MRI Advancements in Musculoskeletal Clinical and Research Practice

Abstract

Over the past decades, MRI has become increasingly important for diagnosing and longitudinally monitoring musculoskeletal disorders, with ongoing hardware and software improvements aiming to optimize image quality and speed. However, surging demand for musculoskeletal MRI and increased interest to provide more personalized care will necessitate a stronger emphasis on efficiency and specificity. Ongoing hardware developments include more powerful gradients, improvements in wide-bore magnet designs to maintain field homogeneity, and high-channel phased-array coils. There is also interest in low-field-strength magnets with inherently lower magnetic footprints and operational costs to accommodate global demand in middle- and low-income countries. Previous approaches to decrease acquisition times by means of conventional acceleration techniques (eg, parallel imaging or compressed sensing) are now largely overshadowed by deep learning reconstruction algorithms. It is expected that greater emphasis will be placed on improving synthetic MRI and MR fingerprinting approaches to shorten overall acquisition times while also addressing the demand of personalized care by simultaneously capturing microstructural information to provide greater detail of disease severity. Authors also anticipate increased research emphasis on metal artifact reduction techniques, bone imaging, and MR neurography to meet clinical needs.

© RSNA, 2023

See also the review “Musculoskeletal CT Imaging: State-of-the-Art Advancements and Future Directions” by Demehri et al in this issue.

Summary

The anticipated expansion of musculoskeletal MRI over the next decade will necessitate more tailored protocols, quantitative techniques to capture microstructural information, and artificial intelligence algorithms for image acceleration, reconstruction, and interpretation.

Essentials

• ■ Technical developments continue to strengthen the role of MRI in the musculoskeletal field, but the rising number of annual examinations and expectations of improved diagnostic specificity will necessitate cultivating techniques to manage increased volume and improve image quality.

• ■ Continued advancements during the past decade in MRI hardware feature wide-bore magnets, more powerful gradients, and conformable phased-array coils, along with trends toward both low-field-strength systems to reduce the magnetic field footprint and operational costs and high-field-strength systems to improve spatial resolution.

• ■ Three-dimensional isotropic data sets amenable to multiplanar reformations are now increasingly generated in clinical musculoskeletal MRI practice, largely driven by advances in acceleration and deep learning reconstruction techniques.

• ■ Short-echo-time sequences enable qualitative and quantitative analyses of rapidly decaying signal from tissues such as menisci, cortical bone, tendons, and ligaments.

• ■ MR neurography, a rapidly growing field among musculoskeletal radiologists and neuroradiologists, uses vascular suppression techniques previously developed for vessel wall imaging.

Introduction

MRI is already indispensable for the diagnostic work-up of many musculoskeletal disorders, and its use in this field is anticipated to increase (1). Modern MRI techniques afford multiplanar and high-spatial-resolution capabilities to depict soft-tissue and osseous pathology without using ionizing radiation. Given its wide dynamic contrast range, MRI provides not only qualitative assessment of macrostructural tissue integrity but also quantitative information about microstructural architecture and biochemical composition.

These capabilities enable characterization of a wide range of pathologies, including acute traumatic injury, degeneration, inflammation, metabolic derangements, and neoplasm. MRI can be used to detect fractures sometimes occult on radiographs and CT images and evaluate tendon, ligament, and cartilage damage following trauma to streamline patient management. Given its high sensitivity to injury, MRI can inform return-to-play times following sports injuries and be used to distinguish acute from chronic injury (Fig 1) (2). Shorter convalescence times are also facilitated by early detection of repetitive microtrauma that may induce stress fractures (3). Stress fractures particularly at higher risk of nonunion (eg, those involving the femoral neck, tibial shaft, or tarsal navicular) benefit from rapid intervention (non–weight-bearing or surgical fixation) (4). Moreover, early assessment of osteoarthritis is primarily achieved with MRI, which enables quantitative evaluation of chondral microstructure for diagnosis, surveillance, and treatment (5). In addition to individual joint assessment, MRI enables monitoring of systemic diseases, including rheumatologic, metabolic, and oncologic conditions. Technical developments have enabled more routine deployment of large-field-of-view or whole-body MRI to assess disease processes such as spondylarthritis, myositis, neurofibromatosis, and multiple myeloma (6–8). More focused MR neurography examinations can help precisely localize the site of injury in traumatic and spontaneous peripheral neuropathies (9).

Figure 1:

Acute-on-chronic low-grade pectoralis major strain in a 42-year-old male patient with right chest pain after bench-pressing 1 week before imaging. (A) Oblique coronal short-tau inversion-recovery (STIR) and (B) T1-weighted MRI scans demonstrate a small amount of fatty deposition (arrowheads) along the pectoralis major sternal head myotendinous junction (arrow) compatible with an old tear. (C) Immediately posterior oblique coronal short-tau inversion-recovery image demonstrates evidence of more recent, low-grade myotendinous junction (arrowheads) and intramuscular (arrow) tear.

The role of MRI in musculoskeletal medicine is still advancing and expanding, and additional demands will need to be addressed. The simultaneous increase in the number of joint replacements in the aging population and greater global availability of MRI scanners has led to a surge in musculoskeletal MRI examinations. Such volume necessitates a stronger emphasis on efficiency and specificity while remaining cost-effective. Strategies to address this challenge will likely involve even faster scanning times, use of low-field-strength magnets (when appropriate) with inherently lower operational costs, and greater attention to individualized care with more tailored protocols and broader adoption of objective, quantitative techniques.

This review focuses on advances in musculoskeletal MRI, including continuous modifications in magnet, gradient, and radiofrequency coil designs as well developments in software acquisition and reconstruction schemes that include synthetic MRI, MR fingerprinting, and deep learning (DL) algorithms. Qualitative and quantitative imaging techniques for cartilage and peripheral nerve and for successfully imaging around orthopedic hardware are also highlighted. Finally, developments in short-echo-time acquisitions to evaluate rapidly decaying T2 species abundant in ligaments, tendons, menisci, and cortical bone that are otherwise invisible with conventional pulse sequences are discussed.

Hardware: Magnets, Gradients, and Radiofrequency Coils

Magnets and Gradients

Wide-bore (70-cm diameter) 1.5- and 3.0-T MRI systems have become mainstream for clinical musculoskeletal MRI, with major vendors offering a selection of these units with varying gradient performance. Although B₀ and B₁ inhomogeneities may be magnified, a wide-bore system facilitates positioning of the target anatomy closer to isocenter, where the magnetic field (important for water excitation or chemical-based fat suppression) and gradient fields (important for resolution uniformity) are more homogeneous. Although earlier wide-bore MRI systems had poorer maximum gradient amplitude (approximately 33 mT/m) and slew rate (approximately 120 mT/m/msec) compared with 60-cm-diameter systems (approximately 50 mT/m amplitude and 200 mT/m/msec slew rate), wide-bore systems introduced between 2017 and 2021 provide higher capacity and peak power gradient amplifiers to enable higher gradient performance (60–80 mT/m, 150–220 mT/m/msec). Stronger gradient amplifiers, however, sometimes require infrastructure upgrades to support power and cooling requirements. Improved gradient performance is critical for minimizing echo spacing (to reduce blurring), particularly for the “workhorse” fast/turbo spin-echo (FSE/TSE) sequences used in musculoskeletal MRI.

While comparisons between 1.5- and 3.0-T musculoskeletal MRI demonstrate similar diagnostic performance (10), the higher signal-to-noise ratio (SNR) efficiency of 3.0-T systems (nearly two times higher than that of 1.5-T systems) has been exploited to shorten scanning time and/or improve spatial resolution (11). This is especially true when dedicated transmit/receive radiofrequency coils are used for knee and wrist MRI, allowing for reduced repetition times or the acquisition of more sections to increase scanning efficiency without facing specific absorption rate limitations. Compared with MRI at 3.0 T, clinical musculoskeletal MRI at 7.0 T, which is used mostly in the knee, can provide higher SNR and may enable novel quantitative MRI applications, such as nonproton imaging (12). Given the high prevalence of orthopedic hardware in the general population, lower field strengths are preferred when imaging near metallic hardware due to reduced susceptibility effects. One possible exception to this is nerve imaging, where spatial resolution is critical (13).

For the lower field strength range, the 0.55-T whole-body system with an 80-cm bore introduced by Siemens Healthineers in 2021 provides almost threefold less susceptibility than 1.5-T systems and may offer even greater benefit for imaging around implants (14). Moreover, its wide bore may further improve patient comfort and isocenter positioning (15). Additional advantages of this system include a smaller magnetic field footprint and less helium required for cooling. Resurgent interest in the lower field strength range may lead to an expansion in the utilization of existing lower-field-strength systems, such as the FONAR upright 0.6-T system and extremity systems (eg, GE 1.5-T Optima MR430s and Esaote 0.25–0.31-T O-scan) for musculoskeletal applications.

Radiofrequency Coils

As musculoskeletal MRI encompasses a broad range of anatomic regions, different phased-array coils for each body part are often required to optimize spatial resolution. The number of coil elements of commercially available phased-array coils now ranges from eight to 72 (Table 1), an increase from the four to 16 coil elements available in prior decades. Furthermore, multiple coil arrays may be combined, and maximum channels available for radiofrequency reception have also increased, from 32 or less before this decade to over 200 for recent commercial systems. While high-element radiofrequency coils add technical complexity and signal nonuniformity, they also offer key advantages: For a fixed surface area, a higher density of radiofrequency coil elements provides (a) higher SNR for superficial anatomy (16) and (b) enhanced parallel imaging (PI) capabilities due to increased localized coil sensitivities. The latter is particularly important in reducing otherwise long scanning times for three-dimensional (3D) imaging, including multispectral imaging techniques for imaging around metal. Additionally, recent surface coil technologies (17), high-impedance coils (18), screen-printed coils (19), and liquid metal conductor coils (20) provide further conformity to the targeted anatomy for higher SNR that may enable future applications such as dynamic joint imaging (18). A selection of commonly used rigid and flexible multipurpose coils for musculoskeletal MRI, in addition to prototype designs, are illustrated in Figure 2.

Table 1:

Survey of Highest–Channel Count Commercially Available Radiofrequency Coils for Musculoskeletal MRI from Three Major MRI Vendors (GE Healthcare, Philips, and Siemens)

Figure 2:

The top row shows commercially available coils. (A) A rigid 18-channel transmit/receive coil is primarily used for knee imaging, whereas (B) a 24-channel flexible “multipurpose” receive-only coil can be used for the knee, (C) brachial plexus or shoulder, and (D) elbow regions, among others. The bottom row shows prototype receive-only coils. (E) A screen-printed 12-channel array has been developed for pediatric applications; note flexibility by the anterior component being folded in the longitudinal direction (arrows). (Adapted, with permission, from reference 19.) (F) A 23-channel array designed for cervical imaging comprises high-impedance coil elements embedded in a flexible mask and a rigid posterior head-rest housing. (Courtesy of Tesla Dynamic Coils.) (G, H) A dual-channel array using a liquid metal conductor and self-tuning stretchable capacitor (arrows) demonstrates form-fitting conformability for different degrees of knee flexion. (Adapted, with permission, from reference 101.)

Acquisition and Reconstruction Schemes for Accelerating Musculoskeletal MRI

Rapid image acquisition for optimal efficiency and motion reduction is highly valued in all MRI disciplines. Traditional acceleration approaches have matured over the past 15 years for two-dimensional (2D) and 3D acquisitions and include PI, compressed sensing, and simultaneous multislice (SMS). These techniques are now largely integrated into 1.5- or 3.0-T musculoskeletal MRI protocols, and it is common to apply at least a twofold acceleration in daily practice. Newer advanced acceleration techniques include synthetic MRI techniques and MR fingerprinting, as well as DL-based reconstruction algorithms, all of which hold enormous research and clinical potential to further accelerate imaging.

“Conventional” Acceleration and 3D Techniques

Image-based and k-space–based PI, compressed sensing, and SMS acquisition (Table 2) approaches are now available for most pulse sequences and can be combined to achieve net four- to eightfold acceleration factors, yielding 80% faster examinations; for example, high-quality joint MRI can now be performed in 5 minutes or less (21). While acceleration schemes are more commonly used at 3.0 T compared with 1.5 T due to higher available SNR, these techniques are generally applicable at any field strength, including on modern 0.55-T low-field-strength scanners (22).

Table 2:

Acronyms and Definitions of MRI Acceleration Techniques

Conventional acceleration techniques have enabled clinical implementation of 3D musculoskeletal MRI acquisitions. Three-dimensional pulse sequences are often accelerated bidirectionally, substantially shortening the formerly long scanning times of isotropic resolution 3D gradient-echo and FSE/TSE sequences by fourfold if twofold acceleration is performed in both phase-encoding directions (23). The commercially available blipped controlled aliasing in PI results in higher acceleration, or CAIPIRINHA, technique has also been shown to improve image reconstruction, noise levels, and aliasing artifacts (24). When this technique was used in 10-minute 3D TSE knee MRI in pediatric patients, it provided surgically confirmed accuracies of 84%–100% for diagnosing internal derangement (25). In addition, meta-analyses have shown comparable diagnostic performances of 2D and 3D FSE/TSE MRI for anterior cruciate ligament and meniscal tears (26,27). In the spine, isotropic 3D MRI (aided by PI and DL reconstruction) facilitates high-spatial-resolution reformations in multiple planes for equivalent image evaluation in the lumbar spine (28) and superior evaluation of foraminal stenosis in the cervical spine (29) compared with 2D sequences. Moreover, high-quality 3D MRI data sets are suitable for advanced 3D visualization, such as cinematic rendering to support physician-to-physician and physician-to-patient communication (Fig 3). In patients with metallic implants, conventional PI and compressed sensing are useful to reduce scanning times of 3D metal artifact reduction sequences or multispectral imaging techniques, such as slice encoding for metal artifact correction and multiacquisition variable-resonance image combination (21). Compressed sensing is readily compatible with time-intensive slice encoding for metal artifact correction acquisitions (Fig 4) (30) and affords shorter acquisition times of up to 70% in clinical practice.

Figure 3:

Acute osteomyelitis of the distal tibia in a 5-year-old boy experiencing lower leg pain. (A) Cinematic three-dimensional (3D) MRI surface rendering shows a markedly swollen (arrow) ankle and foot. (B) Sagittal contrast-enhanced 3D T1-weighted fat-suppressed volumetric interpolated breath-hold examination (T1FS VIBE) and (C) corresponding cinematic rendering (CR) image demonstrate an intraosseous Brodie-type abscess (black arrow, B) decompressing into a subperiosteal abscess (white arrows). (D) The corresponding cinematic rendering vascular map shows marked hyperemia of the distal tibia (arrow).

Figure 4:

Osteochondral defect in a 42-year-old male patient with left knee pain after osteosynthesis of the proximal tibia. Compared with the conventional, high–receiver bandwidth (HBW) (A) proton density (PD) and (C) short-tau inversion-recovery (STIR) images, (B) compressed sensing (CS) slice encoding for metal artifact correction (SEMAC) proton density and (D) short-tau inversion-recovery turbo spin-echo (TSE) images correct the image distortion, unmasking a focally subsided osteochondral defect (arrows). The time stamps indicate the acquisition times of each pulse sequence.

SMS acquisitions increase scanning efficiency in 2D imaging by sampling two- or three-stack sections simultaneously while using PI for section deconvolution, with only minimal SNR loss (31). Combined SMS acquisition and in-plane PI enable fourfold acceleration, yielding 5-minute, 1.5-, and 3.0-T knee MRI protocols with equivalent diagnostic performance compared with 10-minute standard-of-care protocols (32). However, the addition of SMS acquisition for FSE/TSE imaging of larger, deeper-seated joints at 3.0 T has been limited primarily by the higher specific absorption rate, which reduces the section efficiency of SMS.

Synthetic MRI and MR Fingerprinting

Major barriers to incorporate quantitative MRI in current clinical practice include increased scanning time, relatively poor spatial resolution relative to qualitative scans, and the need to improve robustness and accuracy of image registration methods. Synthetic MRI methods that generate multicontrast images, including quantitative maps, from a single acquisition may be one promising approach to overcome these barriers. One example is the GRAPPATINI method (33), which facilitates rapid acquisition of T2 maps combining acceleration with PI and by means of multiecho undersampling of k-space to generate synthetic T2-weighted images that can be interpreted qualitatively and quantitatively (Fig 5). These synthetic images have been shown to be comparable with more conventional qualitative-only TSE sequences performed in the spine (34) and in the knee (35). Another more complex approach is QRAPMASTER (36), which can generate T1, proton density, and T2 parametric maps and qualitative images simultaneously from a single data set (37). Such synthetic MRI techniques can generate arbitrary tissue contrast by modifying inversion, repetition, and echo times and have shown a high level of diagnostic agreement with conventional sequences in the knee, spine, and shoulder regions (21).

Figure 5:

Osteoarthritis in a 49-year-old male patient with chronic left knee pain. (A) Routine two-dimensional sagittal proton density (Sag PD) MRI scan demonstrates chondral high signal intensity (blue arrows), and (B) T2 maps obtained from a sixfold-accelerated GRAPPATINI acquisition (10 echoes, twofold parallel imaging, threefold acceleration in echo time) demonstrate mild surface chondral wear over posterior margins of the lateral femoral condyle with T2 prolongation (black arrows). GRAPPATINI is a combination of a k-based method for undersampling called generalized autocalibrating partial parallel acquisition, or GRAPPA, and the reconstruction technique model-based accelerated relaxometry by iterative nonlinear inversion, or MARTINI (33).

An alternative to synthetic MRI, MR fingerprinting (38) uses pulse sequences with varying settings and parameters (eg, flip angle, sampling trajectory) throughout the acquisition in a pseudorandom fashion to generate “fingerprints” (ie, a physics-based dictionary) that output either quantitative maps or synthetic images on a pixel-by-pixel basis, including B₀, T2/T2*, and T1 values (39). This technique has shown potential in mitigating field inhomogeneities in the presence of orthopedic hardware (40), quantifying cartilage degeneration in osteoarthritis (41), and measuring skeletal muscle extracellular volume fraction in muscular dystrophy (42). Routine clinical adoption of synthetic MRI and MR fingerprinting techniques is hindered by motion artifact associated with requisite longer acquisition times. Additionally, these techniques are predominately 2D, thus lacking benefits associated with 3D isotropic acquisitions. We predict these techniques will become mainstream when further validation of their derived quantitative maps across multiple scanners, field strengths, and patient populations is performed.

Artificial Intelligence Techniques, including DL

Compared with conventional reconstruction approaches (eg, compressed sensing and PI), which exploit data sparsity and coil combinations to reduce acquisition times (typically at the expense of SNR), recent artificial intelligence–based DL techniques use different approaches to achieve higher acceleration or improved image quality. DL acceleration techniques require copious annotated data and processing power to perform supervised learning. These algorithms frequently use convolutional neural networks, which are nonlinear models that obtain parameters from the data itself (43) rather than from physics-based equations. For instance, an early approach, automated transform by manifold approximation (44), learns mapping between the input raw MRI signal in k-space (synonymous to an image aliased with Fourier transform) and the reconstructed MR image (unaliased ground truth). An alternative k-space DL approach learns optimal regularization parameters in constrained image reconstruction (45). Hybrid approaches also exist, which use both image and k-space learning (46). These may outperform earlier approaches, as they ensure that the reconstructed image is more consistent with the original k-space data (47).

In musculoskeletal MRI, DL applications have included end-to-end image networks that permit reconstruction of data sets with even higher acceleration factors into clinically usable images by applying advanced denoising, aliasing and section leakage correction, coil corrections, and superresolution interpolation (48,49). For instance, DL-based superresolution interpolation enables reconstruction of six- and eightfold combined SMS- and PI-accelerated 2D FSE/TSE ankle MRI examinations with exquisite image quality (Fig 6) (49). DL-based reconstruction to denoise and sharpen 2D MR images has been introduced by major MRI vendors and includes AIR Recon DL (GE Healthcare), Deep Resolve Boost (Siemens Healthineers), and SmartSpeed (Philips) (50–52). Additionally, these algorithms have also been extended to 3D MRI to improve spatial resolution in the cervical and lumbar spine (28,29) and to high-resolution MR neurography (Fig 7) (53). Artificial intelligence algorithms have also been developed to generate fat-suppressed data from non–fat-suppressed MRI (54), which can reduce the need for acquiring multiple planes with fat suppression.

Figure 6:

A 3.0-T two-dimensional turbo spin-echo MRI examination was performed in a 58-year-old female patient presenting with ankle pain and swelling by using a 4-minute combined, sixfold acceleration protocol (with simultaneous multislice, parallel imaging, and superresolution deep learning reconstructions). (A) Axial (Ax) proton density (PD), (B) T2-weighted fat-suppressed (T2FS), (C) sagittal (Sag) T1-weighted, (D) sagittal T2-weighted fat-suppressed, and (E) coronal (Cor) proton density fat-suppressed (PDFS) images show an osteochondral lesion within the medial apex of the talar dome (white arrows) characterized by subchondral cyst formation, flattening of the subchondral plate, and proton density and T2 signal hyperintensity of the overlying articular cartilage. The axial (A) proton density and (B) T2-weighted fat-suppressed images also demonstrate midsubstance Achilles tendinopathy (black arrows) characterized by expansion of the anterior-posterior tendon diameter and increased proton density and T2 intrasubstance signal.

Figure 7:

Peripheral nerve-sheath tumor in a 76-year-old male patient with chronic right leg weakness. Oblique coronal three-dimensional multiecho in steady-state acquisition (MENSA)/dual-echo steady state MR neurography images of the lumbosacral plexus reconstructed (A) without and (B) with a commercially available deep learning (DL) algorithm (AIR Recon DL, GE Healthcare) demonstrate a peripheral nerve-sheath tumor (thick arrows) arising from the lower right lumbosacral plexus. Note the increased sharpness and conspicuity of other nerve branches (eg, obturator nerve [thin arrows]) and less noise of the surrounding soft tissues in the DL-reconstructed image.

While DL and other artificial intelligence techniques hold enormous potential to accelerate imaging, they still require appropriate metrics for image integrity. A recent study (55) demonstrated how estimating the reconstruction error might not reflect the diagnostic value of the reconstructed images. Hence, perceptual similarity metrics may better replicate human interpretation of images (56). As artificial intelligence techniques become more mainstream, continued efforts are needed to detect instances when such techniques fail; for instance, reconstruction errors of subtle anomalies may lead to false-negative findings that could impact DL approaches for reliable clinical use. Ongoing monitoring and validation after implementation to verify safety and performance are particularly important, as noninterpretative artificial intelligence tools (ie, those not directly involved in medical image interpretation) are likely to enter the market more rapidly than interpretative ones (tools that assist in medical image interpretation). Artificial intelligence–based image reconstruction algorithms are usually categorized as noninterpretative tools. They are considered less risky for patients than interpretative tools, and hence carry lower regulatory barriers for U.S. Food and Drug Administration approval and clearance or Conformité Européenne marking (57). Additionally, unlike noncommercial open-source artificial intelligence algorithms, commercial artificial intelligence tools may limit disclosures of their methodology due to intellectual property concerns. To enhance future adoption by radiology practices, increased disclosures will likely be needed to allay concerns of altered diagnostic information by these algorithms.

Imaging around Orthopedic Implants

As total joint replacements continue to increase in the United States (58), MRI will likely be increasingly used to identify complications associated with component loosening, including synovial-based soft-tissue reactions to arthroplasty wear and osteolysis. Advanced MRI methods such as 3D multispectral imaging are often used, particularly in the presence of cobalt-chromium alloys that impart strong susceptibility effects on surrounding tissues, as conventional adjustments (eg, higher receiver bandwidths, smaller voxels) do not routinely mitigate susceptibility artifact. Three-dimensional multispectral imaging techniques, such as multiacquisition variable-resonance image combination, have already demonstrated the following impacts for arthroplasty evaluation: (a) distinguishing synovial responses based on the bearing type of the implanted device (59); (b) uncovering associations between MRI synovial reaction classification with revision indication and polyethylene insert damage in total knee arthroplasty (60); and (c) identifying longitudinal changes in synovial patterns following total knee arthroplasty, which is important for noninvasive monitoring, especially when patient-reported outcome measures are ineffective for assessing joint integrity (61).

To increase confidence in determining the presence of component loosening, techniques for improved in-plane resolution and reduced blurring of 3D multispectral imaging are needed to more clearly depict osteolysis at the metallic-bone and cement-bone interfaces. Such improvements in image quality typically incur increased scanning times. Aside from compressed sensing, as previously discussed, a calibration scan can be used with multispectral imaging sequences to reduce scanning time. The calibration scan optimizes the number of spectral bins (62) imaged near an implanted device (63) to minimize susceptibility effects. While these approaches are impactful, applying 3D DL or synthetic MRI may further improve spatial resolution and blurring. Previous metal artifact reduction techniques for arthroplasty imaging have focused on qualitative sequences, but we anticipate that future directions involving multispectral imaging will incorporate quantitative T2 and diffusion mapping (64) to improve characterization of adverse tissue reactions and infection (Fig 8).

Figure 8:

Multiacquisition variable-resonance image combination, or MAVRIC, three-dimensional techniques for morphologic and quantitative imaging in a 67-year-old female patient with pain after left total hip arthroplasty. (A) Coronal MAVRIC proton-density (PD) image demonstrates the total hip arthroplasty in situ. (B) Conventional high-bandwidth coronal proton density image anterior to the total hip arthroplasty shows signal dropout, pileup, and through-plane distortion (arrows) within the iliopsoas bursa, which prevents full visualization of the synovial reaction (arrowheads); these artifacts are mitigated with (C) a MAVRIC acquisition, which also demonstrates intermediate signal intensity–dependent debris (presumed metallosis) (black arrow). (D) MAVRIC T2 mapping and (E) diffusion-weighted imaging (DWI) sequences provide quantitative T2 and apparent diffusion coefficient (ADC) values, which help to characterize this abnormal synovial reaction.

Cartilage Imaging

Over the past 2 decades, multiplanar 2D FSE/TSE with intermediate echo times has been sufficiently optimized to evaluate internal and external joint derangements and specifically cartilage damage (65). However, compared with multiple separate 2D acquisitions, high-through-plane 3D acquisitions with multiplanar reformation capabilities provide potentially superior diagnostic performance for cartilage evaluation due to reduced partial volume effects (66). With the combination of non-DL and DL acceleration techniques, we anticipate that 3D FSE with superior through-plane resolution will supersede use of 2D FSE sequences for cartilage and overall joint evaluation once the aforementioned blurring and SNR challenges are addressed.

Quantitative MRI techniques for cartilage evaluation are well-studied and can be used to detect microstructural changes even earlier than qualitative MRI (Table 3) (67), including evaluating effects of applied load to cartilage (68). Quantitative MRI T2 and T1ρ mapping techniques are complementary validated methods used to depict disruption in the organization and biochemical composition of articular cartilage during osteoarthritis onset and progression. Evaluating the deep zone of articular cartilage or the osteochondral junction, however, is more challenging due to the very short T2 values (<5 msec) of these highly ordered tissues. Ultrashort-echo-time (UTE) imaging techniques, discussed in greater detail later in this review, reduce the effective image echo time to about 30 μsec and permit direct visualization of these zones (Fig 9). Thus, it is now possible to perform concurrent T2 and T2* mapping measurements to comprehensively assess chondral changes following injury or during early degeneration (69).

Table 3:

Summary of Quantitative MRI Techniques for Cartilage Evaluation

Figure 9:

Quantitative T2* maps, derived from a multiecho, three-dimensional ultrashort-echo-time radial-cones acquisition, in an asymptomatic 19-year-old male collegiate basketball player. (A) T2* prolongation of cartilage over posterior margins of the lateral tibial plateau and femoral condyle (blue arrows) are noted, suggesting early chondral degeneration. (B) The lateral meniscus anterior horn also demonstrates early degeneration, reflected by T2* prolongation (white arrows) relative to the posterior horn.

Other novel cartilage applications include sodium imaging and chemical exchange saturation transfer between bulk water protons and glycosaminoglycans on proteoglycan molecule imaging, which directly assess the biochemical composition of articular cartilage. Although SNR and chemical shifts at 3.0 T are current limitations (70,71), we anticipate that increased use of higher-field-strength systems (7.0 T and higher) will enable broader adoption of these techniques. Another novel approach to cartilage evaluation is diffusion imaging that corresponds with the Osteoarthritis Research Society International cartilage score (72), but thus far, this approach has only been used to evaluate ex vivo samples at ultrahigh field strengths (7.0 T and higher). In vivo diffusion imaging is challenging due to long echo times that inadequately capture signal from the short T2 values of cartilage and insufficient spatial resolution.

Over the next decade, we anticipate that the musculoskeletal community will more routinely use phantoms and standardized protocols to help minimize sources of quantitative error that currently impede wider adoption of quantitative cartilage MRI techniques (73). Additionally, deployment of newly developed artificial intelligence–guided automated segmentation techniques to extract cartilage thickness and volume will enable improved lesion detection, prediction of osteoarthritis outcomes, and generation of 3D models for surgical planning (74). Finally, postoperative imaging evaluation may involve labeling of matrix-associated stem cell implants with superparamagnetic ferumoxytol nanoparticles to evaluate graft outcome success (75).

UTE and Zero-Echo-Time MRI

Cortical bone, ligaments, tendons, and menisci comprise a highly organized ultrastructure with very short T2 and T2* values (1 µsec to 11 msec) that result in rapid decay of transverse magnetization. Consequently, these tissues lack signal during conventional FSE/TSE imaging, which uses minimum echo times of about 10 msec. Sequences with sufficiently short echo times sample the free-induction decay, similar to gradient-recalled echo, and include UTE imaging, water- and fat-suppressed proton projection MRI, sweep imaging with Fourier transformation, pointwise encoding time reduction with radial acquisition, and zero-echo-time (ZTE) imaging (76–80). It is only recently, however, that ZTE techniques have become commercially available and UTE prototypes supported by some major MRI vendors. The choice of UTE (echo time <0.1 msec) or ZTE (echo time of 0 msec) technique largely depends on vendor availability and intended applications (Table 4).

Table 4:

Comparison of GRE, UTE, and ZTE Sequences

An immediate clinical translation of either ZTE or UTE MRI is to enhance signal intensity in cortical bone. UTE and ZTE sequences facilitate near-isotropic acquisitions within reasonable scanning times (<7 minutes), approach the voxel resolution of energy-integrating detector CT (81), and support multiplanar reformation, volume rendering, and segmentation tasks. Cortical bone evaluation with ZTE has been applied in the context of shoulder trauma, sacroiliitis, and spine degeneration (Fig 10) (82,83). Compared with conventional gradient-recalled echo acquisitions, ZTE MRI provides superior visualization of osseous detail, with substantial agreement found between ZTE MRI and CT in paired imaging studies (84). Benefits of ZTE sequences include (a) decreasing patient exposure to ionizing radiation from CT, especially for the pediatric and pregnant populations, and (b) conserving health care resources by obviating the need for concurrent CT scans in certain patients. Hence, ZTE sequences will likely revolutionize current algorithms, as evaluation of both cortical bone and soft tissue are now feasible within a single MRI examination.

Figure 10:

Zero-echo-time MRI scans in four patients after acute trauma. (A) Oblique coronal image of the glenohumeral joint in a 50-year-old male patient demonstrates a comminuted greater tuberosity fracture (arrow). (B) Sagittal image of the left elbow in a 33-year-old male patient demonstrates a superiorly displaced triceps avulsion fracture (arrow). (C) Coronal wrist image in a 61-year-old male patient shows small capsular triquetral avulsion fracture (arrow). (D) Sagittal ankle image in a 39-year-old male patient demonstrates Achilles tendon avulsion fracture off the calcaneus, with retraction of osseous fragments (arrows). Images courtesy of Ryan Breighner, PhD, Hospital for Special Surgery.

UTE imaging has greater flexibility than ZTE imaging for morphologic and quantitative evaluation of short-T2 tissues. Protocols that use UTE sequences span a wide echo time range to acquire both short (<10 msec) and ultrashort T2 (<1 msec) musculoskeletal tissue component signals and optionally add either fat suppression or preparation pulses to increase image contrast (Fig 11). The plurality of imaging options that are actively researched include dual-echo UTE with echo subtraction and adiabatic inversion-recovery UTE, including dual inversion-recovery UTE, double inversion-recovery UTE, and inversion-recovery fat-saturated UTE (81). Additionally, UTE techniques are compatible with quantitative T1, T2*, T2, T1ρ, magnetization transfer ratio, magnetization transfer–based modeling of macromolecular fraction, perfusion, diffusion, and susceptibility mapping (81,85). Clinical adoption of UTE imaging will likely increase, as studies have suggested these techniques may provide a more sensitive biomarker of structural change than morphologic evaluation and may be useful for monitoring treatment response in patients (86).

Figure 11:

Three-dimensional (3D) ultrashort-echo-time (UTE) MRI scans of the osteochondral junction in a (A–D) normal cadaveric knee from a 62-year-old male donor and an (E–H) osteoarthritic cadaveric knee from a 87-year-old male donor. (A, E) Two-dimensional T2-weighted (T2) fast spin-echo (FSE), (B, F) 3D T1-weighted fat-saturated (T1-FS) UTE, (C, G) 3D inversion-recovery fat-saturated (IR-FS) UTE, and (D, H) 3D dual inversion-recovery (DIR) UTE images. In the (A–D) normal knee, the osteochondral junction region was dark on the T2-weighted FSE image due to its fast signal decay. The normal osteochondral junction regions were highlighted as bright signal lines in T1-weighted fat-saturated UTE, inversion-recovery fat-saturated UTE, and dual inversion-recovery UTE images. In the (E–H) osteoarthritic knee, the osteochondral junction signals were reduced or completely lost in the abnormal cartilage regions (arrows) of central osteophyte formation. Of note, there was preservation of overlying cartilage in these regions. Images courtesy of Yajun Ma, PhD, University of California San Diego.

Despite current challenges of both ZTE and UTE imaging, including complex postprocessing and hardware requirements, we anticipate that UTE and ZTE sequences will become available across more MRI scanners and vendors. Additionally, UTE and ZTE implementation at 0.55 T may increase to overcome susceptibility artifacts involving metallic hardware (eg, postoperative spine evaluation) or air (eg, temporomandibular joint evaluation).

MR Neurography

MR neurography was first coined by Howe et al (87) in 1992 in a study using T2-weighted fat-suppressed spin-echo and diffusion-weighted imaging to depict the nerve in the rabbit hindlimb. Subsequently, it has been defined by heavily T2-weighted, predominantly qualitative 2D and 3D sequences that maximize spatial resolution as well as nerve-to-background (fat, muscle) tissue contrast. MR neurography is an important adjunct to electrodiagnostic testing to diagnose and localize peripheral neuropathies, particularly when lesions are multifocal or fascicular (rather than involving the entire nerve diameter) (88). Furthermore, MR neurography has been shown to influence surgical planning and outcomes by precisely pinpointing and characterizing lesions to guide the appropriate surgical plan (89,90). While MR neurography is often used in neuroradiology, musculoskeletal-trained radiologists have taken a particular interest in the modality, particularly when imaging peripheral nerves of the extremities distal to the brachial and lumbosacral plexi from where these nerves originate. US is considered complementary to MR neurography for nerve assessment, but MRI provides greater contrast resolution to depict nerve pathology and greater access to deeper nerve segments (such as the lumbosacral plexus) (91).

Specialized vascular suppression techniques, many developed for vessel wall imaging, have been adapted for MR neurography. Vascular suppression is typically required to reliably differentiate small-caliber nerves from adjacent small vessels (arteries or veins) that may demonstrate similar signal intensity. Noncontrast suppression techniques (eg, motion-sensitized driven equilibrium, nerve-sheath signal increased with inked rest-tissue rare imaging) typically implement dephasing gradient waveforms for flow suppression (92,93). A combination of intravenously administered gadolinium contrast agent with T2-weighted MR neurography using a short-tau inversion-recovery pulse can simultaneously suppress both fat and fluid (Fig 12). More recently, ferumoxytol (Feraheme, AMAG Pharmaceuticals) has also been evaluated for off-label use for vascular suppression given its much longer half-life and shorter T2 relaxation rate compared with gadolinium and has shown greater flexibility between administration and timing and a stronger vascular suppression effect (94).

Figure 12:

Parsonage-Turner syndrome (acute brachial neuropathy) in a 38-year-old female patient with severe left shoulder pain that began 2 weeks before imaging, followed by weakness. (A) Coronal two-dimensional T2-weighted Dixon water MRI scan demonstrates denervation edema of the supraspinatus (SS) and infraspinatus (IS) muscles. Due to improved vascular suppression, (C) gadolinium-enhanced three-dimensional (3D) oblique coronal short-tau inversion-recovery (STIR) maximum intensity projection (20-mm thick) shows severe intrinsic constrictions (arrows) more clearly compared with the (B) noncontrast projection using an otherwise identical 3D short-tau inversion-recovery protocol.

MR neurography can also be used to assess skeletal muscle for the presence of active denervation (which manifests as diffusely high T2 signal intensity) and chronic denervation, during which irreversible fatty infiltration may ensue. Because needle electromyography involves individual assessment of muscles (some inaccessible), a key advantage of MR neurography is its ability to simultaneously display multiple muscles with a single field of view. Quantitative MRI markers have also shown early promise for more objectively inferring muscle changes related to nerve injury. One such marker is transverse relaxation time (T2) (derived from a T2 mapping sequence), which depicts muscle edema from extracellular fluid (95), shows superiority to routine qualitative T2-weighted sequences (96), and correlates with electromyography grades of motor unit recruitment and denervation potentials (97). Another example is a diffusion-derived muscle diameter metric (98) that microscopically depicts atrophy following denervation and also correlates with electromyography (97). In addition, an animal study suggests that quantitative T2 increases may be detected earlier than qualitative muscle edema (99), which may be helpful in the early stages of axonal degeneration (Wallerian degeneration) when electromyography is less reliable to initially detect denervation (<21 days).

Future directions in MR neurography include manipulating 3D data sets for advanced 3D visualization, along with cinematic rendering, to facilitate surgical in situ view simulations. Continued improvements in radiofrequency coil designs and potentially imaging at 7.0 T may facilitate higher spatial resolution acquisitions to improve disease characterization. Finally, PET/MRI may hold value in providing functional information and increased sensitivity and specificity in patients with pain that cannot otherwise be localized (100).

Conclusion

Given the dynamic nature of the field, future directions of MRI as they relate to musculoskeletal health are hard to predict. However, we anticipate continuous improvements in field homogeneity and gradient performance, as well as the development of radiofrequency coils specifically designed for musculoskeletal MRI examinations, across a broad range of magnet strengths. Optimization of three-dimensional acquisitions to improve signal-to-noise ratio and reduce blurring will be supported by acceleration and deep learning image reconstruction approaches. Synthetic MRI and MR fingerprinting approaches are predicted to gain further traction due to their potential to simultaneously provide qualitative and quantitative information to characterize pathology. We also anticipate heightened demand for metal artifact reduction techniques, given the increasing rate of joint arthroplasty, and for MR neurography, given the increasing interest in this technique among neuromuscular specialists. Finally, broader adoption and refinement of quantitative MRI techniques to characterize tissue microstructure of both long-(cartilage, muscle) and fast-decaying (tendons, ligaments, cortical bone) T2 species will be needed to meet the ever-increasing demands for greater specificity and personalized care among members of the musculoskeletal community.