Susceptibility-weighted (SW) images can also be acquired (Haacke and Reichenbach, 2011). These images are based on the principle that MR images are generally complex-valued, i.e., effectively two images are always acquired: a commonly used magnitude image that often directly displays the anatomical structures and a phase image that is usually disregarded. The phase image however is sensitive to the so-called magnetic susceptibility (χ). This property of tissues and substances alters the local magnetic field values. Paramagnetic materials have a positive χ and strengthen the magnetic field, and diamagnetic materials have a negative χ and weaken the magnetic field. Tissues with a susceptibility that differs from their surrounding structures, such as tissues with myelin and iron-containing substances, cause local deviations in the magnetic field inside and outside of the structures. This leads to local phase differences, which can then be extracted from the original phase images. In susceptibility-weighted imaging (SWI), these phase images are combined with the magnitude images, which can result in additional contrast, which particularly enhances the brain's (micro)vessels and the small deep brain structures.

Furthermore, post-processing techniques can be employed, to produce so-called T1, T2, T2^*, or (quantitative) susceptibility maps, which display the quantitative T1, T2, T2^*, or susceptibility values of each voxel in an image respectively. Sometimes R1-, R2- or R2^*-values are computed instead, which are defined as 1/T1, 1/T2, and 1/T2^* respectively.

In addition to structural imaging, diffusion-weighted imaging (DWI), which is directionally sensitive to water diffusion, gives complementary information (Le Bihan, 2003). It can provide information on the location and orientation of neuronal fibers, aiding in visualization of these pathways (tractography) (Mori et al., 1999) or super-resolution track-density imaging (TDI) (Calamante et al., 2010). Furthermore, functional MRI (fMRI) can provide information on localized brain activity (Buxton, 2013). Finally, DWI and fMRI can be used to compute the connectivity between two areas by computing the fiber paths between them (structural connectivity) or the correlation of functional activity (functional connectivity) respectively.

In most DBS centers, the MR images are obtained from 1.5T or 3T MR scanners. However, in specialized neuroimaging centers, the possibilities of scanning at ultra-high field are increasingly being explored (Duyn, 2012). Although the number keeps growing, at present an estimate of 61 human ultra-high field MR scanners has been installed or will be installed in the near future (see Table 2). At ultra-high field alterations of physical properties can influence measurements both positively and negatively. Several issues including field strength dependent changes in relaxation times T1, T2, and T2^*; increased B0 and B1 magnetic field inhomogeneities; and increased risks of tissue heating (Duyn, 2012) make ultra-high field scanning more sensitive to inhomogeneous signal-to-noise ratios (SNR) and contrast-to-noise ratios (CNR), geometric distortions, and movement artifacts. This limits the use of T1-weighted, T2-weighted, and proton density weighted turbo spin echo (TSE) scan protocols that are commonly used in clinics (Hennig et al., 1986). However, the alterations in relaxation times and the increased sensitivity to magnetic susceptibility have stimulated the focus of ultra-high field imaging to shift to susceptibility and T2^*-dependent gradient echo sequences (Haase et al., 2011). Furthermore, the SNR increases close to linearly with field strength, which offers the option to scan with higher spatial resolution (Vaughan et al., 2001) and/or CNR in a shorter time (Duyn, 2012). This makes ultra-high field MRI especially beneficial for detailed imaging of structures with altered magnetic susceptibility, such as the basal ganglia, myelin, and blood, which is also important for ultra-high field high-resolution fMRI imaging. Finally, side effects might occur during movement through the gradients of the strong field. The majority of the subjects have been reported to feel sensations when moving into or out of the bore, which was rated as unpleasant vertigo in 5–20% of the subjects (Glover et al., 2007; Theysohn et al., 2008) and a small number of subjects (approximately 3%) experienced a medium or strong metallic taste (Theysohn et al., 2008).

Since the installation of the first ultra-high field MR scanner, several studies investigated the visualization of deep-brain structures at ultra-high field in vivo (Table 4).

In 1999, the basal ganglia were visualized at ultra-high field (8T) using a two-dimensional (2D) multi-slice GE sequence, where high-resolution (195 × 195 μm in-plane) T2^*-weighted axial images of one volunteer were obtained in 13 min (Table 3-1) (Bourekas et al., 1999). On these images the globus pallidus (GP), SN and red nucleus (RN) appeared as hypointense regions. These findings were later confirmed in sagittally recorded slices with similar acquisition parameters (Table 3-2) (Novak et al., 2001). In 2003, the same group showed that on GE phase images (Table 3-3), within the SN, the SN pars dorsalis and SN pars lateralis had a higher signal intensity than the matrix of the SN, and within the RN, the medullary lamella showed a higher signal intensity than the RN pars oralis (Abduljalil et al., 2003). A few years later, again the SN and RN appeared hypointense on 7T axial, sagittal, and coronal GE images (Table 3-4) (Cho et al., 2008) and in 2010, 7T coronal GE images (Table 3-38) were obtained on which the STN and SN could be well distinguished (Cho et al., 2010).

A more detailed description of the visualization of the basal ganglia at 7T with three different scanning sequences, exploiting T1-weighted, T2-weighted and susceptibility-weighted imaging, was published in 2010 (Table 3-5:8) (Abosch et al., 2010). Using SWI, a clear delineation of the STN and the boundary dividing it from the SN were visualized in both axial and coronal planes (Figure 1A). Also, SWI allowed visualization of varying levels of contrast within the RN and two of the laminae within the GP (lamina pallidi medialis and incompleta), thus also distinguishing between the GPi and the GPe. Within the thalamus, it showed intensity variations corresponding to the locations of the ventral intermediate nucleus (Vim), the anterior and medial boundaries of the pulvinar, and the boundary of the nucleus ventralis caudalis as identified with the Schaltenbrand and Wahren atlas (Schaltenbrand et al., 1977).

In 2011, Eapen et al., imaged several deep-brain structures with two different sequences at 7T: T2- and T2^*-weighted gradient and spin-echo (GRASE) and T2^*-weighted GE (Table 3-9:10) (Eapen et al., 2011). Both GRASE and GE scans showed a clear distinction between the densely and the poorly vascularized regions of the RN, but only the GE scan also showed signal intensity differences within the SN, possibly representing the SN pars compacta and SN pars reticulata. In two later studies, susceptibility maps were investigated. Using a multi-echo GE sequence (Table 3-11), a boundary between the STN and the SN was shown (Schafer et al., 2012). The use of susceptibility maps generated from three single-echo GE phase data sets with different head positions (Table 3-12) also facilitated detailed visualization of structures (Deistung et al., 2013b). It provided discrimination between the subnuclei within the SN, and allowed for accurate discrimination of the STN from the SN and surrounding gray matter and white matter. Furthermore, within the GP, these maps showed the lamina pallidi medialis and lamina pallidi incompleta (Figure 1B). The RN displayed substructures in the susceptibility maps, facilitating identification of the medullary lamella, and the RN pars oralis and RN pars dorsomedialis showed a significantly increased susceptibility, compared to the RN pars caudalis. Finally, within the thalamus clear intensity variations were observed on these susceptibility maps corresponding to the Vim, pulvinar, lateral and medial geniculate nucleus, dorsomedial nucleus, and dorsal nuclei group as identified with the Schaltenbrand and Wahren atlas (Schaltenbrand et al., 1977) (Figure 1C).

In two other studies by Kerl et al., investigating the STN and ZI with different sequences at 7T (Kerl et al., 2012c, 2013), a distinction between the STN and the SN and ZI and a clear boundary dividing the rostral ZI from the internal capsule, STN and the pallidofugal fibers could be seen on T2^*-weighted images and the latter also on coronal SW images.

Finally, two studies employed DWI properties to identify substructures within the DBS related structures. In one study, DWI (Table 3-14) was used to estimate the pathways between seven regions of interest: caudate nucleus, putamen, GPe, GPi, SN, STN, and thalamus (Lenglet et al., 2012). Seven pathways could be successfully identified: the nigrostriatal, nigropallidal, nigrothalamic, subthalamopallidal, pallidothalamic, striatopallidal, and thalamostriatal pathway. These projections were also used to create subparcellations of the SN, possibly corresponding to the SN pars reticulata and SN pars compacta; subdivisions of the STN into a dorsolateral and ventromedial part; subdivisions of the GPe into medial, lateral and rostro-ventral parts; subdivisions of the GPi into laterocaudal, rostral, and mid portions; and many subdivisions within the thalamus. In another study, 7T DWI (Table 3-15) was used to construct track-density images of the thalamus (Calamante et al., 2012). These showed high-resolution (200 μm isotropic) substructures within the thalamus with clear intensity differences, not only related to track-density, but also to the directionality of the fibers.

When scanning ex vivo, even higher resolution and higher SNR can be obtained due to the possibility of longer scan times and less movement artifacts. Although fixed tissue may suffer from altered tissue properties, such as decreases in T1 and T2 (Tovi and Ericsson, 1992) and a decreased diffusion coefficient (D'Arceuil et al., 2007), which is especially challenging for DWI, it also has great advantages over in vivo MRI. Several studies employed ex-vivo imaging for investigating the deep-brain structures at ultra-high field (Table 4).

In 2007, the STN and its surroundings were explored at 9.4T with a T2-weighted sequence (Table 3-16:18) in a post-mortem brain sample (Rijkers et al., 2007). Acquiring a high in-plane resolution of 100 × 100 μm, not only the most prominent structures of the basal ganglia were visualized, but also the pulvinar, the lateral and medial geniculate bodies, cerebral peduncle, habenulointerpeduncular tract (fasciculus retroflexus), periaquaductal gray, the medial lemniscus, the spinothalamic tract, the mammillothalamic tract, and the superior colliculus.

Three post-mortem brain stems have also been imaged at 7T for 119 min, acquiring 150 × 150 μm images. On these T1-weighted images (Table 3-19), the RN and SN, which displayed heterogeneous signal intensity, could be visualized (Soria et al., 2011). Even higher in-plane resolutions of 44 × 44 and 88 × 88 μm (Table 3-20:21) were achieved in a different study after scanning post-mortem brain samples for 72 and 10 h respectively (Massey et al., 2012). The obtained T2-weighted images facilitated visualization of the STN, SN, RN, ZI, and thalamus but also allowed a highly detailed identification of many smaller structures surrounding the STN. Furthermore, a hypointense signal band was seen between the SN and STN facilitating easy separation of the two structures. Also the anteromedial part of the STN was relatively hypointense compared to the posterolateral portion, which might be related to the subdivision of the STN in a limbic, associative and sensorimotor part. On the 44 × 44 μm resolution images even the fibers of the subthalamic fasciculus were visualized accurately.

Finally, one study that focused on differences in T2 and T2^* values and iron content between post-mortem brains of progressive supranuclear palsy patients and controls, showed high-resolution (50 μm isotropic) fast low-angle shot (FLASH) GE images (Table 3-22), displaying with much detail the SN, RN, putamen, and the GP with a clear separation into the GPe and GPi (Foroutan et al., 2013).

These studies show that ultra-high field MRI can aid substantially in the identification of small (sub)structures including the separation between the STN and SN and the laminae within the GP both ex vivo and in vivo.

In addition to the qualitative description of the visibility of deep-brain structures with ultra-high field MRI, comparisons between different sequences and image reconstruction methods have been made (see Table 5).

In a previously mentioned study from 2003, magnitude, phase-weighted magnitude (SWI), and phase images of a GE dataset (Table 3-3), were compared for their capability to visualize (sub)structures (Abduljalil et al., 2003). On magnitude images the SN and RN showed up hypointense and on phase images, substructures within the SN could be distinguished as well. The combined magnitude and phase images added little extra to the magnitude and phase images separately.

Later, in 2010, the mean susceptibility difference (Δχ) between compartments in an agar phantom, and between white matter and deep-brain structures of healthy subjects were compared among three different susceptibility mapping methods applied to GE FLASH images acquired at 7T (Table 3-23) (Wharton and Bowtell, 2010). The mapping methods consisted of (a) a multi-orientation method using images acquired with differing head positions, (b) a regularized single-orientation method, and (c) a threshold-based single-orientation method. Although all three methods showed large Δχ in the GP, SN, RN, internal capsule, putamen and caudate nucleus, the multi-orientation method resulted in the least noise related artifacts and good estimation of Δχ values in the phantom.

In another 2010 study, T1-weighted, T2-weighted, and SW imaging (Table 3-5:8) were compared (Abosch et al., 2010). Most structures were identified in the SW images (see Table 4), followed by the T2-weighted images (Figure 2). The T1-weighted images showed no obvious structures. Eapen et al., also quantitatively compared their T2 + T2^*- and T2^*-weighted images (Table 3-9:10) (Eapen et al., 2011). No difference between both sequences could be found in the CNR between the SN and ventral tegmental area (VTA) and between the SN and RN, but in the T2 + T2^*-weighted images, the CNR between RN and VTA was significantly better than in the T2^*-weighted images.

In 2012, again differently reconstructed images derived from a multi-echo GE sequence (Table 3-11) were compared, consisting of T2^*-weighted magnitude images, T2^*-maps, and susceptibility maps (Schafer et al., 2012). In most subjects, the CNR between the SN and STN was highest in the susceptibility maps, suggesting that these are most suitable for differentiating the STN from the SN. The SNR of the STN and the rostral part of the ZI (rZI) and the CNR between these structures and white matter, imaged with different sequences, were investigated in two recent studies that compared T1-weighted GE, T2-weighted TSE, T2^*-weighted FLASH and SW images (Table 3-24:29) (Kerl et al., 2012c, 2013). Furthermore, minimum intensity projections (MIPs) of the SW images were computed. After adjusting the SNR and CNR for differences in voxel size, they were highest on the T2^*-weighted images for both structures. Furthermore, the SNRs of both structures on the T2^*-weighted, T1-weighted, SWI-MIP (and for the STN also on the coronal SW images) were significantly higher than those of the T2-weighted images. The CNRs of both structures on the T2^*-weighted and for the rZI also on the SWI-MIP images were also significantly higher than on the T2-weighted images. Also, a 2013 study compared image reconstruction techniques at 7T consisting of (a) magnitude, (b) frequency, and (c) susceptibility maps derived from GE scans (Table 3-12), and (d) R^*₂ maps derived from multi-echo GE scans (Table 3-13) (Deistung et al., 2013b). Qualitative analysis by a neuroanatomist revealed that susceptibility maps in general facilitated the most detailed visualization of structures. Finally, in a recent study by the same group, the CNR between several brain stem structures and their surroundings were compared between sequences (Table 3-30:32) (Deistung et al., 2013a). For the RN and the SN, the CNR of the R2^*-map and the susceptibility map outperformed those of the R1-map and the T2-weighted image.

Although comparison between studies is difficult due to the differences in scanning conditions, the majority of these studies show that sequences that are sensitive to magnetic susceptibility such as SWI and T2^* related images are most suitable for targeting basal ganglia structures and their subdivisions in DBS at ultra-high field.

In addition to comparisons between different sequences, some studies compared similar sequences between different field strengths (see Table 6). In a 2008 study, the difference between a 7T GE image (Table 3-4) and a 1.5 T image was briefly treated (Cho et al., 2008). Visual inspection showed that the 7T image displayed better contrast, SNR and resolution. However, comparison is difficult because the acquisition parameters of the 1.5T image were unfortunately not provided. In the same year, T2^*-weighted GE images were investigated, acquired at several echo times at three different field strengths: 1.5T, 3T, and 7T (Table 3-33:35) (Yao et al., 2009). This showed that increasing field strength resulted in a higher influence of iron on the value of R2^*, making this contrast useful for iron-rich deep-brain structures, such as the GP, RN, SN, and putamen (Hallgren and Sourander, 1958). A thorough quantitative investigation of the visibility of the STN related to field strength was performed in 2010 (Cho et al., 2010), comparing the contrast between the STN and a baseline (containing the ZI and thalamus), the contrast between the STN and SN, the SNR in gray matter areas, and the slope of signal increase between STN and baseline among 1.5T, 3T, and 7T T2^*-weighted GE images (Table 3-36:38). At higher field strengths, the STN, SN, putamen, GPi, and GPe could be visualized while the boundaries of these structures were unclear on the 1.5T images (Figure 3). Furthermore, all quantitative measures increased with field strength, and the SNR and contrast were significantly improved at 7T compared to 1.5 and 3T. Finally, the two studies by Kerl et al., investigating the STN and rZI at 7T (Kerl et al., 2012c, 2013) were additionally performed at 3T. Again, they compared the SNR and CNR of these structures between different sequences: T1-weighted MPRAGE, T2-weighted fluid attenuated inversion recovery (FLAIR), T2-weighted sampling perfection with application of optimized contrasts using different flip angle evolutions (SPACE), two T2^*-weighted 2D FLASH (FLASH2D) sequences, and SW images and their MIPs (Table 3-39:46) (Kerl et al., 2012a,b). This makes it possible to compare the SNRs and CNRs of the different studies between field strengths, when adjusted for voxel size, although it should be noted that for the T1- and T2-weighted images different sequences were used between field strengths. For both structures, the SNRs of the T2^*-weighted, SWI-MIP and SW images of the 7T images were higher than those of the 3T images, but the SNRs of the 3T T2-weighted SPACE image and T1-weighted images were higher at 3T than at 7T. However, the CNRs of both structures were substantially higher on all the 7T sequences than on the corresponding 3T sequences.

These studies suggest that 7T MRI can better facilitate accurate targeting of deep brain structures than 1.5T or 3T MRI.

The improved visualization of the basal ganglia with ultra-high field MRI discussed here provides good perspectives for clinical practice. The clear delineation of DBS target structures and their possible subdivisions may aid in more accurate targeting, which may reduce negative side effects and shorten surgery duration, or it may even allow surgery under general anesthesia. Furthermore, ultra-high field MRI also shows potential for more accurate diagnosis and monitoring of basal ganglia diseases due to, for example, improved identification of the SN pars compacta and SN pars reticulata, which may in its turn facilitate improved patient specific treatments.

In addition, ultra-high field MRI promises to be a versatile tool in clinically oriented research of the deep brain nuclei. It might help us to improve our current understanding of the functionality of the healthy basal ganglia and its disease processes with high resolution functional MRI and connectivity analyses.

When in the end considering the optimal scan protocol for visualizing the DBS targets for clinical purposes at ultra-high field, both image quality and practical requirements need to be taken into account. In terms of hardware, it is recommended to use a head coil with a high number of receive channels (i.e., 16 or higher). This has been shown to improve the SNR (de Zwart et al., 2004; Wiggins et al., 2006) which is also reflected from the studies described in Table 3. In terms of scan protocol, based on the described literature, we recommend to use a 3D multi-echo GE sequence with an isotropic resolution of 0.5 mm³ and partial brain coverage. The 3D sequence facilitates small and isotropic voxel sizes, which ensures good resolution in every plane which is important for distinguishing the STN from the SN. From the multi-echo GE scan, both T2^*-weighted and susceptibility weighted images as well as T2^*-maps, R2^*-maps, and susceptibility maps can be computed, which were shown in the reviewed literature to display best basal ganglia visibility. Since the basal ganglia are located within the same axial oblique slab of approximately 4–5 cm thickness, we advise to shorten scan time by covering only this part of the brain. If more time reduction is required, partial Fourier imaging, elliptical k-space coverage, or parallel imaging can be considered as well.

To support these guidelines, Figure 4 shows an example of a T2^*-weighted image and an R2^*-map created with these recommendations. The images were obtained by scanning a healthy volunteer on a 7T MR scanner (Magnetom 7T, Siemens, Erlangen, Germany) at Scannexus (Maastricht, The Netherlands) using a 32-channel phased-array coil (Nova Medical, Wilmington, United States) with a multi-echo 3D GE sequence. Scan time was reduced to 12 min and 23 s by partial brain coverage and 75% partial Fourier imaging (other scanning parameters can be found in Table 3–line 47:48). On these 0.5 mm³ isotropic resolution images, the STN can be distinguished from the SN in the coronal plane, and the three laminae of the GP can be identified.

When planning a DBS surgery, the MR images are often registered to CT images, resulting in images that both display the stereotactic frame from the CT image as well as contrast within the brain. This registration may be more reliable, however if a whole brain MR image is available as an intermediate step. Abosch et al. (2010) showed that it is already possible to perform 1 mm³ whole brain T1-weighted imaging in 3.5 min, which may be a good candidate for coregistration.

Despite these promising results concerning accurate and high-resolution visualization of the small deep brain (sub)structures, several issues still need to be addressed before they can routinely be employed in direct targeting for DBS.

Firstly, ultra-high field images have an increased risk of geometrical distortions compared to 1.5T images. The severity of these distortions at 7T in deep-brain regions has been investigated in several studies. One study compared the coordinates of marker points in a phantom imaged with 1.5T and 7T MRI to their locations on computed tomography (CT) images (Dammann et al., 2011). The maximum distortion in either x-, y-, or z-direction at 7T was 1.6 mm, which was slightly larger than at 1.5T (0.9 mm). Furthermore, the fewest distortions were observed in the center of the phantom. In another study the distortions in an anthropomorphic phantom between T2^*-weighted 7T MR and CT images were investigated, revealing a maximum deviation of 0.78 mm (Cho et al., 2010). Finally, registration of 7T T1- and T2-weighted images of the midbrain of PD patients to 1.5T T1- and T2-weighted images showed that mainly rigid body transformations were required and that scaling and skew deformations were small (Duchin et al., 2012). Furthermore, the midbrain region, containing many DBS targets, required the least correction. Quantitative comparison showed that the distances of the T2-weighted images were significantly less than 1 mm suggesting that affine registration of T1- and T2-weighted 7T images to CT images can already provide MR images with midbrain distortions comparable to those of 1.5T images. These few studies suggest that at 7T images can be acquired with distortions smaller than 1 mm in the deep-brain areas.

Secondly, some of the mentioned imaging techniques pose additional challenges in the clinical context. Most studies were performed on young and healthy volunteers. In patients, movement during image acquisition can be less controlled, counteracting the gain in SNR and spatial specificity obtained with ultra-high field. However, newer techniques, such as prospective motion correction might remedy this problem (Maclaren et al., 2013). This approach monitors movement in the scanner with high accuracy and corrects the new image acquisition adaptively according to the new head position. That is, even with large head movements—as observed in many patients—the resulting images are already coregistered and movement artifact free.

In addition, the availability of ultra-high field MR scanners is currently limited. Firstly, the number of scanners that have been installed in the world is limited itself (see Table 2), which is inherent to its high cost in purchase and in operation. Secondly, due to the novel status of ultra-high field MRI, safety precautions regarding metallic objects are often more strict than on 3T systems and the use of ultra-high field MRI is currently only allowed for research purposes.

Finally, direct targeting in DBS suffers from brain shift, intra-operative deformation of the brain compared to preoperative MR images due to difference in head position and cerebrospinal fluid loss. Without compensation for this, it will eventually still limit targeting accuracy. However, this effect is independent of the magnetic field strength and even the pre-operative imaging modality.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.