The brainstem is the primary relay center for afferent and efferent connections between the cerebral cortex, the cerebellum, and the spinal cord (Sutin and Carpenter, 1976). It plays an important role in the regulation of vital functions (e.g., respiratory function, cardio-vascular function, nausea) and coordinates motor control signals sent from the brain to the body. Anatomically, the human brainstem consists of three connected parts, medulla oblongata, pons, and midbrain, and is composed of a multitude of axonal nerve fibers as well as cranial and non-cranial nerve nuclei. Due to the spatial concentration of important neural structures in this relatively small brain region, pathological variations or lesions in the brainstem can introduce severe and multiple neurological effects (Donaldson et al., 2006). The brainstem is also known to be involved in a number of degenerative diseases, such as Alzheimer’s disease, Parkinson’s disease, and Huntington’s disease (Urban and Caplan, 2011). Hence, detailed non-invasive depiction of its morphology and anatomy may help to better relate normal and pathological anatomical variations to medical conditions as well as neurological and peripheral functions.

Magnetic resonance imaging (MRI) at ultrahigh magnetic field strength (B₀ ≥ 7 T) is a powerful means to assess non-invasively normal and abnormal brain tissue with high spatial resolution (Li et al., 2006; Duyn et al., 2007; Deistung et al., 2008; Trampel et al., 2011; Moser et al., 2012; Turner, 2012, 2013; Eichner et al., 2013; Marques and Gruetter, 2013). MRI provides a variety of qualitative and quantitative tissue contrasts, mainly reflecting nuclear relaxation (T₁, T₂, and T2*, or equivalently R₁, R₂, and R2*), diffusion, and magnetic susceptibility. In contrast-sensitized MR images (e.g., T₁-weighted, T₂-weighted, or T2*-weighted images), however, the measured signal is an intricate function of proton density, longitudinal (T₁) and transverse relaxation (T₂) and depends on the applied MRI sequence parameters. Thus, it may be difficult to interpret the underlying tissue composition and structure based on the intensity of these images. Quantitative MRI techniques, on the other hand, provide information that is intrinsically more tissue specific and consequently less dependent on the chosen scan parameters. Hence, quantification of different physical tissue properties by using quantitative MRI techniques also increases intra- and inter-individual comparability and enables objective measurements of disease related brain changes (Tofts, 2003).

The most common approach of quantitative MRI is relaxometry, i.e., mapping of the longitudinal (R₁) and/or transverse (R₂, R2′, R2*) relaxation rates. In the brain these relaxation rates are influenced predominantly by both myelin (Koenig, 1991; Bender and Klose, 2010; Langkammer et al., 2012a; Lee et al., 2012) and tissue iron (Ogg and Steen, 1998; Langkammer et al., 2010). An additional, recently introduced novel quantitative MR method that is unique in its sensitivity to both tissue constituents is quantitative susceptibility mapping (QSM) (Schweser et al., 2011; Reichenbach, 2012). QSM is a post-processing method that uses the phase signal of complex-valued gradient (recalled) echo (GRE) data to produce maps of tissue magnetic susceptibility. Quantification of brain tissue iron content (Wharton et al., 2010; Schweser et al., 2011; Bilgic et al., 2012; Langkammer et al., 2012b; Zheng et al., 2013) and blood oxygenation in vivo (Haacke et al., 2010) as well as assessment of brain myelination (Liu et al., 2011; Li et al., 2012a,b) has recently been reported by applying QSM. Excellent anatomical delineation of cortical and deep gray matter structures and substructures has been demonstrated with QSM (Schweser et al., 2011, 2012; Deistung et al., 2013a), particularly in concert with relaxation rate mapping (Sigalovsky et al., 2006; Fukunaga et al., 2010; Geyer et al., 2011; Lebel et al., 2012; Deistung et al., 2013a), thus rendering this combined approach highly promising for improved delineation of the anatomical structure of the brainstem.

Nerve fibers and their complex network are usually imaged and analyzed by exploiting the translational molecular diffusion of protons. So far, most studies have applied diffusion tensor imaging (DTI) (Basser and Pierpaoli, 1996) for anatomical imaging of the brainstem (Stieltjes et al., 2001; Nagae-Poetscher et al., 2004; Salamon et al., 2005). DTI, however, is commonly limited by the relatively coarse spatial resolution available, which makes detailed analysis of smaller fiber tracts difficult, and often fails in regions of crossing fibers (Jones et al., 2013). One recently suggested approach to overcome this limitation is track-density imaging (TDI), which combines high angular resolution diffusion imaging (HARDI) with fiber tracking information to delineate structures beyond the image resolution of the acquired diffusion-weighted data (Calamante et al., 2010). TDI has been already applied successfully to delineate fiber pathways in the cerebrum and cerebellum, and the substructure of the thalamus (Calamante et al., 2010, 2013).

The present study explores the potential of these high-resolution MRI approaches to resolve non-invasively the intricate anatomy of the human brainstem at 7 T in vivo. To this end, T₂-weighted images, quantitative R₁, R2*, and magnetic susceptibility maps as well as track-density images were compared with respect to their ability of depicting nerve nuclei and nerve fibers. Furthermore, R₁, R2*, and magnetic susceptibility values are presented for selected anatomical regions. It is shown that the combination of multiple image contrasts provides a distinctly improved portrayal of the morphology of the brainstem.

All qualitative findings presented in this study were consistent across all subjects unless otherwise specified.

Figure 1 presents representative MR images and corresponding histological sections of the midbrain. The red nuclei and the substantia nigra were most strikingly discernible on the R2* (Figure 1B) and susceptibility maps (Figure 1C) as well as on the direction-encoded track-density (Figure 1D) and T₂-weighted images (Figure 1E), but were barely visible on the R₁ map (Figure 1A). This qualitative finding was also supported by the CNR measurements presented in Figure 2. On the track-density image (Figure 1D) both nuclei showed lower anisotropy as reflected in the reduced color saturation (arrows b and d). Different fiber bundles traversing the crus cerebri [corticobulbar fibers (arrow c₁), corticospinal fibers (arrow c₂), and corticopontine fibers (arrow c₃)] could be identified on the track-density images. In the midbrain, interestingly, the central tegmental tract (arrow f) and the medial lemniscus (arrow e, five of six subjects) were only identifiable reliably on susceptibility maps, whereas the mammillary body (Figure 1; arrow a) was discernible on all MR image contrasts except the track-density images. The superior (not shown) and inferior colliculi (Figure 1; arrow g) were distinguishable across all subjects only on the R₁ maps. Both of these structures could be assessed on the other MRI contrasts in at least four of six subjects (superior colliculus on susceptibility maps in six of six subjects). The superior colliculus and inferior colliculus presented highest CNR on susceptibility maps and R2* maps, respectively (Figure 2). Anatomic regions with low magnetic susceptibility (Figure 1C; arrows c, e, f), such as the hypointense rim around the red nuclei coincided with regions of high myelin content (Figure 1G; arrows c_1–3, e, f) and had no unique imaging correlate on any of the other MR images. The histological stains (Figures 1G,H) showed increased myelin and decreased cell density in the red nucleus compared to the substantia nigra.

MR images of the rostral and the middle part of the pons are depicted in Figures 3 and 4, respectively. The R2* and magnetic susceptibility maps clearly displayed fibers traversing the pons in medial-lateral direction and fibers of the middle cerebellar peduncle (MCP) (arrow a in Figure 3) as also indicated by the myelin stains (Figures 3G and 4G). These fiber structures were barely visible on the R₁ maps [Figures 2 (TPF), 3A, and 4A]. The direction of the transverse pontine fibers could not be reliably separated from the superior-inferior running corticospinal tract on the track-density images (Figures 3D and 4D). In contrast, the susceptibility maps revealed even smaller fiber tracts [e.g., central tegmental tract (arrow f, two of six subjects), medial longitudinal fasciculus (arrow e)] and nerve nuclei [reticulotegmental nucleus (arrow c)] that were indiscernible on R₁ and R2* maps as well as on the T₂-weighted images. In the direction-encoded track-density images larger [e.g., MCP (arrow a), superior cerebellar peduncle (arrow d), corticospinal tract (arrow h), pontocerebellar fibers (white solid outline in Figure 3D)] and smaller [e.g., central tegmental tract (arrow f)] fiber tracts were clearly identifiable. The MCP was also discernible on the R₁, R2* and susceptibility maps, whereas the superior cerebellar peduncle displayed on the R₁ and the susceptibility maps on all subjects and in two of six subjects on the R2* maps. Both of these structures (MCP, SCP) were also barely visible on T₂-weighted images (MCP in three subjects, SCP in one subject; Figure 2). The susceptibility maps showed the highest CNR for nearly all investigated structures within the pons (Figure 2). Only the CNR value of MCP with respect to adjacent cerebrospinal fluid were outperformed by the R2* maps. Although pontine veins were only seen on both R2* and susceptibility maps (Figures 4B,C; arrow n), their identification compared to adjacent tissue structures (e.g., transverse pontine fibers) was substantially improved on susceptibility maps. The solid white outline in Figures 4A–D encloses the caudal part of the pontine reticular nucleus, the facial nucleus, salivatory nucleus and nucleus abducens. Although R₁, R2*, and susceptibility maps displayed heterogeneous signal intensities in this area, discrimination of these nuclei was not possible. The interruption of the fiber tracts between the MCP seen on the track-density image (dashed outline in Figure 4D) is most likely caused by data acquisition and TDI post-processing issues (incompletely corrected geometric distortions in the unwarped diffusion-weighted images that impeded reliable TDI post-processing) because no such interruption is observed on the susceptibility map (Figure 4C) and on the histological stains (Figures 4G,H). The T₂-weighted images (Figures 3E and 4E) displayed rather homogeneously with low contrast (Figures 2–4), which prevented reliable identification of any substructures across all subjects.

Figures 5 and 6 display the rostral and middle part of the medulla oblongata, respectively. The pyramid (arrow a) is located anterior to the inferior olive and was clearly visible on all contrasts (four of six subjects on R2* maps) except the T₂-weighted images. It could be well discriminated from the inferior olive (arrow b) which was also clearly discernible on the R₁ (five subjects) and susceptibility map (four subjects), and less obvious on R2* maps (three subjects). The inferior cerebellar peduncle was clearly visible on track-density images and was distinguishable on R₁, R2*, and susceptibility maps in four subjects at least. Only the susceptibility maps enabled identification of the spinal trigeminal nucleus (arrow c in Figure 5; four subjects). Although the histological stains (Figures 5G,H) suggested the presence of several cranial nuclei, further nuclei were not visible on the MR images.

Finally, R₁, R2*, and susceptibility differences are listed in Table 1 for selected anatomical regions. The differences of the susceptibility and the R2* values between fibers running nearly parallel (transverse pontine fibers) and nearly perpendicular (corticospinal tract) to the static magnetic field were (−49 ± 11) ppb and (8.9 ± 6) s⁻¹, respectively.

We have investigated the anatomy of the human brainstem in vivo by applying MR imaging at 7 T with high spatial, isotropic resolution, and using multiple image contrasts. QSM and TDI have been applied jointly for the first time to identify anatomical substructures of the brainstem. Both MR methods directly reflect subtle variations in tissue composition that were found to be consistent with histology as demonstrated with corresponding histological stains selected from a data base. In contrast, longitudinal relaxation rate mapping and T₂-weighted imaging both performed inferiorly in their ability to delineate anatomical details (Figure 2).

R2* relaxation comprises both intrinsic microscopic transverse relaxation (R₂) and relaxation due to heterogeneous magnetic susceptibility distributions on a mesoscopic or macroscopic scale (R2′). In contrast to the R2* relaxation rate, which increases with the concentration of both iron (Langkammer et al., 2010) and myelin (Lee et al., 2012), magnetic susceptibility shows a different dependence on the concentration of these substances, i.e., higher magnetic susceptibility values with increasing iron concentration (Langkammer et al., 2012b), and lower magnetic susceptibility values with increasing myelin density (Liu et al., 2011; Schweser et al., 2011; Li et al., 2012a,b). Hence, image contrast on R2* and susceptibility maps appears complementary. The longitudinal relaxation rate, R₁, in turn is heavily influenced by myelin content, but shows little effect with respect to tissue iron concentration (Bridge and Clare, 2006).

Both R2* maps and susceptibility maps provided detailed depiction of the non-cranial nuclei and nerve fibers in the midbrain in accordance with recent studies (Schweser et al., 2011; Deistung et al., 2013a). In particular, the red nuclei and the substantia nigra were clearly discernible on susceptibility and R2* maps because both structures contain substantially higher iron concentrations compared to the surrounding brainstem tissue (Morris et al., 1992). Interestingly, the highly myelinated rim around the red nuclei, which was clearly seen on the myelin stain and on the susceptibility map was not visible on the R₁ map, potentially due to the lower spatial resolution of the MP2RAGE acquisition compared to the gradient-echo acquisition.

Brainstem nuclei were barely seen on the R₁, R2* and susceptibility maps sectioning the pons and medulla oblongata; however, the inferior olive (Figures 5 and 6, arrow b) could be identified and both the spinal trigeminal nucleus (Figure 5C, arrow c) and the reticulotegmental nucleus (Figure 3C, arrow c) were discernible on the susceptibility maps. The inferior olive is the sole source of the climbing fibers to the Purkinje cells in the cerebellar cortex and is involved in learning and timing of motor behavior (De Zeeuw et al., 1998). Lesions in the inferior olive can thus introduce restrictions in performing specialized motor tasks (Martin et al., 1996). The spinal trigeminal nucleus is the largest of all cranial nerve nuclei and one of the three paired sensory nuclei associated with the trigeminal nerve. It receives input from pain, temperature, and some tactile afferents in the trigeminal nerve and is involved in migraine headache (Kaube et al., 2002). The reticulotegmental nucleus of the pons is an important component of the oculomotor circuit that regulates horizontal saccades and smooth pursuit movements of the eyes (Keller and Crandall, 1981; Büttner-Ennever and Horn, 1997). Damage to the reticulotegmental nucleus has been observed in spinocerebellar ataxia (Rüb et al., 2004) and it most probably contributes to hypometric horizontal saccades and slowing of smooth pursuits that characteristically develop in patients suffering from Alzheimer’s disease (Rüb et al., 2001). Therefore, non-invasive assessment of these nuclei may help to improve diagnosis and understanding of these diseases.

These nuclei (inferior olive, spinal trigeminal nucleus, reticulotegmental nucleus) were visible on the susceptibility maps because of their different iron concentration and/or myelin content with respect to surrounding tissue. A histological study by Morris et al. (1992), for instance, demonstrated iron accumulation in the nuclei of the dorsal parts of the brainstem, the pyramids, the spinal trigeminal nucleus, and the inferior olive (although with substantially lower iron concentration than in the red nuclei and substantia nigra) and reported only little or no intrinsic iron reactivity for many other nuclei and tracts of the brainstem (e.g., solitary nucleus, hypoglossal nucleus, raphe pallidus nucleus, cuneate fasciculus, and gracile fasciculus). Brainstem nuclei have a similar, relatively low myelin density as seen from the histological myelin stains. Only the inferior olive (Figure 5G, arrow b) and the spinal trigeminal nucleus (Figure 5G, arrow c) and reticulotegmental nucleus [see (Paxinos et al., 2012) and Figure 3C, arrow c] are embedded in tissue with substantially increased myelin density.

In addition to myelin’s bulk diamagnetic nature, a dependence of the magnetic susceptibility on the axons’ alignment with respect to the external magnetic field has been recently reported (Liu, 2010; Li et al., 2012a; Schweser et al., 2012; Wharton and Bowtell, 2012). Similar direction dependent observations have been made for R2* (Bender and Klose, 2010; Denk et al., 2011; Lee et al., 2011; Sati et al., 2012). We measured differences of the susceptibility and the R2* values between fibers running nearly parallel (transverse pontine fibers) and nearly perpendicular (corticospinal tract) to the static magnetic field of (−49 ± 11) ppb and (8.9 ± 6) s⁻¹, respectively. This susceptibility difference is nearly threefold as large as the values that have been derived for a hollow cylinder model of myelin (−18 to −16 ppb) (Li et al., 2012a; Wharton and Bowtell, 2012). In vivo, however, Li et al., 2012a also reported susceptibility differences between parallel and perpendicular fibers of up to −40 ppb in individual white matter regions, a value which is almost the same as reported here. The result for the orientation dependence of R2* is in good agreement with the recently reported ex vivo value of −6.44 ± 0.15 s⁻¹ at 7 T (Lee et al., 2011). The orientation-dependent differences of magnetic susceptibility and R2*in our study may be caused by the presence of additional contributors other than myelin, such as pontine nuclei and neuropil within the investigated volumes-of-interest. Nevertheless, consistent with the theory of myelin’s orientation dependency, the contrast between fibers running in medial-lateral direction (oriented nearly perpendicular to the magnetic field; e.g., transverse pontine fibers, arrow o in Figure 4) and fibers running inferior-superiorly (oriented nearly parallel to the magnetic field; corticospinal tract) is enhanced on the R2* and susceptibility maps. More sophisticated approaches, such as susceptibility tensor imaging (STI) (Liu, 2010; Li et al., 2012a) or multipole susceptibility tensor imaging (MSTI) (Liu and Li, 2013) may further improve discrimination between fibers and nuclei of the pons and medulla, particularly in combination with high spatial resolution.

Direction-encoded TDI enabled discrimination of the directionality of nerve fibers within the brainstem and yielded complementary information to the relaxation and susceptibility maps. The depiction of larger fiber bundles was consistent with the results of other DTI studies conducted in vivo at lower magnetic fields (Nagae-Poetscher et al., 2004; Salamon et al., 2005). We were, however, unable to resolve unambiguously the transverse pontine fibers, which were clearly visible on the R2* and susceptibility maps (Figure 4). This may be due to the lower base voxel resolution of 1.2 mm of the DWI data, which was chosen as a compromise between spatial resolution, SNR, and acquisition time. To overcome this limitation we interpolated the DWI data using a super-resolution approach (TDI) to obtain an isotropic virtual resolution of 0.43 mm. It should be noted, however, that the TDI technique is not able to fully recover the whole information that would be present if the data would be acquired with such a high spatial resolution (Calamante et al., 2011). Due to the larger errors in tractography by performing fiber tracking on data with a lower resolution, the higher resolved track-density images appeared slightly blurred and lost some fine details; explaining the difficulties in identifying the transverse pontine fibers in the presented track-density images (Figures 3D and 4D). As might be expected, the level of blurring in the track-density images depends on the desired virtual resolution (Calamante et al., 2011). Higher spatial resolution is thus needed to delineate transverse pontine fibers unambiguously, which requires sophisticated modifications of the diffusion MRI sequence such as combining zoomed imaging and parallel imaging (Heidemann et al., 2012; Eichner et al., 2013). Karampinos et al. (2009) were recently able to resolve transverse pontine fibers in direction-encoded fractional anisotropy maps with a resolution of 0.8 mm × 0.8 mm × 3 mm at 3T by applying a dedicated acquisition scheme (self-navigated, multi-shot, variable-density, spiral-imaging with outer volume suppression).

One particular problem of brainstem imaging with MRI are the induced field inhomogeneities caused by nearby air-tissue or bone-tissue interfaces. In the vicinity of such interfaces significant image distortions of EPI-based diffusion-weighted images occur and additional spin dephasing arises in gradient-echo images. Despite unwarping the diffusion-weighted images based on gradient-echo field-map information, it was not possible to completely compensate geometric distortions, thus impeding accurate depiction of ventral fiber structures of the middle pons and medulla oblongata on track-density images. These induced field inhomogeneities manifest themselves as steep gradients in gradient-echo phase images and lead to signal reductions in gradient-echo magnitude images and decreased SNRs, affecting both susceptibility maps [due to increased phase noise (Gudbjartsson and Patz, 1995)] and R2* maps. In future studies, the impact of spin dephasing due to air-tissue or bone-tissue interfaces on R2* maps may be minimized by taking into account the macroscopic field inhomogeneity reflected in the phase images (de Leeuw and Bakker, 2012). R₁ maps and T₂-weighted images, on the other hand, are distinctly more immune against these field inhomogeneity effects.

The quantitative analysis revealed similar values for the longitudinal relaxation rates (R₁) for nuclei and myelinated fibers (Table 1), whereas susceptibility and R2* values varied more substantially. This broader spread of R2* and susceptibility values is reflected in the large CNR values (Figure 2) underlining the good discrimination of nerve fibers on susceptibility maps and the exquisite delineation of midbrain nuclei on susceptibility and R2* maps. It should, however, be noted that the discrimination of brain tissue with respect to cerebral spinal fluid (CSF) on susceptibility maps is not as striking as on the R₁, R2*, and T₂-weighted images [Table 1; Figure 4 (arrow p)]. Due to the quantitative nature of R₁, R2*, and magnetic susceptibility maps, these contrasts may be combined by projections along support vectors observed from discriminant analysis or support vector machines to generate composite images with improved depiction of anatomical features while providing improved discrimination of CSF (Deistung et al., 2013b).

In conclusion, maps of magnetic susceptibility displayed the largest number of brainstem structures, including larger and smaller fiber pathways as well as several nerve nuclei. Usage of multiple image contrasts enables a detailed non-invasive view into tissue structure and composition. Hence, multi-contrast MR imaging of the brainstem at ultra-high magnetic fields that utilizes relaxation, diffusion, and magnetic susceptibility information is a versatile tool to assess anatomy individually in great detail.

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.