opinion less likely to find
their way into clinical routine in the near future.

6.3. T2 Mapping
The articular cartilage matrix comprises a collagen network and proteoglycans (PG).
Early stages of OA have been shown to exhibit disruption of that matrix with a concomitant
decrease in water and proteoglycans in cartilage [68]. This decrease in water content can
be noticed as alteration in T2-weighted images, albeit with a lack of sensitivity [69]. Lüsse
et al. demonstrated a greater sensitivity of T2-relaxation-time measurements, known as
T2 mapping, for minimal changes in cartilage [70]. Several studies have confirmed the
diagnostic accuracy of the technique [71]. Images are acquired via a multiecho spin echo
technique to measure T2 values. Those are displayed as a grayscale or color-encoded map
highlighting areas at risk for OA (Figure 5) [42,70]. T2 maps can be acquired with most
commercially available scanners. However, as Koff et al. demonstrated, T2 measurements
Diagnostics 2023, 13, x FOR PEER REVIEW 10 of 21
do not correlate with manifestation of OA as seen on radiography, and thus are limited for
early-stage detection of OA [72].

Figure5.5.T2
Figure T2map
mapof
offemoral
femoraland
andpatellar
patellarcartilage
cartilagein
inthe
theknee
knee(Image
(Imagekindly
kindlyprovided
providedby
bySiemens
Siemens
Healthineers). The colorbar on the right indicates the T2 relaxation time.
Healthineers). The colorbar on the right indicates the T2 relaxation time.

6.4.
6.4.T1-Rho
T1‐RhoMapping
Mapping
T1rho
T1rho relaxation
relaxation is also known
known asasspin-lattice
spin-latticerelaxation
relaxationininthe
the r(h)otational
r(h)otational frame.
frame. In
In contrast
contrast to to conventional
conventional T1T1 measurements,
measurements, anan additional
additional spin-locking,
spin-locking, long-duration
long-duration ra-
radiofrequency pulseisisapplied,
diofrequency pulse applied,that
thatlocks
locks spins
spins in the transverse
transverse plane.
plane. T1rho
T1rho measures
measures
the interaction between motion-restricted water molecules and their molecular environ-
ment and is sensitive to changes in the extracellular matrix [42,73]. Like in T2 mapping,
color-encoded maps can be computed that display altered cartilage as an increase in T1rho
relaxation time. T1rho relaxation time shows strong correlation with proteoglycan con-
Diagnostics 2023, 13, 2586 10 of 19

the interaction between motion-restricted water molecules and their molecular environ-
ment and is sensitive to changes in the extracellular matrix [42,73]. Like in T2 mapping,
color-encoded maps can be computed that display altered cartilage as an increase in T1rho
relaxation time. T1rho relaxation time shows strong correlation with proteoglycan con-
centration but is also influenced by other factors such as collagen fiber orientation [26,74].
The disadvantages of this technique are long acquisition times and the need for special
pulse sequences.

6.5. dGEMRIC
dGEMRIC (short for delayed Gadolinium enhanced MRI of Cartilage) is an imaging
technique that requires intravenous or intraarticular administration of an anionic gadolin-
ium (Gd)-based contrast agent. Articular proteoglycans carry a large number of negatively
charged sidechains, called glycosaminoglycans (GAG). The theory is that after a fixed delay
(i.e., 90 min after injection, including 10 min of joint movement), negatively charged Gd
ions diffuse into the cartilage and are repelled by negatively charged GAG. Accordingly, in
areas of GAG depletion, Gd ions accumulate, resulting in higher T1 values at readout. T1
mapping is one way to display the dGEMRIC indices [42]. Even though the technique is
well validated, total acquisition time ranges around two hours owing to the long waiting
time. Additionally, it requires contrast administration with the risk of adverse events, such
as nephrogenic systemic fibrosis or anaphylactoid reaction. Some studies have also investi-
gated dGEMRIM (delayed Gadolinium enhanced MRI of the Menisci) for the evaluation
of menisci [75,76]. This technique is not well validated. A recent study by Hangaard et al.
showed no correlation of meniscal damage and dGEMRIM [77].

6.6. DWI
These sequences first acquire a T2*w image. Then, opposing gradient magnetic fields
are successively applied. This causes phase changes in the protons. Protons in healthy
cartilage are relatively constrained in their motion by the surrounding matrix components.
Thus, they experience two opposing gradient magnetic fields that result in a net zero phase
shift. Freely moving protons such as in structurally damaged cartilage, on the other hand,
acquire motion-induced phase i.e., two gradients that do not cancel each other out. DWI is
to date not well validated in cartilage imaging but has the potential to monitor the efficacy
of the therapy and cartilage quality. Diffusivity is expressed as the apparent diffusion
coefficient (ADC). Nevertheless, ADC measurements are only relative to the previously
acquired image and do not represent absolute diffusion [78–81].

6.7. gagCest Imaging

gagCEST, short for “glycosaminoglycan chemical exchange saturation transfer”, is
a property of endogenous diamagnetic metabolites with interchangeable protons such
as the negatively charged side chains of articular proteoglycans. Imaging is based on a
long saturating radiofrequency pulse at the resonance frequency of interchangeable GAG
protons resulting in zero net magnetization. Proton interchange occurs with adjacent water
molecules until a steady state is reached. Image readout can be performed by using normal
acquisition parameters, allowing for measurement of decreased water signal as a function
of GAG abundance [82].
This technique has limitations due to magnetic field inhomogeneities and spillover
effects, i.e., radiofrequency pulses are not highly selective resulting in an effect on water
molecules. Spillover can be reduced at higher field strengths such as 7T MRI, which allows
for a more selective GAG pulse. While contrast has been described as negligible at 3T,
7T enables gagCEST mapping. gagCEST may provide a useful diagnostic tool to detect
onset of early OA and is sensitive to minor changes in cartilage. However, the technique
has not been confirmed in larger cohort studies [83]. Identification of CEST-compatible
metabolites may lead to more applications of the technique, e.g., acidoCEST-MRI that
allows for determination of articular acidity [84].
Diagnostics 2023, 13, 2586 11 of 19

6.8. Sodium Imaging

As previously described, PG are large molecules with negatively charged side chains.
The negative charges need to be compensated by positive ones. Sodium is abundant in
the extracellular space and is therefore highly concentrated in the interstitium of cartilage.
Sodium—just like hydrogen—performs a precessional motion with a characteristic Larmor
frequency of 11.3 MHz/Tesla and can therefore be specifically excited using radiofrequency
pulses. Images can be acquired at very short TE. Limitations of the method include a low
SNR owing to the lower sensitivity of radiofrequency pulses, lower abundance compared
to hydrogen, and the limited hardware availability. Moreover, partial volume effects from
synovial fluid can be challenging. As sodium follows a biexponential decay with very low
T2 time, the technique requires acquisition at ultra-short echo times [85,86].

6.9. Semiquantitative Scoring Methods

In an effort to standardize joint assessment, several semiquantitative scoring methods
have been established (Table 2). Going back to a time when MRI was not an option,
visualization of cartilage could be achieved arthroscopically. In the 1960s, the Outerbridge
classification system was introduced, which graded cartilage defects according to their
depth as seen on arthroscopy [87]. With the advent of MRI, the scoring system was
modified to meet the needs of radiologists. Grade 1 refers to focal areas of hyperintensity
with normal cartilage contours, grade 2 includes cartilage defects up to 50% of the cartilage
depth, grade 3 describes a defect >50%, and grade 4 a full thickness lesion with denudation
and reactive changes of the underlying bone [88].

Table 2. Semiquantitative scoring methods for Knee MRI.

Number of
Intrarater Interrater
Scoring Method Features Assessed Compartments
Kappa Kappa
Assessed
volume fill of cartilage defect, integration into
MOCART 0.57–0.87 0.57–1.0 adjacent cartilage, surface, structure signal intensity, -
bony defect (overgrowth, subchondral changes)
cartilage, BML, subarticular cysts, subarticular
bone attrition, osteophytes, meniscal integrity,
anterior and posterior cruciate ligament integrity,
WORMS 0.61–0.99 (ICC) 15
medial and lateral collateral ligament integrity,
synovitis, loose bodies, and periarticular
cysts/bursae
BML, cartilage, osteophytes, synovitis effusion,
BLOKS 0.51–0.79 meniscal abnormalities, ligaments, 9
periarticular features
cartilaginous lesions, osteophytes, subchondral
KOSS 0.56–0.91 0.63–0.91 cysts, bone marrow edema, meniscal abnormalities, 9
effusion, synovitis, and Baker’s cyst
BML, cartilage, synovitis, osteophytes, effusion,
MOAKS 0.42–1.0 0.36–1.0 14
menisci, ligaments, periarticular features
cartilage, BML, osteophytes, menisci, inflammation
ROAMES 0.92–1.0 0.85–1.0 3
(Hoffa-synivitis, effusion)
Synovitis score 0.67–1.0 0.67–0.92 synovial thickness 9

To create a scoring method to monitor treatment efficacy in non-OA cartilage defects,

orthopedic surgeons and radiologists created the MOCART (Magnetic Resonance Observa-
tion of Cartilage Repair Tissue) score, which is currently available in its revised version,
the MOCART 2.0 [89]. The score evaluates the articular surface by scoring the volume of
cartilage defect filling, integration with adjacent cartilage, the surface and structure of the
Diagnostics 2023, 13, 2586 12 of 19

repair tissue, and signal-intensity of repair tissue on PD-weighted-images. The maximum

score is 100 points. The intra- and interreader reliability have been demonstrated to be good
for experts, albeit rather poor for inexperienced readers. The performance of inexperienced
readers could be improved by using an atlas published along with the scoring method [89].
Aiming to implement a whole-organ scoring method with a focus on articular struc-
tures believed to be involved in the pathogenesis of OA, Peterfy et al. published WORMS
(Whole-Organ Magnetic resonance imaging Score) in 2004. The authors defined 14 compart-
ments of the knee and scored each compartment on a predefined scale for cartilage integrity,
BML, cysts, bone attrition, and osteophytes. Additionally, synovitis/effusion, loose bodies,
and synovial cysts were scored on a three-grade scale. The final score was defined as the
sum of points assigned to each of these features with a maximum of 332 points [48].
A similar approach was taken by Kornaat et al. in 2005 with the introduction of the
Knee Osteoarthritis Scoring System (KOSS). The KOSS is based on a simplified assessment
of cartilage defects, osteophytes, subchondral cysts, BML, effusion, and meniscal lesions,
but does not assess surrounding ligaments [90].
In 2008, Hunter et al. reported that the effect size and the standard response means
of WORMS were small [91], resulting in the development of BLOKS (Boston–Leeds Os-
teoarthritis Knee Score). Here, assessment scales were adapted, which led to a better
correlation with pain on the Visual Analogue Scale (VAS). BLOKS assesses nine intraarticu-
lar regions and includes eight items rated on a three-point scale [92].
Applied to data from the Osteoarthritis Initiative, BML-measurement in BLOKS turned
out to be complex and somewhat redundant as well as inferior to WORMS. On the other
hand, meniscal scoring was superior in BLOKS compared to WORMS [93]. Consequently,
MOAKS was developed by Hunter et al. in 2011 in order to facilitate assessment and evolve
existing tools towards a novel scoring method with a high interrater reliability. MOAKS
considers the same 14 subregions as WORMS scored on a scale from 0–3. The scored items
are BML/cysts, articular cartilage, osteophytes, Hoffa’s synovitis and synovitis-effusion,
menisci, ligaments/tendons, and periarticular features [94].
All of the above scoring methods use Non-Contrast-Enhanced (NCE) Images. For
synovitis, a study by Roemer et al. showed superior precision of Dynamic Contrast
Enhanced (DCE)-MRI [95]. In 2011, Guermazi et al. published a semiquantitative scoring
method for synovitis. The method assesses nine joint sites and grades synovitis from 0 to 2
depending on maximum synovial thickness. In the presence of loose bodies or Baker cysts,
these sites are scored additionally [96].
In 2020, Roemer et al. introduced a scoring method for the rapid determination of
OA phenotypes, called ROAMES (Rapid OsteoArthritis MRI Eligibility Score). The tool
is based on a three-compartmental approach (PFJ—patellofemoral joint, MTFJ—medial
tibiofemoral joint, LTFJ—lateral tibiofemoral joint) determining the maximum grades for
cartilage lesions, BML, osteophytes, menisci, and inflammation. These items were adapted
from the MOAKS and WORMS scoring systems [97].
The treatment of OA relies on the effective collaboration of several clinical disciplines.
Therefore, it is important to agree on a scoring method that is understood by all involved
professions. All of the above scoring methods have been shown to be highly reproducible
and can be used as outcome measures in clinical trials.
Here, we performed a systematic evaluation of clinical trials and randomized con-
trolled trials over the last 5 years with MRI-parameters as an outcome measure using
PubMed- and Medline-Databases. Searching for the keywords “osteoarthritis”, “MRI”, and
“knee” returned 131 results, of which 97 utilized articular, Knee-MRI-based measures as an
outcome parameter (Figure 6, Table S1). Publications that did not use Knee-MRI features as
an outcome parameter were excluded along with studies that used MRI diagnosis as an
inclusion criterion. Reports on planned, but not yet conducted trials were also excluded.
Of the remaining trials, 15 relied on the WORMS and 11 on MOAKS, being the most
commonly used semiquantitative scores. KOSS and BLOKS were not used in any of the
trials. A four-grade cartilage assessment, like the modified Outerbridge approach, was
Diagnostics 2023, 13, 2586 13 of 19

used in seven cases. Eight studies investigating cartilage treatment used the MOCART
score. Overall, most trials analyzed did not use any of the structured, semiquantitative
Diagnostics 2023, 13, x FOR PEER REVIEW 14 of 21
assessments, but evaluated independent items, such as cartilage thickness (n = 14), cartilage
volume (n = 15), or BML-volume (n = 5). Eight trials used T2 maps for measurements.

Figure 6.
Figure 6. Literature
Literature review
review for
for commonly
commonly used
used scoring
scoringmethods
methodsand
andMRI
MRIoutcome
outcomeparameters.
parameters.

6.10.In
Reduction in Acquisition
our experience, Time
the ICRS-approach is the most practical score for clinical reporting.
However,
Despite the different acquisition times would
unless otherwise established, we recommend
of various protocolsMOAKS for whole-organ
in use, efforts have been
disease monitoring, especially in a trial setting. MOCART can be
made to further accelerate imaging. Over the last decades, prevalence andused as a powerful tool to
incidence of
assess the efficacy of surgical treatment of chondral defects. As therapy becomes
OA showed a substantial increase due to a number of risk factors [98] creating a growing more indi-
vidualized,
need for MRI phenotypic characterization
examinations. Reducingmay become times
acquisition more important.
may lead to Currently, ROAMES
an increase in the
can
totalbenumber
considered
of as the best option
examinations andforreduced
phenotyping withlists,
waiting a sensitivity of 86% [97].
which would subsequently
enable further integration of MRI into clinical practice.
6.10. Reduction in Acquisition Time
A first step towards faster acquisition times was proposed by Hutchinson et al. in
1988:Despite
The usetheof different
a multiple acquisition timescoil
detector array of various protocols
placed around theinpatient
use, efforts havespatial
to allow been
made to further accelerate imaging. Over the last decades, prevalence and incidence of
localization of signal, and to reduce the number of phase encoding steps to encode spatial
OA showed a substantial increase due to a number of risk factors [98] creating a growing
information [99]. Parallel imaging—like most acceleration techniques—is based on
need for MRI examinations. Reducing acquisition times may lead to an increase in the
undersampling of k-space. The acceleration factor R represents the reduction in
total number of examinations and reduced waiting lists, which would subsequently enable
acquisition time. As less signal is acquired during the acquisition process with increasing
further integration of MRI into clinical practice.
R, the SNR decreases, consequently limiting the possible acceleration [100].
A first step towards faster acquisition times was proposed by Hutchinson et al. in
Simultaneous multislice (SMS) is another parallel imaging approach that allows
1988: The use of a multiple detector array coil placed around the patient to allow spatial
acquisition of signal from multiple slices within one repetition time (TR). As TR is usually
localization of signal, and to reduce the number of phase encoding steps to encode spa-
much longer than TE, several acquisitions can be fitted into one TR by applying multiple
tial information [99]. Parallel imaging—like most acceleration techniques—is based on
slice-selective excitations. Unlike the in-plane technique mentioned above, there is only a
undersampling of k-space. The acceleration factor R represents the reduction in acquisition
marginal loss of SNR here [101].
time. As less signal is acquired during the acquisition process with increasing R, the SNR
Further
decreases, acceleration
consequently of MRI
limiting theacquisition can be achieved
possible acceleration [100]. by a technique called
Compressed Sensing (CS). This relatively new approach
Simultaneous multislice (SMS) is another parallel imaging approach is also based on the
that allows ac-
undersampling of k-space data. This undersampling is performed in a pseudo-random
quisition of signal from multiple slices within one repetition time (TR). As TR is usually
pattern
much acquiring
longer more
than TE, samples
several from thecan
acquisitions center of the
be fitted intok-space
one TRthan the surrounding
by applying multiple
slice-selective excitations. Unlike the in-plane technique mentioned above, reconstruction
regions. This results in image noises that can be denoised via an iterative there is only a
algorithm.
marginal Luckily,
loss of SNRCS-effectiveness
here [101]. is increased in 3D acquisitions, which are the most
timeFurther
consuming [102].
acceleration of MRI acquisition can be achieved by a technique called Com-
pressed Sensing (CS). This relatively new approach is also based on the undersampling
7. Recent Developments—The Advent of Deep Learning
Recent developments in the field of deep learning (DL) have led to another
substantial decrease in acquisition time. Models can be trained with preexisting data sets
enabling them to reconstruct images from undersampled data while preserving or even
improving image quality [103]. Additionally, Müller-Franzes et al. found that DL has the
Diagnostics 2023, 13, 2586 14 of 19

of k-space data. This undersampling is performed in a pseudo-random pattern acquiring

more samples from the center of the k-space than the surrounding regions. This results
in image noises that can be denoised via an iterative reconstruction algorithm. Luckily,
CS-effectiveness is increased in 3D acquisitions, which are the most time consuming [102].

7. Recent Developments—The Advent of Deep Learning

Recent developments in the field of deep learning (DL) have led to another substantial
decrease in acquisition time. Models can be trained with preexisting data sets enabling them
to reconstruct images from undersampled data while preserving or even improving image
quality [103]. Additionally, Müller-Franzes et al. found that DL has the potential to accu-
rately quantify T2 relaxation times, resulting in an accelerated acquisition of T2 maps [104].
Other approaches use deep learning methods in order to reconstruct higher-resolution
images at regular acquisition times [105]. A study by Hammernik et al. demonstrated a
fourfold acceleration using this approach [106]. However, DL reconstruction has to date
not been sufficiently evaluated for diagnostic accuracy, rather only for image appearance,
which makes further research in the field necessary [107]. On the other hand, Johnson et al.
demonstrated diagnostic equivalence of DL-reconstructed images at a twofold reduction in
scan time [108].
The three techniques mentioned above can be combined to achieve further acceleration.
Some vendors offer sequences that allow acquisition of a standard protocol in just below
two minutes.
The increasing workload, as more scans can be performed in the clinical routine, could
be compensated in the future by implementing deep learning for joint assessment into the
clinical workflow. A recent publication by Kijowski et al. reviews several studies on this
topic and looks forward to future developments in this field [66].
A recent study by Kulseng et al. demonstrated that CNNs are highly capable of
correctly identifying joint structures [109,110]. Similar studies on cartilage lesion detection
and cartilage segmentation have been very promising and with quality close to radiology
fellows [111]. Although further research is necessary, another application for DL is OA risk
assessment. Recent findings suggest limited diagnostic performance of existing models, but
future combination of clinical and imaging information prospectively, even from different
time points and modalities, is quite encouraging [66].
Despite the optimism about these new developments, future studies should focus
on this topic as the existing data are still insufficient. Research is currently limited by a
lack of training data sets, i.e., MRI scans that have been manually annotated to train a DL
algorithm. In addition, in contrast to classic machine learning where rules are defined
by a human programmer, DL algorithms define these rules themselves. These need to be
carefully examined to analyze the reliability of such self-generated rulesets [111,112].

8. Outlook
Imaging in OA continues to evolve and will prospectively be dominated by MRI.
The latest advancements in LF-MRI represent a significant stride towards enhancing MRI
accessibility, particularly in middle-to-low-income countries. Detection of early changes
in the joint and the surrounding tissues may allow to start curative therapies on time,
halting the course of the disease and reducing the socioeconomic impact of OA. Future
imaging protocols might incorporate additional sequences for comprehensive whole-organ
assessments, encompassing biomechanics, cartilage composition, and surrounding tis-
sues. Three-dimensional acquisition bears the potential to replace standard multiplane
protocols that are currently in use. With the advent of deep-learning-based methods, quan-
titative measurements might change the way we monitor disease progression and measure
treatment efficacy.
Nevertheless, deep learning may soon play an important role in image reconstruction
with either increasing image quality or decreasing acquisition time substantially. Prospec-
tively, this will lead to a greater availability of MRI and an increase in the number of
Diagnostics 2023, 13, 2586 15 of 19

examinations performed. Thus, MRI may gain further importance in the diagnostic process
of OA. The increasing workload may soon be offset by deep-learning-based algorithms that
can alert the radiologist to cartilage lesions or even assess the individual risk for OA pro-
gression. With deep learning currently being heavily researched and multiple approaches
being explored, we can expect the landscape of MRI in OA to be reshaped soon.

Supplementary Materials: The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/diagnostics13152586/s1, Table S1: Literature Review.
Author Contributions: Conceptualization, G.E. and B.P.; investigation, J.E.; writing—original draft
preparation, J.E.; writing—review and editing G.E., A.F.S. and S.T.; visualization, J.E.; supervision,
A.S.A.H., W.L. and J.L.; project administration, B.P. All authors have read and agreed to the published
version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Acknowledgments: Figure 5 was kindly provided by Siemens Healthineers without influence on the
content of this review.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Glyn-Jones, S.; Palmer, A.J.R.; Agricola, R.; Price, A.J.; Vincent, T.L.; Weinans, H.; Carr, A.J. Osteoarthritis. Lancet 2015,
386, 376–387. [CrossRef] [PubMed]
2. Hunter, D.J.; March, L.; Chew, M. Osteoarthritis in 2020 and beyond: A Lancet Commission. Lancet 2020, 396, 1711–1712.
[CrossRef] [PubMed]
3. Palazzo, C.; Nguyen, C.; Lefevre-Colau, M.-M.; Rannou, F.; Poiraudeau, S. Risk factors and burden of osteoarthritis. Ann. Phys.
Rehabil. Med. 2016, 59, 134–138. [CrossRef] [PubMed]
4. Nüesch, E.; Dieppe, P.; Reichenbach, S.; Williams, S.; Iff, S.; Jüni, P. All cause and disease specific mortality in patients with knee
or hip osteoarthritis: Population based cohort study. BMJ 2011, 342, d1165. [CrossRef]
5. March, L.M.; Bachmeier, C.J. 10 Economics of osteoarthritis: A global perspective. Baillieres Clin. Rheumatol. 1997, 11, 817–834.
[CrossRef]
6. Hall, M.; van der Esch, M.; Hinman, R.S.; Peat, G.; de Zwart, A.; Quicke, J.G.; Runhaar, J.; Knoop, J.; van der Leeden, M.; de Rooij,
M.; et al. How does hip osteoarthritis differ from knee osteoarthritis? Osteoarthr. Cartil. 2021, 30, 32–41. [CrossRef]
7. Murphy, N.J.; Eyles, J.P.; Hunter, D.J. Hip Osteoarthritis: Etiopathogenesis and Implications for Management. Adv. Ther. 2016,
33, 1921–1946. [CrossRef]
8. Delpachitra, S.; Dimitroulis, G. Osteoarthritis of the temporomandibular joint: A review of aetiology and pathogenesis. Br. J. Oral
Maxillofac. Surg. 2022, 60, 387–396. [CrossRef]
9. Shorter, E.; Sannicandro, A.J.; Poulet, B.; Goljanek-Whysall, K. Skeletal Muscle Wasting and Its Relationship with Osteoarthritis:
A Mini-Review of Mechanisms and Current Interventions. Curr. Rheumatol. Rep. 2019, 21, 40. [CrossRef]
10. Adams, B.G.; Houston, M.N.P.; Cameron, K.L.P. The Epidemiology of Meniscus Injury. Sports Med. Arthrosc. Rev. 2021,
29, e24–e33. [CrossRef]
11. Wang, L.-J.; Zeng, N.; Yan, Z.-P.; Li, J.-T.; Ni, G.-X. Post-traumatic osteoarthritis following ACL injury. Arthritis Res. Ther. 2020,
22, 57. [CrossRef] [PubMed]
12. Fernandez-Madrid, F.; Karvonen, R.L.; Teitge, R.A.; Miller, P.R.; Negendank, W.G. MR features of osteoarthritis of the knee. Magn.
Reson. Imaging 1994, 12, 703–709. [CrossRef] [PubMed]
13. Zhu, X.; Chan, Y.T.; Yung, P.S.H.; Tuan, R.S.; Jiang, Y. Subchondral Bone Remodeling: A Therapeutic Target for Osteoarthritis.
Front. Cell Dev. Biol. 2021, 8, 607764. [CrossRef] [PubMed]
14. Taheri, S.; Winkler, T.; Schenk, L.S.; Neuerburg, C.; Baumbach, S.F.; Zustin, J.; Lehmann, W.; Schilling, A.F. Developmental
Transformation and Reduction of Connective Cavities within the Subchondral Bone. Int. J. Mol. Sci. 2019, 20, 770. [CrossRef]
[PubMed]
15. Taheri, S.; Yoshida, T.; Böker, K.; Foerster, R.; Jochim, L.; Flux, A.; Grosskopf, B.; Hawellek, T.; Lehmann, W.; Schilling, A. Changes
of the subchondral bone microchannel network in early osteoarthritis. Osteoarthr. Cartil. 2023, 31, 49–59. [CrossRef] [PubMed]
16. Chen, Y.; Hu, Y.; Yu, Y.E.; Zhang, X.; Watts, T.; Zhou, B.; Wang, J.; Wang, T.; Zhao, W.; Chiu, K.Y.; et al. Subchondral Trabecular
Rod Loss and Plate Thickening in the Development of Osteoarthritis. J. Bone Miner. Res. 2018, 33, 316–327. [CrossRef]
17. Dennison, E.M. Osteoarthritis: The importance of hormonal status in midlife women. Maturitas 2022, 165, 8–11. [CrossRef]
Diagnostics 2023, 13, 2586 16 of 19

18. Zoli, A.; Lizzio, M.M.; Capuano, A.; Massafra, U.; Barini, A.; Ferraccioli, G. Osteoporosis and bone metabolism in postmenopausal
women with osteoarthritis of the hand. Menopause 2006, 13, 462–466. [CrossRef]
19. Wang, A.; Zawadzki, N.; Hedlin, H.; LeBlanc, E.; Budrys, N.; Van Horn, L.; Gass, M.; Westphal, L.; Stefanick, M. Reproductive
history and osteoarthritis in the Women’s Health Initiative. Scand. J. Rheumatol. 2021, 50, 58–67. [CrossRef]
20. Jung, J.H.; Bang, C.H.; Song, G.G.; Kim, C.; Kim, J.-H.; Choi, S.J. Knee osteoarthritis and menopausal hormone therapy in
postmenopausal women: A nationwide cross-sectional study. Menopause 2019, 26, 598–602. [CrossRef]
21. Noehren, B.; Kosmac, K.; Walton, R.; Murach, K.; Lyles, M.; Loeser, R.; Peterson, C.; Messier, S. Alterations in quadriceps muscle
cellular and molecular properties in adults with moderate knee osteoarthritis. Osteoarthr. Cartil. 2018, 26, 1359–1368. [CrossRef]
22. Alnahdi, A.H.; Zeni, J.A.; Snyder-Mackler, L. Muscle Impairments in Patients with Knee Osteoarthritis. Sports Health Multidiscip.
Approach 2012, 4, 284–292. [CrossRef]
23. Di Pietro, G.; Scimeca, M.; Iundusi, R.; Celi, M.; Gasbarra, E.; Tarantino, U.; Capuani, S. Differences between muscle from
osteoporotic and osteoarthritic subjects: In vitro study by diffusion-tensor MRI and histological findings. Aging Clin. Exp. Res.
2020, 32, 2489–2499. [CrossRef]
24. Pereira, D.; Peleteiro, B.; Araújo, J.; Branco, J.; Santos, R.A.; Ramos, E. The effect of osteoarthritis definition on prevalence and
incidence estimates: A systematic review. Osteoarthr. Cartil. 2011, 19, 1270–1285. [CrossRef]
25. Guermazi, A.; Niu, J.; Hayashi, D.; Roemer, F.W.; Englund, M.; Neogi, T.; Aliabadi, P.; McLennan, C.E.; Felson, D.T. Prevalence of
abnormalities in knees detected by MRI in adults without knee osteoarthritis: Population based observational study (Framingham
Osteoarthritis Study). BMJ 2012, 345, e5339. [CrossRef]
26. Roemer, F.W.; Guermazi, A.; Demehri, S.; Wirth, W.; Kijowski, R. Imaging in Osteoarthritis. Osteoarthr. Cartil. 2022, 30, 913–934.
[CrossRef]
27. Bijlsma, J.W.; Berenbaum, F.; Lafeber, F.P. Osteoarthritis: An update with relevance for clinical practice. Lancet 2011, 377, 2115–2126.
[CrossRef]
28. Kellgren, J.H.; Lawrence, J.S. Radiological Assessment of Osteo-Arthrosis. Ann. Rheum. Dis. 1957, 16, 494–502. [CrossRef]
29. Crema, M.D.; Nevitt, M.; Guermazi, A.; Felson, D.; Wang, K.; Lynch, J.; Marra, M.D.; Torner, J.; Lewis, C.; Roemer, F. Progression
of cartilage damage and meniscal pathology over 30 months is associated with an increase in radiographic tibiofemoral joint
space narrowing in persons with knee OA—The MOST study. Osteoarthr. Cartil. 2014, 22, 1743–1747. [CrossRef]
30. Kijowski, R.; Blankenbaker, D.G.; Davis, K.W.; Shinki, K.; Kaplan, L.D.; De Smet, A.A. Comparison of 1.5- and 3.0-T MR Imaging
for Evaluating the Articular Cartilage of the Knee Joint. Radiology 2009, 250, 839–848. [CrossRef]
31. Masi, J.N.; Sell, C.A.; Phan, C.; Han, E.; Newitt, D.; Steinbach, L.; Majumdar, S.; Link, T.M. Cartilage MR Imaging at 3.0 versus
That at 1.5 T: Preliminary Results in a Porcine Model. Radiology 2005, 236, 140–150. [CrossRef] [PubMed]
32. Arnold, T.C.; Freeman, C.W.; Litt, B.; Stein, J.M. Low-field MRI: Clinical promise and challenges. J. Magn. Reson. Imaging 2023,
57, 25–44. [CrossRef] [PubMed]
33. Khodarahmi, I.M.; Keerthivasan, M.B.; Brinkmann, I.M.; Grodzki, D.; Fritz, J. Modern Low-Field MRI of the Musculoskeletal
System. Investig. Radiol. 2023, 58, 76–87. [CrossRef] [PubMed]
34. Sutter, R.; Tresch, F.; Buck, F.M.; Pfirrmann, C.W.A. Is Dedicated Extremity 1.5-T MRI Equivalent to Standard Large-Bore 1.5-T
MRI for Foot and Knee Examinations? Am. J. Roentgenol. 2014, 203, 1293–1302. [CrossRef]
35. Klein, H.-M. Low-Field Magnetic Resonance Imaging. RöFo-Fortschritte Geb. Röntgenstrahlen Bildgeb. Verfahr. 2020, 192, 537–548.
[CrossRef]
36. Lutterbey, G.; Behrends, K.; Falkenhausen, M.V.; Wattjes, M.P.; Morakkabati, N.; Gieseke, J.; Schild, H. Is the body-coil at 3 Tesla
feasible for the MRI evaluation of the painful knee? A comparative study. Eur. Radiol. 2007, 17, 503–508. [CrossRef]
37. Link, T.M. MR Imaging in Osteoarthritis: Hardware, Coils, and Sequences. Radiol. Clin. N. Am. 2009, 47, 617–632. [CrossRef]
38. Guglielmi, G.; Lennart, J.; Simoni, P.; Mascarenhas, V. Knee. MRI Protocols of the ESSR Arthritis Subcommittee. 2018. Available
online: https://www.essr.org/content-essr/uploads/2018/05/Knee.pdf (accessed on 22 February 2023).
39. Sudoł-Szopińska, I.; Jurik, A.G.; Eshed, I.; Lennart, J.; Grainger, A.; Østergaard, M.; Klauser, A.; Cotten, A.; Wick, M.C.; Maas, M.;
et al. Recommendations of the ESSR Arthritis Subcommittee for the Use of Magnetic Resonance Imaging in Musculoskeletal
Rheumatic Diseases. Semin. Musculoskelet. Radiol. 2015, 19, 396–411. [CrossRef]
40. Freeman, D.M.; Bergman, G.; Glover, G. Short TE MR microscopy: Accurate measurement and zonal differentiation of normal
hyaline cartilage. Magn. Reson. Med. 1997, 38, 72–81. [CrossRef]
41. Link, T.M.; Stahl, R.; Woertler, K. Cartilage imaging: Motivation, techniques, current and future significance. Eur. Radiol. 2007,
17, 1135–1146. [CrossRef]
42. Crema, M.D.; Roemer, F.W.; Marra, M.D.; Burstein, D.; Gold, G.E.; Eckstein, F.; Baum, T.; Mosher, T.J.; Carrino, J.A.; Guermazi, A.
Articular Cartilage in the Knee: Current MR Imaging Techniques and Applications in Clinical Practice and Research. RadioGraphics
2011, 31, 37–61. [CrossRef]
43. Delfaut, E.M.; Beltran, J.; Johnson, G.; Rousseau, J.; Marchandise, X.; Cotten, A. Fat Suppression in MR Imaging: Techniques and
Pitfalls. RadioGraphics 1999, 19, 373–382. [CrossRef]
44. Dixon, W.T. Simple proton spectroscopic imaging. Radiology 1984, 153, 189–194. [CrossRef]
45. Guerini, H.; Omoumi, P.; Guichoux, F.; Vuillemin, V.; Morvan, G.; Zins, M.; Thevenin, F.; Drape, J.L. Fat Suppression with Dixon
Techniques in Musculoskeletal Magnetic Resonance Imaging: A Pictorial Review. Semin. Musculoskelet. Radiol. 2015, 19, 335–347.
[CrossRef]
Diagnostics 2023, 13, 2586 17 of 19

46. Chen, Y.; Wang, T.; Guan, M.; Zhao, W.; Leung, F.-K.; Pan, H.; Cao, X.; Guo, X.; Lu, W. Bone turnover and articular cartilage
differences localized to subchondral cysts in knees with advanced osteoarthritis. Osteoarthr. Cartil. 2015, 23, 2174–2183. [CrossRef]
47. Muratovic, D.; Cicuttini, F.; Wluka, A.; Findlay, D.; Wang, Y.; Otto, S.; Taylor, D.; Humphries, J.; Lee, Y.; Labrinidis, A.; et al. Bone
marrow lesions detected by specific combination of MRI sequences are associated with severity of osteochondral degeneration.
Arthritis Res. Ther. 2016, 18, 54. [CrossRef]
48. Peterfy, C.; Guermazi, A.; Zaim, S.; Tirman, P.; Miaux, Y.; White, D.; Kothari, M.; Lu, Y.; Fye, K.; Zhao, S.; et al. Whole-Organ
Magnetic Resonance Imaging Score (WORMS) of the knee in osteoarthritis. Osteoarthr. Cartil. 2004, 12, 177–190. [CrossRef]
49. Deveza, L.; Melo, L.; Yamato, T.; Mills, K.; Ravi, V.; Hunter, D. Knee osteoarthritis phenotypes and their relevance for outcomes:
A systematic review. Osteoarthr. Cartil. 2017, 25, 1926–1941. [CrossRef] [PubMed]
50. Roemer, F.W.; Kwoh, C.K.; Hayashi, D.; Felson, D.T.; Guermazi, A. The role of radiography and MRI for eligibility assessment in
DMOAD trials of knee OA. Nat. Rev. Rheumatol. 2018, 14, 372–380. [CrossRef]
51. Ryd, L.; Flodström, K.; Manley, M. Patient-Specific Implants for Focal Cartilage Lesions in The Knee: Implant Survivorship
Analysis up to Seven Years Post-Implantation. Surg. Technol. Online 2020, 38, 379–386. [CrossRef]
52. Siepmann, D.B.; McGovern, J.; Brittain, J.H.; Reeder, S.B. High-Resolution 3D Cartilage Imaging with IDEAL–SPGR at 3 T. Am. J.
Roentgenol. 2007, 189, 1510–1515. [CrossRef] [PubMed]
53. Bruder, H.; Fischer, H.; Graumann, R.; Deimling, M. A new steady-state imaging sequence for simultaneous acquisition of two
MR images with clearly different contrasts. Magn. Reson. Med. 1988, 7, 35–42. [CrossRef] [PubMed]
54. Welsch, G.H.; Scheffler, K.; Mamisch, T.C.; Hughes, T.; Millington, S.; Deimling, M.; Trattnig, S. Rapid estimation of cartilage T2
based on double echo at steady state (DESS) with 3 Tesla. Magn. Reson. Med. 2009, 62, 544–549. [CrossRef]
55. Friedrich, K.M.; Reiter, G.; Kaiser, B.; Mayerhöfer, M.; Deimling, M.; Jellus, V.; Horger, W.; Trattnig, S.; Schweitzer, M.; Sa-
lomonowitz, E. High-resolution cartilage imaging of the knee at 3T: Basic evaluation of modern isotropic 3D MR-sequences. Eur.
J. Radiol. 2011, 78, 398–405. [CrossRef] [PubMed]
56. Fritz, J.; Raithel, E.; Thawait, G.K.; Gilson, W.; Papp, D.F. Six-Fold Acceleration of High-Spatial Resolution 3D SPACE MRI of
the Knee Through Incoherent k-Space Undersampling and Iterative Reconstruction—First Experience. Investig. Radiol. 2016,
51, 400–409. [CrossRef]
57. Van Dyck, P.; Smekens, C.; Roelant, E.; Vyvere, T.V.; Snoeckx, A.; De Smet, E. 3D CAIPIRINHA SPACE versus standard 2D TSE
for routine knee MRI: A large-scale interchangeability study. Eur. Radiol. 2022, 32, 6456–6467. [CrossRef]
58. Roemer, F.W.; Engelke, K.; Li, L.; Laredo, J.-D.; Guermazi, A. MRI underestimates presence and size of knee osteophytes using CT
as a reference standard. Osteoarthr. Cartil. 2023, 31, 656–668. [CrossRef]
59. Cheng, K.Y.; Moazamian, D.; Ma, Y.; Jang, H.; Jerban, S.; Du, J.; Chung, C.B. Clinical application of ultrashort echo time (UTE) and
zero echo time (ZTE) magnetic resonance (MR) imaging in the evaluation of osteoarthritis. Skelet. Radiol. 2023, 1–9. [CrossRef]
60. Aydıngöz, Ü.; Yıldız, A.E.; Ergen, F.B. Zero Echo Time Musculoskeletal MRI: Technique, Optimization, Applications, and Pitfalls.
Radiographics 2022, 42, 1398–1414. [CrossRef]
61. Du, J.; Carl, M.; Bae, W.; Statum, S.; Chang, E.; Bydder, G.; Chung, C. Dual inversion recovery ultrashort echo time (DIR-UTE)
imaging and quantification of the zone of calcified cartilage (ZCC). Osteoarthr. Cartil. 2012, 21, 77–85. [CrossRef]
62. Chu, C.R.; Williams, A.A.; West, R.V.; Qian, Y.; Fu, F.H.; Do, B.H.; Bruno, S. Quantitative Magnetic Resonance Imaging UTE-T2*
Mapping of Cartilage and Meniscus Healing After Anatomic Anterior Cruciate Ligament Reconstruction. Am. J. Sports Med. 2014,
42, 1847–1856. [CrossRef] [PubMed]
63. Garetier, M.; Rousset, J.; Makki, K.; Brochard, S.; Rousseau, F.; Ben Salem, D.; Borotikar, B. Assessment and comparison of image
quality between two real-time sequences for dynamic MRI of distal joints at 3.0 Tesla. Acta Radiol. 2023, 64, 1093–1102. [CrossRef]
[PubMed]
64. Garetier, M.; Borotikar, B.; Makki, K.; Brochard, S.; Rousseau, F.; Ben Salem, D. Dynamic MRI for articulating joint evaluation on
1.5 T and 3.0 T scanners: Setup, protocols, and real-time sequences. Insights Imaging 2020, 11, 66. [CrossRef] [PubMed]
65. Bowes, M.A.; Kacena, K.; Alabas, O.A.; Brett, A.D.; Dube, B.; Bodick, N.; Conaghan, P.G. Machine-learning, MRI bone shape
and important clinical outcomes in osteoarthritis: Data from the Osteoarthritis Initiative. Ann. Rheum. Dis. 2021, 80, 502–508.
[CrossRef]
66. Kijowski, R.; Fritz, J.; Deniz, C.M. Deep learning applications in osteoarthritis imaging. Skelet. Radiol. 2023, 1–14. [CrossRef]
67. Desai, A.D.; Caliva, F.; Iriondo, C.; Mortazi, A.; Jambawalikar, S.; Bagci, U.; Perslev, M.; Igel, C.; Dam, E.B.; Gaj, S.; et al. The
International Workshop on Osteoarthritis Imaging Knee MRI Segmentation Challenge: A Multi-Institute Evaluation and Analysis
Framework on a Standardized Dataset. Radiol. Artif. Intell. 2021, 3, e200078. [CrossRef]
68. Aigner, T.; McKenna, L. Molecular pathology and pathobiology of osteoarthritic cartilage. Cell. Mol. Life Sci. 2002, 59, 5–18.
[CrossRef]
69. Hohe, J.; Faber, S.; Stammberger, T.; Reiser, M.; Englmeier, K.-H.; Eckstein, F. A technique for 3D in vivo quantification of proton
density and magnetization transfer coefficients of knee joint cartilage. Osteoarthr. Cartil. 2000, 8, 426–433. [CrossRef]
70. Lüssea, S.; Claassen, H.; Gehrke, T.; Hassenpflug, J.; Schünke, M.; Heller, M.; Glüer, C.-C. Evaluation of water content by spatially
resolved transverse relaxation times of human articular cartilage. Magn. Reson. Imaging 2000, 18, 423–430. [CrossRef]
71. Dunn, T.C.; Lu, Y.; Jin, H.; Ries, M.D.; Majumdar, S.; Roemer, F.W.; Demehri, S.; Omoumi, P.; Link, T.M.; Kijowski, R.; et al. T2
Relaxation Time of Cartilage at MR Imaging: Comparison with Severity of Knee Osteoarthritis. Radiology 2004, 232, 592–598.
[CrossRef]
Diagnostics 2023, 13, 2586 18 of 19

72. Koff, M.; Amrami, K.; Kaufman, K. Clinical evaluation of T2 values of patellar cartilage in patients with osteoarthritis. Osteoarthr.
Cartil. 2007, 15, 198–204. [CrossRef]
73. Wang, L.; Regatte, R.R. T1ρ MRI of human musculoskeletal system. J. Magn. Reson. Imaging 2015, 41, 586–600. [CrossRef]
74. Li, X.; Cheng, J.; Lin, K.; Saadat, E.; Bolbos, R.I.; Jobke, B.; Ries, M.D.; Horvai, A.; Link, T.M.; Majumdar, S. Quantitative MRI
using T1ρ and T2 in human osteoarthritic cartilage specimens: Correlation with biochemical measurements and histology. Magn.
Reson. Imaging 2011, 29, 324–334. [CrossRef]
75. Sigurdsson, U.; Müller, G.; Siversson, C.; Lammentausta, E.; Svensson, J.; Tiderius, C.-J.; Dahlberg, L.E. Delayed gadolinium-
enhanced MRI of meniscus (dGEMRIM) and cartilage (dGEMRIC) in healthy knees and in knees with different stages of meniscus
pathology. BMC Musculoskelet. Disord. 2016, 17, 406. [CrossRef]
76. van Tiel, J.; Kotek, G.; Reijman, M.; Bos, P.K.; Bron, E.E.; Klein, S.; Verhaar, J.A.N.; Krestin, G.P.; Weinans, H.; Oei, E.H.G. Delayed
gadolinium-enhanced MRI of the meniscus (dGEMRIM) in patients with knee osteoarthritis: Relation with meniscal degeneration
on conventional MRI, reproducibility, and correlation with dGEMRIC. Eur. Radiol. 2014, 24, 2261–2270. [CrossRef]
77. Hangaard, S.; Gudbergsen, H.; Daugaard, C.L.; Bliddal, H.; Nybing, J.D.; Nieminen, M.T.; Casula, V.; Tiderius, C.-J.; Boesen, M.
Delayed gadolinium-enhanced MRI of menisci and cartilage (dGEMRIM/dGEMRIC) in obese patients with knee osteoarthritis:
Cross-sectional study of 85 obese patients with intra-articular administered gadolinium contrast. J. Magn. Reson. Imaging 2018,
48, 1700–1706. [CrossRef]
78. Guermazi, A.; Alizai, H.; Crema, M.D.; Trattnig, S.; Regatte, R.; Roemer, F. Compositional MRI techniques for evaluation of
cartilage degeneration in osteoarthritis. Osteoarthr. Cartil. 2015, 23, 1639–1653. [CrossRef]
79. Miller, K.L.; Hargreaves, B.A.; Gold, G.E.; Pauly, J.M. Steady-state diffusion-weighted imaging of in vivo knee cartilage. Magn.
Reson. Med. 2004, 51, 394–398. [CrossRef]
80. Mlynárik, V.; Sulzbacher, I.; Bittšanský, M.; Fuiko, R.; Trattnig, S. Investigation of apparent diffusion constant as an indicator of
early degenerative disease in articular cartilage. J. Magn. Reson. Imaging 2003, 17, 440–444. [CrossRef]
81. Quirbach, S.; Trattnig, S.; Marlovits, S.; Zimmermann, V.; Domayer, S.; Dorotka, R.; Mamisch, T.C.; Bohndorf, K.; Welsch, G.H.
Initial results of in vivo high-resolution morphological and biochemical cartilage imaging of patients after matrix-associated
autologous chondrocyte transplantation (MACT) of the ankle. Skelet. Radiol. 2009, 38, 751–760. [CrossRef]
82. Kogan, F.; Hariharan, H.; Reddy, R. Chemical Exchange Saturation Transfer (CEST) Imaging: Description of Technique and
Potential Clinical Applications. Curr. Radiol. Rep. 2013, 1, 102–114. [CrossRef] [PubMed]
83. Soellner, S.; Welsch, G.; Gelse, K.; Goldmann, A.; Kleyer, A.; Schett, G.; Pachowsky, M. gagCEST imaging at 3 T MRI in patients
with articular cartilage lesions of the knee and intraoperative validation. Osteoarthr. Cartil. 2021, 29, 1163–1172. [CrossRef]
[PubMed]
84. Lombardi, A.F.; Ma, Y.; Jang, H.; Jerban, S.; Tang, Q.; Searleman, A.C.; Meyer, R.S.; Du, J.; Chang, E.Y. AcidoCEST-UTE MRI
Reveals an Acidic Microenvironment in Knee Osteoarthritis. Int. J. Mol. Sci. 2022, 23, 4466. [CrossRef] [PubMed]
85. Madelin, G.; Lee, J.-S.; Regatte, R.R.; Jerschow, A. Sodium MRI: Methods and applications. Prog. Nucl. Magn. Reson. Spectrosc.
2014, 79, 14–47. [CrossRef]
86. Kamp, B.; Frenken, M.; Henke, J.M.; Abrar, D.B.; Nagel, A.M.; Gast, L.V.; Oeltzschner, G.; Wilms, L.M.; Nebelung, S.; Antoch,
G.; et al. Quantification of Sodium Relaxation Times and Concentrations as Surrogates of Proteoglycan Content of Patellar
CARTILAGE at 3T MRI. Diagnostics 2021, 11, 2301. [CrossRef]
87. Outerbridge, R.E. The etiology of chondromalacia patellae. J. Bone Jt. Surg. 1961, 43, 752–757. [CrossRef]
88. Jungius, K.-P.; Schmid, M.R.; Zanetti, M.; Hodler, J.; Koch, P.; Pfirrmann, C.W.A. Cartilaginous Defects of the Femorotibial Joint:
Accuracy of Coronal Short Inversion Time Inversion-Recovery MR Sequence. Radiology 2006, 240, 482–488. [CrossRef]
89. Schreiner, M.M.; Raudner, M.; Marlovits, S.; Bohndorf, K.; Weber, M.; Zalaudek, M.; Röhrich, S.; Szomolanyi, P.; Filardo, G.;
Windhager, R.; et al. The MOCART (Magnetic Resonance Observation of Cartilage Repair Tissue) 2.0 Knee Score and Atlas.
Cartilage 2021, 13, 571S–587S. [CrossRef]
90. Kornaat, P.R.; Ceulemans, R.Y.T.; Kroon, H.M.; Riyazi, N.; Kloppenburg, M.; Carter, W.O.; Woodworth, T.G.; Bloem, J.L. MRI
assessment of knee osteoarthritis: Knee Osteoarthritis Scoring System (KOSS)—Inter-observer and intra-observer reproducibility
of a compartment-based scoring system. Skelet. Radiol. 2005, 34, 95–102. [CrossRef]
91. Hunter, D.; Conaghan, P.; Peterfy, C.; Bloch, D.; Guermazi, A.; Woodworth, T.; Stevens, R.; Genant, H. Responsiveness, effect size,
and smallest detectable difference of Magnetic Resonance Imaging in knee osteoarthritis. Osteoarthr. Cartil. 2006, 14, 112–115.
[CrossRef]
92. Hunter, D.J.; Lo, G.H.; Gale, D.; Grainger, A.J.; Guermazi, A.; Conaghan, P.G. The reliability of a new scoring system for knee
osteoarthritis MRI and the validity of bone marrow lesion assessment: BLOKS (Boston–Leeds Osteoarthritis Knee Score). Ann.
Rheum. Dis. 2008, 67, 206–211. [CrossRef]
93. Felson, D.; Lynch, J.; Guermazi, A.; Roemer, F.; Niu, J.; McAlindon, T.; Nevitt, M. Comparison of BLOKS and WORMS scoring
systems part II. Longitudinal assessment of knee MRIs for osteoarthritis and suggested approach based on their performance:
Data from the Osteoarthritis Initiative. Osteoarthr. Cartil. 2010, 18, 1402–1407. [CrossRef]
94. Hunter, D.; Guermazi, A.; Lo, G.; Grainger, A.; Conaghan, P.; Boudreau, R.; Roemer, F. Evolution of semi-quantitative whole joint
assessment of knee OA: MOAKS (MRI Osteoarthritis Knee Score). Osteoarthr. Cartil. 2011, 19, 990–1002. [CrossRef]
Diagnostics 2023, 13, 2586 19 of 19

95. Roemer, F.W.; Guermazi, A.; Zhang, Y.; Yang, M.; Hunter, D.J.; Crema, M.D.; Bohndorf, K. Hoffa’s Fat Pad: Evaluation on
Unenhanced MR Images as a Measure of Patellofemoral Synovitis in Osteoarthritis. Am. J. Roentgenol. 2009, 192, 1696–1700.
[CrossRef]
96. Guermazi, A.; Roemer, F.W.; Hayashi, D.; Crema, M.D.; Niu, J.; Zhang, Y.; Marra, M.D.; Katur, A.; Lynch, J.A.; El-Khoury, G.Y.;
et al. Assessment of synovitis with contrast-enhanced MRI using a whole-joint semiquantitative scoring system in people with, or
at high risk of, knee osteoarthritis: The MOST study. Ann. Rheum. Dis. 2011, 70, 805–811. [CrossRef]
97. Roemer, F.; Collins, J.; Kwoh, C.; Hannon, M.; Neogi, T.; Felson, D.; Hunter, D.; Lynch, J.; Guermazi, A. MRI-based screening for
structural definition of eligibility in clinical DMOAD trials: Rapid OsteoArthritis MRI Eligibility Score (ROAMES). Osteoarthr.
Cartil. 2020, 28, 71–81. [CrossRef]
98. Wallace, I.J.; Worthington, S.; Felson, D.T.; Jurmain, R.D.; Wren, K.T.; Maijanen, H.; Woods, R.J.; Lieberman, D.E. Knee osteoarthri-
tis has doubled in prevalence since the mid-20th century. Proc. Natl. Acad. Sci. USA 2017, 114, 9332–9336. [CrossRef]
99. Hutchinson, M.; Raff, U. Fast MRI data acquisition using multiple detectors. Magn. Reson. Med. 1988, 6, 87–91. [CrossRef]
100. Glockner, J.F.; Hu, H.H.; Stanley, D.W.; Angelos, L.; King, K. Parallel MR Imaging: A User’s Guide. RadioGraphics 2005,
25, 1279–1297. [CrossRef]
101. Barth, M.; Breuer, F.; Koopmans, P.J.; Norris, D.G.; Poser, B.A. Simultaneous multislice (SMS) imaging techniques. Magn. Reson.
Med. 2016, 75, 63–81. [CrossRef]
102. Geethanath, S.; Reddy, R.; Konar, A.S.; Imam, S.; Sundaresan, R.; Ramesh Babu, D.R.; Venkatesan, R. Compressed Sensing MRI: A
Review. Crit. Rev. Biomed. Eng. 2013, 41, 183–204. [CrossRef] [PubMed]
103. Akai, H.; Yasaka, K.; Sugawara, H.; Tajima, T.; Kamitani, M.; Furuta, T.; Akahane, M.; Yoshioka, N.; Ohtomo, K.; Abe, O.; et al.
Acceleration of knee magnetic resonance imaging using a combination of compressed sensing and commercially available deep
learning reconstruction: A preliminary study. BMC Med. Imaging 2023, 23, 5. [CrossRef] [PubMed]
104. Müller-Franzes, G.; Nolte, T.; Ciba, M.; Schock, J.; Khader, F.; Prescher, A.; Wilms, L.M.; Kuhl, C.; Nebelung, S.; Truhn, D. Fast,
Accurate, and Robust T2 Mapping of Articular Cartilage by Neural Networks. Diagnostics 2022, 12, 688. [CrossRef] [PubMed]
105. Knoll, F.; Hammernik, K.; Zhang, C.; Moeller, S.; Pock, T.; Sodickson, D.K.; Akcakaya, M. Deep-Learning Methods for Parallel
Magnetic Resonance Imaging Reconstruction: A Survey of the Current Approaches, Trends, and Issues. IEEE Signal Process. Mag.
2020, 37, 128–140. [CrossRef] [PubMed]
106. Hammernik, K.; Klatzer, T.; Kobler, E.; Recht, M.P.; Sodickson, D.K.; Pock, T.; Knoll, F. Learning a variational network for
reconstruction of accelerated MRI data. Magn. Reson. Med. 2018, 79, 3055–3071. [CrossRef]
107. Roemer, F.W. The Role of Speed and Possible Implications. Radiology 2023, 307, e222872. [CrossRef]
108. Johnson, P.M.; Lin, D.J.; Zbontar, J.; Zitnick, C.L.; Sriram, A.; Muckley, M.; Babb, J.S.; Kline, M.; Ciavarra, G.; Alaia, E.; et al. Deep
Learning Reconstruction Enables Prospectively Accelerated Clinical Knee MRI. Radiology 2023, 307, e220425. [CrossRef]
109. Kulseng, C.P.S.; Nainamalai, V.; Grøvik, E.; Geitung, J.-T.; Årøen, A.; Gjesdal, K.-I. Automatic segmentation of human knee
anatomy by a convolutional neural network applying a 3D MRI protocol. BMC Musculoskelet. Disord. 2023, 24, 41. [CrossRef]
110. Almajalid, R.; Zhang, M.; Shan, J. Fully Automatic Knee Bone Detection and Segmentation on Three-Dimensional MRI. Diagnostics
2022, 12, 123. [CrossRef]
111. Yeoh, P.S.Q.; Lai, K.W.; Goh, S.L.; Hasikin, K.; Hum, Y.C.; Tee, Y.K.; Dhanalakshmi, S. Emergence of Deep Learning in Knee
Osteoarthritis Diagnosis. Comput. Intell. Neurosci. 2021, 2021, 4931437. [CrossRef]
112. Siouras, A.; Moustakidis, S.; Giannakidis, A.; Chalatsis, G.; Liampas, I.; Vlychou, M.; Hantes, M.; Tasoulis, S.; Tsaopoulos, D. Knee
Injury Detection Using Deep Learning on MRI Studies: A Systematic Review. Diagnostics 2022, 12, 537. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.