Introduction

By the end of the 1990s, high-field MRI at 3 Tesla had become the benchmark for routine clinical applications, as well as for clinical MRI research. The clinical benefits of the signal-to-noise ratio (SNR) that is twice that of standard 1.5 T MR consist mainly of the combined morphological and functional high-field MR methods possible at this field strength, such as functional MR, metabolic imaging, and diffusion-weighted imaging [1].

The next important step was taken in the early 2000s [2–5] and involved increasing the MRI field strength by another factor of approximately two, and, presently, at least three major MR vendors provide commercial 7 T units for human clinical research under ethics board-approved conditions. Approximately 10 years ago, there were fewer than five ultra-high-field MR installations in the world. During the last several years, the number of installations finalized or under preparation has increased to about 40 (0.2% of the total installation number). This rapid increase indicates the growing interest in ultra-high-field MRI in the bio-imaging community, brought to the fore by promising results with regard to morphological detail, as well as functional imaging capability [6–8].

Most clinical MR research centers operating at 7 T primarily focus on neuroradiological applications, with only a few sites performing whole-body clinical research at 7 T. This is because of the limited availability of suitable coils for 7 T, which must be transmit-and-receive coils, the higher B0 and B1 heterogeneity related to the ultra-high field, as well as safety issues, particularly the increased specific absorption rate, which is about 22 times higher at 7 T compared to 3 T.

In this review, we want to provide an overview of the initial clinical results of musculoskeletal studies performed at 7 T, with a special focus on sodium imaging, new techniques such as chemical exchange saturation transfer (CEST) and T2* imaging, and multinuclear metabolic imaging or MR spectroscopy (MRS) in tissues such as cartilage, tendons, and muscles.

Sodium MRI at 7 T

Sodium imaging of cartilage and cartilage repair

Proton MRI at ultra-high-field (7 T and above) MR systems poses various challenges, including radiofrequency power deposition, increased chemical shift, susceptibility artifacts, homogeneity of RF coils, and changes in relaxation times, compared to lower-field clinical MR systems. These problems are much less pronounced in nuclei with a low gyromagnetic ratio, such as sodium (23Na). However, a 3.8-fold smaller gyromagnetic ratio, a lower resonance frequency (23Na: 78.6 MHz vs. 1H: 297.2 MHz), a significantly shorter T2 relaxation time, and a lower concentration of sodium nuclei (23Na: 300 mM vs. 1H: 110 M) result in sodium MR signal in articular cartilage that is 1/4,000–1/5,000 smaller than the proton MR signal. In order to achieve sufficiently high SNR (SNR > 15), sodium MRI requires longer measurement times (15–40 min) and results in low-resolution images. Ultra-high-field MR systems can provide higher intrinsic 23Na SNR and/or higher spatial/temporal resolution and improved contrast in the image [9, 10]. The MR relaxation properties of sodium nuclei in biological systems are dominated by quadrupolar interactions and result in short longitudinal (T1) and transverse relaxation (T2) times. Transverse relaxation is characterised by biexponential decay with a short component (T2 short) that represents 60% of the total sodium signal and a long component (T2 long) that represents 40% of the sodium signal. Due to the short T1 of sodium, rapid averaging can be used in sodium MRI in order to improve the SNR. However, the short T2 long (∼12 ms) and the even shorter T2 short (∼1 ms) result in significant loss of sodium signal when using conventional MR sequences with an echo time longer than 2 ms [11, 12]. Therefore, many recent studies on sodium MRI of articular cartilage at 7 T have used MR techniques with non-Cartesian k-space trajectories, such as 3D radial [13–17] or 3D cones [10], which enable ultra-short echo time (UTE) acquisition.

Madelin et al. applied compressed sensing to undersample a 3D radial UTE sequence for sodium MRI at 7 T [15]. These authors showed that compressed sensing can be applied to reduce the acquisition time by a factor of two at 7 T without losing sodium quantification accuracy. The nonlinear reconstruction used in compressed sensing can be used as an alternative to denoise fully sampled images. Parallel imaging techniques combined with sequences using non-Cartesian k-space trajectories at 7 T seem to be promising in further reducing the acquisition time of sodium MRI.

Staroswiecki et al. compared the sodium SNR at 3 and 7 T and verified that SNR scales linearly with field strength [10]. Sodium T2* values in patellar cartilage, measured with 3D cone trajectory, were not significantly different between 3 and 7 T and were consistent with previously reported relaxation times.

The sodium relaxation times in four different regions of the cartilage in the knee joint were measured in vivo on eight healthy volunteers at 7 T and reported by Madelin et al. [17]. The average sodium T1 was about 20 ms, but T2*short ranged from 0.5 to 1.4 ms and T2*long from 11.4 to 14.8 ms. Statistically significant differences in T1, T2*short, and T2*long were observed between the different regions in cartilage. Presented relaxation time measurements can be applied as correction factors for the calculation of sodium content in the articular cartilage.

Sodium MRI offers many potential clinical applications. Sodium content measured by 23Na MRI has been shown to be proportional to glycosaminoglycan (GAG) content in articular cartilage [18, 19]. Sodium MRI can be used to detect early signs of cartilage degeneration or injury before morphological changes can be detected by proton MRI [20]. Sodium MRI may enable the noninvasive in vivo evaluation of disease-modifying treatments for osteoarthritis (OA) and methods for cartilage repair. Affecting millions of people, OA is the most common degenerative disease of the musculoskeletal system.

In a study by Wang et al., cartilage sodium concentration in five healthy controls and five patients with OA was measured with 3D-radial UTE acquisition [13]. These authors reported a statistically significant reduction in sodium concentration of 30–60% in OA patients when compared to healthy subjects. The findings are in good agreement with previous publications [11, 21] and suggest that sodium MRI at 7 T could be used for the detection of changes in cartilage associated with OA and may help to improve clinical diagnosis.

Because of thinness of cartilage, typically on the order of 3–5 mm, and the low resolution of sodium images, ROI evaluations are prone to partial volume errors, which could result in overestimation of sodium signal due to sodium signal from the synovial fluid. Madelin et al. addressed this issue by employing 3D radial UTE sodium sequences with fluid suppression and inversion recovery (IR) preparation of the magnetisation [14]. These investigators showed that a sequence with an adiabatic IR pulse generates a more uniform fluid suppression over the whole sample than a rectangular IR pulse, and therefore, can better separate sodium signal from the cartilage. Because of the specific absorption rate limits, the total acquisition time was about 17 min for the 3D radial sequence without IR or with the rectangular IR, and about 24 min for the adiabatic IR sequence.

Madelin et al. documented the reproducibility and repeatability of sodium quantification in cartilage in vivo by using a radial 3D sequence, with and without fluid suppression at 3 and 7 T [16]. These investigators did not observe any significant intermagnet, intersequence, intraday, or interday differences in the coefficients of variation. The reproducibility and repeatability were comparable to other proton-based MRI techniques for cartilage. Moreover, the fluid-suppressed sequence resulted in sodium concentrations (252 mM at 3 T; 278 mM at 7 T) that were closer to the values for healthy cartilage reported in the literature (220–310 mM) when compared to results from sequences without fluid suppression (171 mM at 3 T; 177 mM at 7 T). This difference was likely due to a reduction in partial volume errors from synovial fluid in sequences with fluid suppression.

The very short T2 times of sodium in articular cartilage result in sequences with short echo times that are very effective with regard to SNR . However, radial sequences are prone to blurring and artifacts from off-resonance effects, fast components of T2* decay, and variations in gradient timing. Although the Cartesian sampled sequences are less effective in regards to the SNR, they can offer much sharper images. In an article by Trattnig et al., a Cartesian 3D gradient echo sequence was used for sodium MRI at 7 T in patients after matrix-associated autologous chondrocyte transplantation (MACT) [22] (Fig. 1). These authors observed significant differences in sodium normalised values between repaired cartilage transplant tissue and healthy cartilage. Moreover, a strong significant correlation was found between sodium normalised values from 7 T and intravenously enhanced T1 values from another GAG-specific technique—delayed gadolinium-enhanced MRI of cartilage (dGEMRIC) at 3 T. These findings suggest that sodium MRI can differentiate between repaired tissue and native cartilage in MACT patients and may be useful for the noninvasive evaluation of different cartilage repair techniques.

Fig. 1
figure 1

A proton-density-weighted image measured at 3 T (TR/TE = 2,400/38 ms, resolution of 0.2 × 0.2 × 2.0 mm3) (a) and a corresponding sodium image at 7 T (TR/TE = 10/3.8 ms, resolution of 1.6 × 3.1 × 3.0 mm3) (b) of a 23-year-old man, acquired 67 months after matrix-associated autologous chondrocyte transplantation (MACT) on the medial femoral condyle of the left knee. Arrows mark the borders of the transplanted area. The patient had a good outcome and MOCART score of 80 points. Please note lower intensity of sodium signal in repair area, which is suggestive of lower GAG content in the repair tissue than in surrounding native cartilage

Full size image

Sodium imaging of tendons

Degeneration of the Achilles tendon leads to thickening of the tendon [23]. Tendinopathy is also accompanied by disaggregation of the microfibrillar bundles due to the greater quantities of water and proteoglycan [24]. Almost double the GAG content was observed in pathologic tendons in studies using biochemical assays [25]. In a study by Juras et al., the feasibility of sodium magnetic resonance imaging for the diagnosis of Achilles tendinopathy was investigated at ultra-high field [26]. Their cohort comprised 20 healthy volunteers with no history of pain in the Achilles tendon and 8 patients with clinical findings of chronic Achilles tendinopathy. The study found that the mean bulk sodium SNR was 4.9 ± 2.1 in healthy control subjects and 9.3 ± 2.3 in patients with Achilles tendinopathy, and that the difference between the means was statistically significant. Figures 2 and 3 depict the increased sodium SNR in the tendinopathy group. This study not only showed a statistically significant increase in sodium SNR in patients with Achilles tendinopathy, compared with healthy tendons, but also revealed abnormal sodium signal values in the whole tendon, as well as morphologically focal abnormalities with focal thickening. Sodium signal values may correspond to GAG content in the Achilles tendon, which, as shown in in vitro studies, is increased with tendinopathy [26].

Fig. 2
figure 2

Magnetic resonance images from a 27-year-old healthy volunteer. a Proton-density-weighted turbo spin-echo (TSE) image with arrows pointing at the centers of each evaluated region of interest. Parameters: TR/TE 4,000/26 ms; field of view, 140 × 140; slice thickness, 3 mm. Please note that the Achilles tendon thickness and signal intensity are normal. b Corresponding 3D gradient-echo sodium (TR/TE 17/8.34 ms, field of view, 199 × 199; slice thickness, 3 mm) image depicts Achilles tendon sodium signal (arrows), which is relatively low. The color bar represents sodium signal intensity (courtesy of Medical University of Vienna)

Full size image
Fig. 3
figure 3

Magnetic resonance images of a 46-year-old patient with chronic Achilles tendinopathy. a Proton-density-weighted TSE image of the patient with chronic Achilles tendinopathy, with region-of-interest centers (arrows). b Corresponding color-coded 3D-gradient-echo sodium image depicts Achilles tendon sodium signal (arrows), which is obviously higher than that in the healthy volunteer. The sodium signal increase was observed in the whole tendon and not only in regions with clinical findings. Color scale represents sodium signal intensity values (courtesy of Medical University of Vienna)

Full size image

T2* mapping of the Achilles tendon at 7 Tesla

MRI is quite frequently used to evaluate the Achilles tendon (AT). It was successfully employed to detect partial or total tendon rupture or even the degenerative processes in the tendon tissue. Most imaging findings are related to the pathologic processes of tendon degeneration and repair progression [27]. In addition to the morphological MRI evaluation (imaged predominantly by T2*-weighted sequences), the quantitative MRI analysis of tendon tissue may be helpful in identifying the early pathological changes in the tissue. Quantitative analysis of the relaxation or diffusion constants [such as T1, T2*, T1ρ, and apparent diffusion coefficient (ADC)] of the AT may provide additional information about the overall condition of the tendon. On conventional high-field MR systems (1.5–3 T), several parameters have been investigated as prospective markers for Achilles tendon degeneration, such as T2* [28], T1ρ [29], ADC [30], magnetisation transfer [31], or spectroscopically, even T2 [32]. Using spectroscopic methods, it was shown that the T2* decay in the Achilles tendon is a multicomponent process [33]. However, with clinical sequences, it is difficult to acquire signal from the second, third, and fourth components, since these have a small component ratio. The first component has the largest ratio, but it is in the submillisecond range, and thus, can barely be acquired with conventional echo times.

In general, MR imaging of rapidly relaxing tissues, such as tendons, menisci, ligaments, and bones, is rather difficult with clinical sequences. Recent developments in new hardware and sequence design allow the acquisition of a signal directly from these tissues. Moreover, ultra-high field provides a substantial increase in the SNR [9]. The 3D-UTE sequence provides the ability to detect MR signal from a large variety of rapidly relaxing tissues and materials, including tendons. Juras et al. used a 3D-UTE sequence at 7 T to estimate T2* in tendons in order to investigate the potential feasibility of using this parameter as a marker for Achilles tendinopathy [34]. The SNR increase between 3 and 7 T was validated by this study as well. Ten volunteers with no history of pain in the AT and five patients with chronic Achilles tendinopathy were recruited. T2* was acquired by fitting bi-component exponential functions, and both short (T2*s) and long (T2*l) were evaluated. With regard to the comparison of patients and volunteers at 7 T, the bulk T2*s was significantly higher in patients, although the bulk T2*l difference was not statistically significant. It seems that ultra-short bi-component T2* measurements in the human Achilles tendon in vivo are feasible using a 3D-UTE sequence. Higher SNR at 7 T allows the calculation of both T2* components very accurately. Figure 4 demonstrates a sample fit of signal intensities acquired by the 3D-UTE sequence. The observed differences between T2*s in healthy and abnormal tendons suggest that advanced quantitative imaging of the human AT may provide additional information beyond standard clinical imaging in reasonably short MR data acquisition times. T2* may be a promising marker for the diagnosis of early tendinopathy in the AT [34].

Fig. 4
figure 4

An example of the accuracy of bi-component T2* fitting in the Achilles tendon at 7 T (courtesy of Medical University of Vienna)

Full size image

Chemical exchange saturation transfer (CEST) at 7 T

Saturation transfer (ST) is a commonly used technique in nuclear magnetic resonance (NMR) [35] and has been proposed as a method for the direct detection of chemical exchange between bulk water protons and protons bound to solutes [36]. The resultant MR imaging scheme is referred to as CEST MRI [37, 38]. The basic principle of CEST imaging is a reduction in bulk water MR signal after off-resonant spins are selectively pre-saturated by radiofrequency (RF) irradiation and then undergo chemical exchange with bulk water protons [39].

The hydroxyl and amide protons of glycosaminoglycans (GAG) provide exchange properties that render them principally suited for CEST experiments [40]. In vitro experiments at 11.7 T demonstrated that CEST imaging can be used as a biomarker for cartilage GAG content (gagCEST) in bovine cartilage samples [41, 42]. However, GAG –OH protons resonate at frequency offsets (Δω) of only 1 and 1.5–2 ppm downfield from bulk water, and rate constants of chemical exchange (k) can be on the order of 1,000 s−1 [42]. At a magnetic field strength of 3.0 T, the Δω of hydroxyl protons corresponds to a separation from bulk water of 128 Hz and 192–256 Hz in frequency units. As a consequence, radiofrequency power intended to selectively saturate –OH resonances simultaneously attenuates the bulk water signal (RF spillover), which impairs quantification of the CEST effect at 3.0 T. At higher fields, i.e., with increasing frequency differences, the RF spillover decreases, making ultra-high-field strengths, such as 7.0 T, ultimately favorable for CEST experiments.

After the introduction of the gagCEST approach, further studies have been performed with animals and humans to assess the feasibility of gagCEST imaging in vivo [43–45], and several imaging sequences for fast and reliable detection of gagCEST effects have been proposed [43, 46–48]. Most techniques are based on multiple image acquisition with pre-saturation at different offset frequencies (Δω). The remaining bulk water signal (MSat), normalised to a reference (MRef), is then plotted against the RF presaturation offset (z-spectrum). CEST effects are asymmetric with respect to the water resonance, and thus, calculation of the asymmetry of the magnetisation transfer ratio (Eq. 1) \( ({\text{MT}}{{\text{R}}_{{asym}}}(\Delta \omega ) = ({{\text{M}}_{\text{Sat}}}( - \Delta \omega ) - {{\text{M}}_{\text{Sat}}}( + \Delta \omega ))/{{\text{M}}_{\text{Ref}}}) \) provides a positive measure of CEST effects. Cartilage magnetisation transfer resonance asymmetry (MTRasym) values scale linearly with the ratio of GAG protons to bulk water protons, given that temperature and pH values inside cartilage tissue are spatially equivalent within a sample, and MTRasym is evaluated at frequency offsets with underlying resonances of exchangeable GAG protons [42].

From the initial studies, the key requirements for in vivo gagCEST examinations can be derived: For accurate quantification of gagCEST effects, it is essential to account for inhomogeneities of the static magnetic field B0 in a sample. For this purpose, fitting of the absorption curve and a separate reference MR data acquisition have been proposed as possible methods [43, 49, 50]. Furthermore, compensation for sample movement during the course of a measurement is crucial for accurate assignment of signal evolution to a certain location. Data from a separate B0 reference also have to be aligned with the measurement data, which induces a greater registration effort.

Despite the requirements mentioned above, gagCEST imaging is a valuable tool for the non-invasive assessment of GAG content in vivo. A recent study demonstrated that gagCEST can be used to reliably detect GAG in the knee cartilage of patients who had undergone cartilage repair surgery [51] (Fig. 5). This study was conducted at 7.0 T with a 3D GRE-based measurement technique, and 23Na MRI was used as a reference for GAG measurements. Moreover, the potential of gagCEST for GAG evaluation in intervertebral discs at 3.0 T has been demonstrated [52, 53]. Given the results from the latter studies, it seems possible that gagCEST can also be used to detect cartilage GAG content at 3 T, which indicates the potential of this approach for use in the clinical routine. The main strength of gagCEST compared to other GAG-sensitive imaging techniques, such as dGEMRIC and sodium imaging, is the relatively short acquisition time, which covers the entire volume of a knee joint in ∼10 min [51], and gagCEST does not require administration of contrast agent and can easily be implemented into a standard imaging protocol.

Fig. 5
figure 5

Sagittal knee images from a 49-year-old patient with a secondary cartilage defect after a tear of the anterior cruciate ligament. The central part of the lesion can be delineated in the morphologic proton-density-weighted image (a). The gagCEST (b) and sodium images (c) show a signal loss in the lesion compared to normal reference tissue. Color bars in b and c represent magnetisation transfer resonance asymmetry values summed over offsets from 0 to 1.3 ppm (gagCEST) and sodium signal-to-noise ratio, respectively

Full size image

Metabolic imaging of muscles at 7 T

Another MR method that has the potential to become increasingly important in clinical musculoskeletal MR at 7 T is metabolic imaging or MR spectroscopy (MRS). MRS is a powerful noninvasive tool for the investigation of metabolite concentrations and studies of bioenergetics that could otherwise only be assessed by invasive muscle biopsies [54]. MRS provides information on a cellular level beyond the anatomical information assessed by standard imaging methods and aids in the understanding of various lesions [54–56], clinical diagnosis [56, 57], and treatment monitoring [58, 59].

The improvements in data quality for MRS techniques resulting from the higher B0 (i.e., 7 T) are larger than for standard MR imaging methods [60, 61]. In addition to SNR improvements, 7 T MRS also offers higher spectral resolution. SNR and spectral resolution together contribute to more reliable quantification of MRS data (i.e., more than three times higher than at 3 T [61, 62]) and enable a significantly more rapid acquisition time for MRS experiments [60, 62–64]. Based on the detected nuclei, MRS techniques can be classified as 1H-MRS, 31P-MRS, or 13C-MRS. Multinuclear MRS (i.e., 31P-MRS and 13C-MRS), in particular, improves significantly at 7 T [60, 62–66].

1H-MRS provides insights into both lipid metabolism by quantification of intramyocellular (IMCL)/extramyocellular lipids (EMCL) [54, 67] and cancer metabolism by quantification of Cho compounds [68]. Recent reports on studies of lipid metabolism of the calf muscle at 7 T show more accurate and reliable results due to improved separation of individual IMCL and EMCL resonances by both 1D [69, 70] and 2D [70] correlated 1H-MRS. This can aid in clinical investigations of pathologic or training-related alterations in lipid muscle metabolism.

The 13C-MRS technique allows the in vivo measurement of glycogen, which is one of the major energy sources in the skeletal muscle besides IMCL/EMCL [71, 72]. One preliminary study at 7 T provides evidence of the significantly improved spectral quality along with a reproducibility that is more than three times better than that at 3 T [61]. Although 13C-MRS is not likely to enter clinical use, even with these improvements, 13C-MRS can nevertheless improve the understanding of pathological changes in glucose metabolism in muscle disease and could be applied in clinical trials that are designed to evaluate new drugs and therapies.

31P-MRS is a popular tool that reveals the major factors of muscle metabolism noninvasively in a resting state and during muscle exercise [54, 62, 66, 73]. It allows the quantification of temporal changes in high-energy phosphates [66, 73], intracellular pH [54], and fluxes of creatine kinase and ATPase reactions [54, 62], which are important indicators of muscle metabolism. Several groups use 31P-MRS regularly for the investigation of muscle diseases in basic [57, 59, 74] and clinical [54, 56] research. In addition, 31P-MRS provides more specific markers of cancer metabolism than 1H-MRS [68].

In addition to SNR and spectral resolution improvements, T1 relaxation of 31P metabolites at 7 T is twice as fast as at 3 T (note that 31P is the only nucleus with decreasing T1) [60]. This improves the measurement speed, allowing acquisition of high quality data within a few minutes [60, 62–64] and improved spatial specificity [63, 64, 66]. In addition, at 7 T, there was an ∼3 times higher reproducibility of 31P-MRS experiments than at 3 T [62] (Fig. 6). Apart from improved quantification of known metabolites, 7 T 31P-MRS might even be able to separately detect intramitochondrial inorganic phosphate pools [65].

Fig. 6
figure 6

a Sample spectra with high spectral quality obtained in a total of 6 min by a magnetisation transfer experiment in the calf muscle of a healthy volunteer. To determine the forward rate constant of the ATPase reaction in the resting state accurately, it is necessary to acquire one spectrum with suppression of γ-ATP at −2.48 ppm and one spectrum with the suppression pulse set to 12.52 ppm (i.e., mirrored frequency position relative to Pi). b A box plot diagram illustrates the significantly higher intra- and intersubject reproducibility of measuring the time-resolved forward rate constant of the ATPase reaction (kATP) in the calf muscle of six healthy volunteers at 7 T compared to 3 T. For this assessment each subject was measured four times

Full size image

The availability of fast and robust MRS methods at 7 T will provide new opportunities for imaging a large clinical spectrum of musculoskeletal diseases, such as mitochondrial disorders [75, 76], glycolytic defects [77], systemic diseases affecting muscle metabolism [56], muscle injury [78], or diabetes [74, 79], for diagnostic use [56, 57], therapy monitoring [58, 59], and clinical research [56].

In conclusion, these initial clinical studies demonstrate the potential of ultra-high-field MR at 7 T, with the advantage of significantly improved sensitivity for other nuclei, such as 23Na (sodium) and 31P (phosphorus). This will provide new insights into normal and abnormal physiology of musculoskeletal tissues and the metabolism of muscle, and will, therefore, provide new in vivo clinical applications.