Morphological MRI of the brain is a powerful method in assessing atrophy in patient cohorts, healthy subjects and normal aging [6,7]. There is no general rule about which contrast should be used to assess morphology, but generally T₁-weighted images are used to segment neuroanatomical structures such as the hippocampus, gray/white matter boundaries and the ventricles. T₂ sequences provide high image intensity for structures, which are filled with cerebrospinal fluid (CSF) and is, therefore, suitable to segment the ventricles. MP-RAGE (Magnetization Prepared RApid Gradient Echo) [8] is a very common gradient echo sequence used to acquire T₁-weighted data. MP-RAGE uses a non-selective 180° pulse to prepare the magnetization in order to achieve a certain contrast followed by a RAGE (RApid Gradient Echo) sequence for image acquisition. The major advantage of MP-RAGE is that it can be used for 3D image acquisition, increasing the signal to noise ratio (SNR) in proportion with the number of slices. Another advantage of MP-RAGE is that gradient echoes are used for data acquisition, which is important in order to keep the Specific Absorption Rate (SAR; see Appendix) low. This is especially important for ultra-high field MR-systems (>3 T). The achievable image resolution using MP-RAGE under clinical conditions, with scan times of about 10-15 min for the brain, is about 0.80 mm isotropic (0.51 mm³) on a 3 T system and 0.65 mm isotropic (0.27 mm³) on a 7 T system.

In order to acquire T₂-weighted data, Turbo Spin Echo (TSE) or Fast Spin Echo (FSE) measurement sequences [9] are mainly used, where TSE and FSE are in principal very similar, the difference in naming being a manufacturers’ convention. As T₂-weighted sequences work with spin echoes, requiring higher RF-energy, the SAR limit can be reached easily on 3 T systems. This problem becomes even worse on 7 T systems where only a small number of slices can be acquired without exceeding the SAR limit. Currently, there are a number of methods under development in order to reduce the RF energy deposition and to acquire T₂-weighted data on 7 T systems [10].

Susceptibility Weighted Imaging (SWI) is a novel imaging technique that involves both MRI and image processing, exploiting the magnetic susceptibility difference between different tissues to enhance contrast in MR magnitude images [11].

Magnetic susceptibility, which is the degree of magnetization of a substance in response to an external magnetic field B₀, depends on the tissue’s chemical composition, state, and density. A difference in magnetic susceptibility between neighboring tissues produces a different signal phase that is proportional to the echo time (TE) and the magnetic field strength B₀. This phase difference can be measured and reconstructed as an image using a SWI protocol. Unlike commonly acquired magnitude images, SWI does not require additional scan time as complex data, i.e., magnitude and phase, are acquired in one scan. Phase images extracted from this complex data are used to enhance contrast in MR magnitude images, but also provide clinically interesting contrast in their own right.

SWI has clear advantages as a high field method, as contrast is proportional to the magnetic field strength B₀, and a gradient-echo sequence is used for data acquisition which allows high resolution imaging on 7 T systems without running into SAR limits. These features shall enable new insights to be gained into Parkinson’s [12] and Alzheimer’s diseases [13–15], where associated iron accumulations can be mapped with high spatial resolution, and in Multiple Sclerosis, in which lesion-related venous vessels can be visualized. Also, imaging micro bleeds (trauma) or monitoring the neurovascular system of a growing, malignant tumor is feasible. SWI images are sometimes displayed using the minimum intensity projection (mIP) of several slices, which is then called HRBV (high-resolution BOLD venography) [16,17]. To date, SWI has been used in MR venography [16], arterial venous malformations [18], occult venous disease [17], multiple sclerosis [19], tumors [20,21], and brain iron mapping [22–24].

Recently, high- and ultra-high-field whole-body MR imaging systems have played an increasingly important role in musculoskeletal imaging. However, ultra-high field imaging systems are currently only used in research centers. In clinical settings, 1.5 T scanners have been the gold standard for musculoskeletal MR imaging, but are now increasingly being replaced by 3 T systems. The major advantage of MR imaging systems at 3 T and above is the increased capability to perform high resolution morphological and especially biochemical imaging.

At 1.5 T, most musculoskeletal tissues (e.g., cartilage, muscle, menisci, bone, tendon, ligaments) are rendered with quite a low signal-to-noise ratio (SNR). Modern 3 T systems offer not only an increase in field strength, but also have advanced radiofrequency coil designs, and new imaging sequences. These yield increased SNR, which can be translated into either higher temporal or spatial resolution for improved morphological imaging. The field strength dependent SNR gain is outweighed by the frequency dependency of relaxation times, and there is the need to adjust sequence parameters (e.g., bandwidth, RF pulse parameters) to reduce artifacts, and not exceed SAR limits.

Imaging the knee: MRI at 1.5 T already provides very high image quality for knee imaging. Eckstein et al., however, evaluated the precision of assessment of cartilage volume, thickness, and surface area of the femorotibial cartilage at 1.5 T and 3 T, and found that results tend to be more reproducible at 3 T than at 1.5 T [25].

Chang et al. performed the first in vivo 7 T MRI study of the knee. They investigated the trabecular bone microarchitecture, and detected activity-related changes in Olympic fencers compared to healthy controls [26]. It has been shown that dual X-ray measurement of bone density, as currently used for the diagnostics of osteoporosis, is not sufficient to characterize bone quality, because, apart from the bone density, the trabecular bone architecture contributes significantly to the mechanical strength, and thus, fracture risk [27,28]. Such bone structure is routinely characterized by microscopic computed tomography [29,30]. However, in vivo trabecular bone imaging is one of the emerging applications for high-resolution morphological MRI; this application—due to the increased sensitivity to susceptibility and the resulting shortened T2* relaxation times—truly benefits from moving to a higher field strength [31,32].

For cartilage imaging, up to a 2.4-fold increase in SNR has been reported in fully balanced steady-state, free-precession imaging comparing 3 T and 7 T. One study detected an increase in SNR from 3 T to 7 T that was significant only for the gradient-echo, but not for the fast spin-echo sequences [33]. In that comparison, however, scan parameters for the fast spin-echo sequence had to be modified due to SAR limits, which resulted in incomplete coverage of the knee joint, extensive artifacts, and a less effective fat saturation. Contrast-to-noise ratio (CNR) and image quality were increased for the gradient-echo and decreased for the fast spin-echo sequences. When comparing 3 T and 7 T, the level of confidence for diagnosing cartilage lesions was higher in the gradient-echo and lower in the fast spin-echo sequences; however, overall, the fast spin-echo sequences at 3 T had the highest confidence score. Evaluation of bone marrow edema was decreased at 7 T due to the limited performance of the fast spin-echo sequences.

Kraff et al. have performed a comparison between knee imaging at 1.5 T and 7 T and also reported that, due to the SAR limitations with the fast spin-echo sequences, subsequent measurements were needed for complete coverage of the knee joint [34]. They also confirmed that bone marrow edema was better visualized at 1.5 T than at 7 T. To date, no study has verified a statistically significant increase in diagnostic accuracy for the detection of common pathologies in the knee when moving to MR scanners with field strengths above 3 T.

Imaging the ankle: Bauer et al. performed trabecular bone imaging of the calcaneus and concluded that 3 T results were more accurate than those calculated from the 1.5 T [35]. Notably, the advances in visualizing trabecular bone structure were partially SNR-independent. They assumed that the better performance at 3 T might have been due to increased sensitivity to susceptibility, enhancing the visualization of thin trabecular structures. The same group published data indicating that ankle abnormalities are better visualised at 3 T than at 1.5 T, and applying parallel imaging reduced the scan time by 56% [36].

Another group confirmed that image quality for imaging the ankle was significantly higher at 3 T than at 1.5 T [37]. In two of four readers, the sensitivity in detecting cartilaginous pathology at the ankle joint was enhanced significantly by imaging at 3 T compared to 1.5 T, using a fast spin-echo sequence. Differences between 1.5 T and 3 T when balanced Steady-State Free Precession (bSSFP) and the Spoiled Gradient Recalled (SPGR) sequences were used were insignificant, based on ratings by all the radiologists in that study. The sensitivities for detecting tendon pathologies were significantly higher at 3 T as opposed to 1.5 T for two of the radiologists in the same study. Specificities were similar for all readers at 1.5 T and 3.0 T.

Imaging the ankle at 7 T is mainly limited by the lack of commercially available, dedicated ankle coils. Thus, one group used an 8-channel head array coil for performing trabecular bone imaging at 7 T, and found that the relative noise enhancement factor was lower at 7 T than at 3 T for parallel imaging acceleration factors higher than three [38,39].

Imaging the wrist: The wrist is an especially challenging region because of the small size of the relevant anatomic structures, such as the intrinsic and extrinsic ligaments, the articular cartilage, and the Triangular Fibro Cartilage Complex (TFCC). Sensitivity, specificity, and accuracy for imaging the TFCC pathology are higher at 3 T than 1.5 T [40]. In addition, trabecular bone imaging has already been performed at the wrist using 3 T MR scanners, which resulted in a 16-fold increase in SNR as compared to 1.5 T and a maximum spatial resolution of 0.20 × 0.20 × 2.00 mm³ [41].

Farooki et al. were the first to image the wrist above 3 T; they qualitatively compared 1.5 T with 8 T and concluded that 8 T was superior for the visualization of the infrastructure of the median nerve and in the definition of the boundaries of the carpal tunnel [42]. Other authors went on to quantify the difference between 1.5 T and 7 T MRI of the wrist and found that SNR was about five times higher in tendon, bone, muscle, and nerve [43]. These authors achieved a maximum spatial resolution of 0.16 × 0.16 × 1.50 mm³.

The New York University research group uses an in-house built 8-channel receive and 4-channel transmit RF-coil for wrist imaging, which provides the possibility to perform parallel imaging; the maximum spatial resolution they achieved in vivo was 0.08 × 0.08 × 2.00 mm³ [44].

In summary, morphological MR imaging at 3 T has already been proven to be superior to 1.5 T in many brain and musculoskeletal imaging applications and is widely used in the clinical routine; nevertheless, imaging at 3 T continues to improve. At 7 T, initial studies have demonstrated the great potential of those systems for ultra-high resolution imaging, but several technical challenges remain to be solved. In the future, we hope to be able to fully realize the benefits of high- and ultra-high-field imaging by combining ultra-high resolution morphological imaging with functional imaging techniques, not only in the brain but also in the small joints of the human body.

During the second half of the last century biology, in particular brain research, became increasingly important in the field of psychology. A new scientific discipline evolved—Cognitive Neuroscience. The rapid growth of knowledge in this discipline has been tellingly documented already by four books entitled “The Cognitive Neurosciences”, edited by Michael Gazzaniga between 1995 and 2009. Two dominant approaches, pursuing ultimately the same aim, are manifest in these works: the first begins with sub-cellular and cellular processes in order to explain functions of small and finally large scale neuronal networks, while the other observes activities of large scale networks during experimentally defined behavioural and cognitive challenges, subdividing them into smaller networks to describe their interconnectivity and interaction. Both utilize the knowledge gained by the other approach.

The latter strategy has been driven by the development of powerful imaging techniques, in particular MRI, fMRI, MRS and PET. These ‘sensor-systems’ have added a new dimension to the data as gathered previously with more simple methods like electrophysiology, providing highly resolved localization of distributed processes. This feature initiated not only the study of whole brain sensory systems but also widely distributed processing systems, that is neuronal networks that mediate complex mental processes such as language, spatial ability, emotion, empathy, etc.

The most widely used contrast mechanism in the mapping of brain activation is the so-called BOLD (blood-oxygen level dependent) contrast, which is based on blood oxygenation changes in the venous vessels and capillaries following electric activity of nerve cells and related metabolic changes [45]. In BOLD-based functional MRI, ultra-fast echo-planar imaging (EPI) is the workhorse for functional studies. Time-series of rapid whole- or part-brain imaging slabs are recorded during rest and/or task performance, technically limited by temporal and spatial resolution, SNR and physiological artifacts. Major artifacts result from intra-voxel signal dephasing [46–48], voxel blurring due to broadened point-spread function (PSF), physiological artifacts due to breathing and heart-beat, and gross head motion. The consequences of these artifacts and the correction schemes which may be applied are different in different brain areas as well as being dependent of magnetic field strength and scanner performance, leading to the need for specifically optimized measurement protocols (e.g., Robinson, Moser and Peper, in [5]). Cortical regions, due to better magnetic homogeneity than subcortial or limbic areas, are typically scanned with 27–64 mm³ nominal resolution, resulting in a temporal resolution (TR) of about 3 s for a 20-slice slab. In studies of emotion, for example, high-resolution EPI with ≤10 mm³ (3 T) or 1 mm³ (7 T) voxels should be used, however, this may reduce temporal resolution to >5 s or brain coverage (also depending on the gradient system employed). In studies where high temporal resolution is required (TR ≥ 100 ms), Brain Activity Movies, or BAM’s [49], may be generated. Such rapid acquisition reduces the SNR per unit time, however, as well as the brain coverage that may be achieved. Functional connectivity data may be collected during rest/non-activity using EPI time series [50]. Spontaneous BOLD-signal fluctuations may indicate resting-state networks in the brain [51–53], and also pathological disturbance [54]. Image preprocessing is necessary to reduce artifacts (e.g., Weissenbacher et al. [55]) and statistical mapping to either create brain activity maps or visualize white matter bundles. Finally, software to co-register and map functional onto anatomical data may be used to help visualize the wealth of information appropriately.

Research on unconscious visual processes and their capacity to influence human behaviour has a long and controversial history within Psychology (see Kouider et al. [56] for a review). An influential view in the Cognitive Neurosciences proposes that such unaware processing of stimuli can occur in sub-cortical structures without any significant involvement of the cortex, while these processes may in turn modulate cortical activity [57]. Other models argue for an essential role of long-distance networks linking, e.g., the higher sensory areas and the prefrontal and parietal cortex [58], or link awareness to early activations within primary sensory areas or functionally specialized brain areas such as the fusiform face area (FFA) [59].

Experimentally, a reduction of conscious awareness can be achieved by reducing the stimulus presentation time below the individual visual sensory threshold. Such an approach enables the investigation of the modulation of brain activity induced by truly sub- (unconscious) and superliminally (conscious) presented neutral stimuli.

Although modern high-field MR scanners offer TRs down to 100 msec, high temporal resolution comes at the cost of reduced spatial resolution and/or brain coverage. This fact initiated efforts to co-record fMRI and the electroencephalogram (EEG) in parallel; the more so because electroencephalography has also evolved into a functional brain imaging technique. Algorithms have been developed and tuned to estimate sources of brain activity based on scalp potential distributions captured via multi-channel EEG recording, however, with the restriction to activities within the cerebral cortex [60,61]. Additionally, EEG is certainly closer linked to the neuronal activity since it is composed of a sum of neuronal field potentials (albeit a complex one), whereas the BOLD signal indirectly mimics neuronal activity via blood oxygen consumption showing an individual but constant delay of 3 s to 8 s in the various brain regions. Parallel recording of fMRI and EEG, therefore, enables short latency imaging with the aid of the fMRI’s high-resolution localization feature. This way both the temporal and spatial course of brain activity can be traced.

Since the first publication of the simultaneous recording of electroencephalogram (EEG) and BOLD contrasted MRI in 1993 [62] great efforts have been made to turn this powerful tool into a standardized technique. However, the technical problems arising from the integration of these two techniques are manifold in terms of developing suitable amplifiers and procedures to correct for MR related artifacts [63]. While MR-compatible EEG amplifiers are becoming increasingly available, elimination of scanner-induced artifacts in the EEG recordings is still challenging. Basically, these artifacts can be divided into two groups: (1) artifacts related to the cardiac pulse, which causes movements of the electrodes within the static magnetic field (B₀) of the MR-scanner [64] and (2) artefacts produced by the fast switching of the B₀-gradients [65]. For event-related potentials (ERP), the efficacy of several software algorithms in the compensation of these artifacts has been shown repeatedly (for a review see Ritter & Villringer [66]) allowing the method to become an essential tool in the investigation of neuronal correlates of cognitive functions [63].

Slow cortical potentials (SCP), however, form a special class of ERPs that accompany temporally extended (up to 10–20 s) sensory/cognitive processes. Unlike the regular ERPs, SCPs neither show specific peaks nor a well-structured temporal course but are comprised of spatially and temporally extended waveforms in the frequency range far below 1 Hz—therefore also referred to as direct-current EEG (DC-EEG). While the neural origin of these potential changes is the same as that of ERPs, they are also tightly related to post-synaptic potentials (PSPs), but possibly not only generated by a superposition of many short-living PSPs, but also of long-lasting PSPs. Furthermore, contributions of astroglia-depolarizations possibly induced by neuron-glia communication [67,68] have been shown. Although a complete and coherent view of the neural basis of SCPs is missing, work on animal models has shown a tight correlation between the fMRI BOLD signal and local field potentials measured extracellularly within the cortex (for a review see Logothetis [69]). In humans, this relationship has been shown by Lamm et al. [70] and Khader et al. [71], using spatial imagery tasks, and by He et al. [72] studying spontaneous fluctuations of the BOLD signal.

The studies cited above suggest a tight relationship between SCP-variations and the BOLD response. However, these studies did not record both signals simultaneously. fMRI and EEG data were acquired separately since: (1) MR-compatible EEG amplifiers capable of recording DC-EEG were not available until recently, and (2), because existing scanner-artifact reduction methods would fail for SCPs since they would suppress the DC component of the EEG-signal. In the Results section we present an approach where NeuroConn Inc. (Ilmenau, Germany), in close collaboration with the University of Vienna and the Medical University of Vienna, developed an adequate artifact compensation procedure for simultaneous SCP and fMRI data acquisition.

Articular cartilage consists mainly of an extracellular matrix made of type II collagen, proteoglycan, chondrocytes, and water [73]. Proteoglycan (PG) is made of a linear protein core to which many glycoproteins known as glycosaminoglycans (GAG) [74] are attached. Proteoglycan serves to cross-link the collagen fibrils in the extracellular matrix to provide both compressive and tensile strength to the matrix. In addition, the proteoglycan links restrict fluid flow through the matrix and thereby increase resistance to structural deformation.

The sulfate and carboxyl groups of the glycosaminoglycans impart a negative fixed charge density (FCD) to the matrix. These negative ions attract positive counter-ions (sodium) and water molecules and provide a strong electrostatic repulsive force between the proteoglycans. These osmotic and electrostatic forces are responsible for the swelling pressure of cartilage. The configuration of the PG macromolecules also contributes to the resistance of the matrix to the passage of water molecules and hence affects the mechanics of the cartilage in this fashion.

The onset of osteoarthritis (OA) is understood to be associated primarily with biochemical phenomena. The ability to quantify these molecular changes using novel sensors will provide a tool for the early diagnosis of osteoarthritis and treatment monitoring. The loss of proteoglycan from the extracellular matrix has been hypothesized to be the event that initiates the onset of OA [75,76].

The loss of proteoglycan with the onset of OA results in a reduction of FCD in cartilage. Maroudas et al. have shown that the fixed charge density (FCD) of cartilage is correlated to the GAG content of cartilage [77]. Since the FCD is counter-balanced by the Na ions, loss of PG (hence GAG and FCD) due to cartilage degeneration results in the loss of sodium ions from the tissue. The loss of the negatively charged PG lowers the FCD in the tissue, thereby releasing positively charged sodium ions.

Sodium (²³Na) MR imaging has been validated as a quantitative method of computing FCD and, hence, proteoglycan content, in healthy humans [78–80]. Healthy human cartilage FCD ranges from a concentration of 50 mM to 250 mM, depending on the age and location in the tissue [81]. As demonstrated in controlled cartilage degradation experiments [78,82], the sensitivity of sodium MR imaging is adequate for detecting small changes in proteoglycan content on the order of 5%. Sodium MRI experiments were performed on the knee cartilage of healthy as well as early stage OA patients at 4 T and demonstrated the feasibility of sodium MRI in computing PG loss in early stage OA. [79]

The major advantage of ²³Na MRI, especially of cartilage, is that it is highly specific to PG content and, since the sodium from surrounding structures in the joint is low (<50 mM), cartilage can be visualized with very high contrast without the requirement for any exogenous contrast agent such as that in dGEMRIC [83]. It can be used to quantify early molecular changes in osteoarthritis.

The disadvantage of sodium MRI is that it requires field strengths of >3 T to obtain quality sodium images that enable accurate quantification of cartilage FCD. Furthermore, due to the limitations of gradient strengths and other hardware requirements, most of the sodium imaging experiments reviewed here employed echo times of >2 ms. Since the fast T₂ decay of cartilage lies in the range 1–2 ms, substantial signal is lost before the acquisition. This is a major contributor to the low SNR of sodium compared with conventional proton MRI. Additionally, the sodium gyromagnetic ratio (γ) is one-quarter that of protons, meaning that sodium MRI requires four times stronger gradients to obtain images with the same resolution to that of proton MRI. Resolution, however, is not only a function of the gyromagnetic ratio but also of the receiver bandwidth. The MR sensitivity for ²³Na is only 9.2% of that of protons (¹H), and the in vivo concentration is ∼360 times lower than the in vivo water proton concentration. The combination of these factors results in a ²³Na signal which is approximately 4000 times smaller than the ¹H signal.

Multi-nuclear magnetic resonance spectroscopy (MRS) is a non-invasive tool for investigating metabolism in vivo. Its applications range from investigation of normal physiology in healthy humans and studying metabolic changes under the influence of ageing, exercise, nutrition or medication [84–87] to the diagnosis of a wide range of diseases which include, but are not limited to, metabolic and neurological disorders (e.g., diabetes, myopathies, Parkinson’s disease, epilepsy), psychiatric disorders, and many forms of cancer (e.g., in the brain, breast or prostate) [88–90]. In vivo magnetic resonance is most commonly associated with its use as an imaging modality for clinical diagnostics (MRI, see previous sections).

MRS is based on the same physical principle as MRI and can be performed on the same equipment; however an additional component of information is obtained from the tissue under investigation: the chemical environment, i.e., which molecule or functional group the observed nuclei is surrounded by, can be distinguished via a shift in resonance frequency caused by the molecules’ electron configuration, which leads to shielding of the magnetic field. Additionally, spatial information can be retained to various extents, from no specific localization information, beyond the radio frequency (RF) coil’s position for signal detection, through one dimensional and single voxel localization up to full 3D chemical shift imaging, capable of generating a map of spectra for each point on a three dimensional grid. Of course, the choice of methods is governed by a trade-off between sensitivity, temporal and spatial resolution (both influencing specificity) and total measurement time, limited by motion artifacts, patient compliance and potentially also availability of the scanner. Using ¹H MRS it is possible to quantify total creatine, trimethylammonium (TMA) compounds, intra- and extramyocellular lipids (IMCL/EMCL) in muscle, involved in the pathogenesis of insulin resistance and type-2 diabetes [91,92]. The metabolites N-acetylaspartate (NAA), creatine, choline, inositol, glutamine, glutamate, γ-aminobutyric acid (GABA), lactate and others are accessible when ¹H MRS is applied to brain [93–97], rendering the method valuable in the study of physiological metabolism and pathologies, as the metabolites’ concentrations, typically abundant in the millimolar range, change [90–100].

MRS is not limited to the hydrogen nucleus. Other physiologically relevant nuclei may be measured by transmitting the RF pulses with a frequency specific to the particular nucleus. This requires dedicated antennas (RF coils) specifically designed to match the resonance frequency of the nucleus and the given magnetic field strength B₀. This offers a wider view on the metabolism and state of tissue under investigation, as either a different set of metabolites can be observed or molecular properties (e.g., binding to cell walls, orientation effects) are exploited to obtain specific information. Multi-nuclear MRS is most frequently employed using the phosphorus (³¹P) or carbon (¹³C) nuclei. Phosphorus is of physiological interest as high-energy phosphates play an important role in metabolic turnover and can be detected directly [84]. In particular, adenosine triphosphate (ATP) turnover and proton handling can be studied by ³¹P MRS of muscle [101,102], rate constants of changes in phospho-creatine (PCr) concentration during exercise [103] and, especially, recovery from exercise [101,104] can teach us about mitochondrial function and oxidative ATP re-synthesis [105,106]. As has been shown recently, a combination of different multi-nuclear MRS methods can further increase their value beyond the sum of their respective contributions by adding complementary information to a view on metabolism not accessible to each of the methods alone [85,107,108].

Magnetic resonance at ultra-high field strength has many advantages, particularly for MR spectroscopy [1,2,109]: firstly, the signal intensity, which is often the key factor for the feasibility of a spectroscopy study, increases with field strength and secondly, the chemical shift dispersion, i.e., separation of the signal components which are attributed to the different metabolites to be quantified, obeys a linear relationship with field strength. This stronger separation makes the identification of resonances easier, more reliable or even possible at all, as signals of individual metabolites may not be adequately resolved at lower field strength. Additionally, the attraction of ultra-high field MRS is enhanced by the reduction of T₁ relaxation times for certain metabolites (e.g., phospho-creatine) [110,111], allowing shorter repetition times and thus a further increase in signal-to-noise ratio (SNR) by averaging measurements, increasing time resolution or reducing total measurement time.

To fully exploit the advantages of increased magnetic field strength some technical challenges inherent to the MR physics in this regime need to be addressed:

Higher demands on RF power, due to the higher resonance frequency, leads to increased specific absorption rate (SAR).

Examples for high-resolution structural MRI of the brain are given in Figure 1. These images where acquired on a healthy volunteer and demonstrate T₁ and T₂ weighted contrast. Note the strong contrast between gray and white matter in the T₁ weighted image as well as the excellent contrast between CSF and brain tissue in the T₂ weighted image.

An example of susceptibility-weighted contrast in the human brain at 7 T is given in Figure 2. Note the excellent definition of veins, due to deoxygenated blood, as well as the good delineation of the basal ganglia, due to higher iron content.

Examples for high-resolution structural MRI of the knee and cartilage are given in Figure 3. A dedicated knee-coil was used to visualize details in cartilage and menisci. In addition to structural proton imaging, we have recently applied sodium imaging in patients after matrix-associated autologous cartilage transplantation (MACT). The resolution achieved was sufficient to visualize even the thin cartilage layer adjacent to the proximal tibio-fibular joint (Figure 4). In our patient group sodium imaging allowed to differentiate between different concentrations of sodium and hence GAG in different portions within the transplants which underscores its strength in the overall evaluation of cartilage transplants in contrast to the limited biopsy sample evaluation. We found a good correlation between sodium imaging and dGEMRIC in the quantification of the GAG concentration in patients after MACT [112].

In this particular study healthy subjects were examined while performing a forced choice detection task. The subjects’ task was to identify black/white target faces, which were sequentially embedded within two pictures of equiluminiscent white noise. Targets were randomly chosen images from six equiluminiscent faces of men and women. Following each presentation, subjects were requested to rate if they were able to identify the faces’ gender, saw a face but could not rate the gender, or had not seen any presentation at all. Presentation durations were adjusted to three subliminal and three superliminal values according to the subjects’ individual visual sensory thresholds, and were presented 12 times each in randomized order. Individual thresholds were determined before fMRI scanning at 7 T. For each subject the percentage of correct detections per duration was calculated and finally a logistic curve fitted across all durations. Individual thresholds were defined as that duration which resulted in 50 percent correct recognitions. Exact timing of the presentations was provided by a custom build LC-shutter tachistoscope [113]. For single subject analyses statistical parametric maps were calculated using the general linear model with both regressors corresponding to the six presentation durations as well as sigmoidal regressors corresponding to the probability with which the faces’ gender was detected. This approach enabled on the one hand identification of those neuronal regions that were activated by the different presentation durations (Figure 5) and on the other hand identification of those brain areas that were involved in subjective perception (Figure 6). As expected, neuronal areas related to the different presentation durations were the thalamus and the primary visual cortex and areas that followed the perception curve were within the prefrontal and the cingulate cortex, additionally to the predominantly sensory areas.

Understanding the neuronal correlates of consciousness is one of the most challenging problems within the Cognitive Neurosciences. As mentioned above, probing stimuli around their sensory/perceptual thresholds is one way to identify brain structures that are essential for conscious experience. However, responses to stimuli presented for these short durations are weak and so is their BOLD response. In such applications the high sensitivity of a 7 T MR scanner is quite advantageous as can be seen from preliminary results presented here.

The brain areas identified with the two analysis approaches employed are in line with the so-called global neuronal workspace model [58]. Central to this theory is the hypothesis that two clearly distinguishable hierarchy levels can be described within the functional and anatomical architecture of the brain. The first level includes cortical and subcortical areas responsible for specific tasks within a certain sensory modality. According to the Dehaene model these areas are activated automatically and therefore work mostly unconscious. The primary visual cortex and the thalamus may be seen as part of this level since their activity linearly increased with presentation duration. The second level consists of prefrontal, cingulate, and parietal cortices and is characterised by many widespread connections to other areas within the brain. This high interconnectivity allows for fast and flexible exchange of information between the different sensory modules, a prerequisite for awareness. Additional activities within the prefrontal and cingulate cortex, as shown by the second analysis, confirms this view.

In order to demonstrate the feasibility of simultaneous EEG and fMRI data acquisition a simple experiment using visual stimulation will be presented. As stimulus a slowly rotating checkerboard-propeller was presented in the center of the visual field for 8 s. Stimuli were repeated 40 times with a fixed 7 s black screen inter-stimulus interval in two runs, with and without parallel fMRI acquisition. EEG was recorded via 61 silver/silver chloride scalp electrodes using a NEURO PRAX® MR full-band EEG amplifier (NeuroConn Inc., Ilmenau, Germany). For offline eye movement artifact correction vertical and horizontal electrooculograms were recorded bipolarly from above and below the left eye and from the left and right outer canthi. The subject’s skin was slightly scratched at all recording sites in order to minimize skin potential artifacts and to ascertain homogeneous electrode impedance. The electrodes were filled with degassed electrode gel (Electro-Gel, ElectrodeCap International Inc., Eaton/OH, USA) to prevent ultra-slow baseline shifts [114]. Gradient and pulse artifacts were suppressed on-line using an adaptive template-matching algorithm. To this end up to 100 channel-specific templates were generated in a previous acquisition run [115]. Thereafter eye movement artefacts were eliminated using a linear regression approach with channel-specific correction coefficients [116]. Finally, single trial signals were low-pass filtered at 8 Hz and averaged. In order to identify the cortical structures involved, the resulting SCPs were subjected to current density analyses using sLoreta [61].

MR-images were acquired on a 3 Tesla TIM Trio system. For single subject analysis statistical parametric maps were calculated using the general linear model with regressors corresponding to the visual stimulation onsets. Activity in occipital areas was found in the SCP recordings of both acquisition runs (see Figure 7). A clear negative going SCP can easily be identified at occipital recording sites, which is slightly smaller during fMRI acquisition. As expected this waveform is restricted to sensory visual areas, i.e., to parietal and occipital recording sites, which can also be seen in the corresponding topographic potential pattern (color coded in blue). SCP based source localization (Figure 8) revealed strong activation in and around the primary visual cortex. The corresponding fMRI results showed activation primarily in the middle occipital gyrus (Figure 9).

The experiment presented resulted in SCP waveforms, topographies, and source localizations, comparable to data acquired outside the MR scanner. These results clearly demonstrate that simultaneous acquisition of DC-EEG and fMRI data, i.e., data acquired by means of different sensor systems within one experiment covering different aspects of brain activity, is feasible at high magnetic fields.

For an illustration of ultra-high field MRS applied to human skeletal muscle see Figure 10. Localized ¹H spectra were acquired from human soleus muscle, at 3 and 7 Tesla, using a STEAM (“stimulated echo acquisition mode”) singel voxel localization scheme with an echo time T_E = 20 ms and a voxel size of approx. 2 mL, in MR systems from the same manufacturer (Tim Trio and Magnetom 7 T, both by Siemens Medical Solutions, Erlangen, Germany). Parameters at 3 Tesla were T_E = 20 ms, T_M = 30 ms, T_R = 4 s, VOI = 12 × 12 × 12 mm, 32 averages. At 7 Tesla, measurement parameters were T_E = 20 ms, T_M = 10 ms, T_R = 1.5 s, VOI = 12 × 12 × 20 mm, 32 averages. Proton (¹H) MR spectra of skeletal muscle allow quantification of e.g., total creatine, trimethylammonium compounds and lipids. Due to a bulk magnetic susceptibility effect, it is possible to distinguish between intra- and extracellular lipids, which becomes more pronounced at 7 Tesla. Several further advantages of ultra-high field strength, as compared to spectra acquired at 3 Tesla, become evident from the example given in Figure 10. As chemical shift dispersion increases linearly with B₀, the separation of resonances improves (e.g., between TMA and creatine), provided that macroscopic field inhomogeneities can be compensated using high order shim routines. Signal-to-noise ratio is increased at higher field, improving reliability of quantification, for example the signal attributed to the CH₂ group of creatine (Cr2) is clearly visible at 7 T.

Another example of ultra-high field MRS is shown in Figure 11. Dynamic localised ³¹P spectra of skeletal muscle were acquired on a 3 T whole body scanner (Bruker Biospin, Ettlingen, Germany) and a 7 T system (Siemens Medical Solutions, Erlangen, Germany). Signal gain at 7 Tesla—further improved here by using an advanced acquisition method which was recently implemented on the Siemens platform (³¹P semi-LASER, “localization by adiabatic selective refocusing” [104,117])—was utilized to achieve higher temporal resolution by avoiding averaging of consecutively acquired spectra. In exercising muscle, phosphocreatine is depleted under concomitant accumulation of inorganic phosphate. During recovery, PCr is resynthetised via oxidative phosphorylation, with a time constant typically on the order of minutes. These processes can be followed by dynamic ³¹P MRS. Measurements at 3 Tesla were performed with STEAM (T_E = 7.5 ms, T_M = 30 ms, T_R = 7.6 s, 1024 complex data points per acquisition). At 7 Tesla, semi-LASER was used (T_E = 53 ms, T_R = 8 s, 2048 complex data points). The quality of 7 T data shown for each time point surpasses the quality of measurements at 3 T, so that no averaging was necessary at all for quantifying PCr in the 7 T spectra using a time domain fit routine for MR spectra. Reliable quantification of phosphocreatine (PCr) and inorganic phosphate [87] time courses during and after aerobic exercise of human muscle [84], from localized ³¹P spectra is thus possible with higher temporal resolution at 7 T. This allows determination of PCr resynthesis kinetics, a measure for oxidative capacity of skeletal muscle, with improved specificity.

As a diagnostic technique based on non-ionizing radiation, magnetic resonance imaging and spectroscopy can be considered a low risk procedure, as long as several safety-relevant parameters are monitored and kept within certain limits. Three main potential sources of hazards are connected with the use of magnetic resonance systems: (1) The radio frequency (RF) field, B₁, required to tilt the magnetization vector from its equilibrium value and thereby induce a signal which can be measured, causes energy deposition in the tissue, (2) rapidly switched B₀ field gradients, mainly applied for the purpose of encoding spatial information the MR signal, may lead to the induction of voltages in tissue, potentially causing nerve stimulation, and finally (3) the static magnetic field B₀ poses a source of risks, predominantly due to the strong attraction forces on ferromagnetic objects. The potentially hazardous effects connected with in vivo magnetic resonance are subject to legal limits and recommendations issued by national and international organizations. Particularly, the United States Food and Drug Administration (FDA), the International Commission on non-ionizing Radiation Protection (ICNIRP), and the International Electrotechnical Commission (IEC) publish standards and guidelines defining limits of applicable RF-energy, the so-called specific absorption rate (SAR), magnetic gradient amplitudes and switching times and B₀ field strength. Regulations limit the specific power absorption and/or resulting temperature increase [129] and define distinct thresholds for acceptable deposition, depending on the parts of the human body investigated. Limits distinguish between total average and average per gram of tissue in the head, abdomen and extremities, as well as averages on various time scales (typically over 5 or 6 min or short term averages of several seconds). It can be shown that the deposited power is proportional to the square of the static magnetic field B₀² and to the square of the irradiating RF field B₁² [130,131]. The calculation of the deposited energy is complex [122] due to the geometrical distribution pattern of B₁ in biological tissue, particularly at high field strengths, where the wavelength of the RF wave in tissue is on the order of (or smaller than) the dimensions of the organs.

MP-RAGE parameters at 3 T: TR/TI/TE = 2,300/900/3.59 ms, image-matrix = 320 × 320, 208 sagittal slices, GRAPPA factor = 2, resulting in an isotropic resolution of 0.8 mm, and scan time: 12.18 minutes.,

MP-RAGE parameters at 7 T: TR/TI/TE = 4,660/1700/4 ms, image-matrix = 320 × 320, 228 sagittal slices, GRAPPA factor = 2, resulting in an isotropic resolution of 0.65 mm, and scan time: 12.50 minutes.

TSE parameters at 3 T: 2D-Turbo Spin Echo (TSE), GRAPPA factor = 2, TR/TE = 13,070/92 ms, MA = 256 × 256, slices = 128, resulting in a resolution of 1 × 1 × 1.2 mm, and scan time of 5.15 minutes.

Advanced cartilage imaging at 7 T: PD_FSE; TR = 2,400 ms, TE = 24 ms, 36 slices, FOV = 120 cm², Matrix = 1024 × 1024, slice thickness: 2 mm, scan time: 6min37sec.

SWI parameters at 3 T: A three-dimensional, fully first-order flow-compensated GE sequence with a TE of 29 ms; TR = 36 ms; image-matrix = 320 × 320 pixel; slices = 104; parallel imaging (GRAPPA) factor = 2; resolution = 0.64 × 0.64 × 1.3 mm.

SWI parameters at 7 T: SWI sequence; TE = 15 ms; repetition time = 28 ms; image-matrix = 704 × 528 pixel; slices = 96; parallel imaging (GRAPPA) factor = 2; scan time = 9.48 min; resolution = 0.3 × 0.3 × 1.2 mm³.

FMRI-EPI parameters at 3 T: single-shot gradient-recalled echo-planar imaging; 12 axial slices (4 mm thickness, 3 mm gap) aligned to the connection line between anterior and posterior commissure; matrix size = 64 × 96; FOV = 210 × 210 mm², and TE/TR = 42/1,000 ms.

FMRI-EPI parameters at 7 T: single-shot gradient-recalled echo-planar imaging; 4 axial slices (5 mm thickness), aligned to include the thalamus and the fusiform gyrus, matrix size = 128 × 128; FOV = 240 × 240 mm², and TE/TR = 26/200 ms.

²³Na-MRI parameters at 7 T: Modified 3D-GRE sequence: TR/TE = 10.0/3.77 ms; FOV = 199 × 199 mm², 48 slices; MA = 64 × 128; resulting in a resolution of 3.11 × 1.55 × 3.0 mm³, and scan time of 30:45 min.