Direct In Vivo MRI Discrimination of Brain Stem Nuclei and Pathways

BACKGROUND AND PURPOSE: The brain stem is a complex con ﬁ guration of small nuclei and pathways for motor, sensory, and au-tonomic control that are essential for life, yet internal brain stem anatomy is dif ﬁ cult to characterize in living subjects. We hypothesized that the 3D fast gray matter acquisition T1 inversion recovery sequence, which uses a short inversion time to suppress signal from white matter, could improve contrast resolution of brain stem pathways and nuclei with 3T MR imaging. MATERIALS AND METHODS: After preliminary optimization for contrast resolution, the fast gray matter acquisition T1 inversion recovery sequence was performed in 10 healthy subjects (5 women; mean age, 28.8 6 4.8 years) with the following parameters: TR/ TE/TI ¼ 3000/2.55/410ms, ﬂ ip angle ¼ 4°, isotropic resolution ¼ 0.8 mm, with 4 averages (acquired separately and averaged outside k -space to reduce motion; total scan time ¼ 58minutes). One subject returned for an additional 5-average study that was combined with a previous session to create a highest quality atlas for anatomic assignments. A 1-mm isotropic resolution, 12-minute ver-sion, proved successful in a patient with a prior infarct. RESULTS: The fast gray matter acquisition T1 inversion recovery sequence generated excellent contrast resolution of small brain stem pathways in all 3 planes for all 10 subjects. Several nuclei could be resolved directly by image contrast alone or indirectly located due to bordering visualized structures (eg, locus coeruleus and pedunculopontine nucleus). CONCLUSIONS: The fast gray matter acquisition T1 inversion recovery sequence has the potential to provide imaging correlates to clinical conditions that affect the brain stem, improve neurosurgical navigation, validate diffusion tractography of the brain stem, and generate a 3D atlas for automatic parcellation of speci ﬁ c brain stem structures.

T he human brain stem represents the complex interdigitation of compact anatomic pathways and nuclei that control or modulate motor, sensory, autonomic, and cognitive functions. Many brain stem structures are essential for survival. 1 Patients can experience profound symptoms and disability even from small focal brain stem lesions due to ischemic stroke, multiple sclerosis, infections, vascular malformations, or tumors. [2][3][4][5][6] Histopathologic changes in specific brain stem structures can lead to neurodegeneration associated with movement disorders, such as Parkinson disease, progressive supranuclear palsy, and multiple system atrophy. [7][8][9][10][11] Even with state-of-the-art clinical MR imaging, it remains challenging to confidently localize the spatial locations of specific brain stem structures or to detect early pathologic changes to these structures in individual patients before postmortem examination.
To estimate the location of a particular brain stem structure in current practice, clinicians triangulate on the basis of surface landmarks, limited MR imaging-visible internal features, and mental representations of histology-stained sections from the same superior-inferior level. Systematic dissections of normal cadaver brains may improve the accuracy of this qualitative approach, particularly to better identify safe entry zones for surgical procedures of the brain stem. 12,13 However, this overall approach does not account for anatomic distortion that frequently occurs with pathology (ie, no one performs surgery on the normal brain stem). Unfortunately, most stereotactic imaging-based brain atlases have emphasized resolution of cortex, white matter, or specific functional neurosurgery targets within the diencephalon. [14][15][16][17][18] Detailed image-based parcellation for internal brain stem anatomy remains scant. 19,20 The widely used FreeSurfer (http://surfer.nmr.mgh.harvard.edu) parcellation provides a single atlas label for the entire brain stem, and the more recent brain stem substructures algorithm only divides the brain stem into "midbrain," "pons," and "medulla." [21][22][23] High-field MR imaging of postmortem, isolated brain stem specimens reveals the potential for MR imaging contrast resolution of internal brain stem anatomy comparable with histology, [24][25][26][27] though external validity is limited because this requires 8+ hours of scanning-dissected, coil-optimized individual specimens that may have altered MR imaging properties due to fixation and the postmortem interval. 28,29 Improved internal brain stem contrast also can be obtained with in vivo MR imaging using 7T systems, [30][31][32][33] susceptibility mapping, 34,35 relaxometry, 30,35,36 or various representations/models of diffusion contrast. [37][38][39] Diffusion tractography approaches can putatively identify many internal pathways and features of the brain stem. 36,39 However, diffusion tractography is prone to spurious tract generation and true-positive segmentations are challenging to validate. 40 Furthermore, diffusion contrast is often altered by underlying pathology (unfortunately when identifying brain stem structures is most needed clinically).
Several groups have previously reported novel MR imaging sequences that manipulate T1 relaxation to suppress signal from white matter and maximize internal contrast within the thalamus and basal ganglia for direct targeting during functional neurosurgery. [41][42][43][44][45][46] These sequences have various acronyms, but perhaps the most relevant was fast gray matter acquisition T1 inversion recovery (FGATIR). 42 Optimization of image contrast by manipulation of MR imaging sequence parameters was described in detail previouslythe key concept is suppression of signal from voxels that contain myelin, even if the tissue does not contain classic white matter bundles by histology (eg, individual thalamic nuclei). Here, we demonstrate the tremendous advantages of also applying this MR imaging contrast mechanism for the direct identification of many internal brain stem structures in individual living subjects.

MATERIALS AND METHODS
This study was compliant with the New York University institutional review board and requirements of the Health Insurance Portability and Accountability Act. The data that support the findings of this study are available from the corresponding author on reasonable request. Healthy adult volunteers were imaged using a 3T Magnetom Prisma scanner with a 64-channel head and neck coil (Siemens). The sequence used here differs minimally from the originally reported FGATIR MR imaging sequence, 42 or the more recent fluid and white matter suppression (FLAWS) component 43 of the updated MP2RAGE sequence. 44 For simplicity, we refer to the sequence used as FGATIR throughout the remainder of this report. For all experiments described below, parallel imaging was not used, the radiofrequency mode was "fast," and "prescan normalize" was used (per manufacturer). Complete protocol details will be provided to interested readers. We empirically developed a protocol to obtain excellent contrast and spatial resolution to discriminate brain stem structures as determined by consensus between 2 board-certified diagnostic neuroradiologists with expertise in brain stem neuroanatomy.
We first explored how to use signal averaging during image reconstruction to improve contrast resolution of brain stem structures using a 1-mm isotropic resolution sequence with the following parameters: TR/TE/TI ¼ 3000/2.4/410 ms, nonselective 180°inversion pulse, flip angle ¼ 6°, 256 Â 256 matrix, 256-mm-square FOV, 160 Â 1 mm sagittal slices, bandwidth ¼ 320 Hz/pixel, time ¼ 12 minutes 9 seconds per 1 average. Due to the relatively long scan time per signal average and potential for subtle-but-compounding motion artifacts, we compared obtaining 4 averages in k-space before Fourier image transformation (the standard approach to signal averaging) versus when 4 averages were obtained independently, reconstructed to image space, spatially coregistered using a 6-df rigid-body transform with FMRIB Linear Image Registration Tool (FLIRT; http://www.fmrib.ox.ac.uk/fsl/fslwiki/FLIRT), and finally averaged together. While there is a signal-to-noise penalty for not obtaining multiple averages in k-space before Fourier image transformation, 47 this approach (4 separate 12-minute 9-second scans versus a continuous scan time of 48 minutes 36 seconds) resulted in less image degradation from subtle head motion, even in cooperative, tolerant, and experienced volunteers.
Next, we compared contrast resolution between 1-and 0.8mm isotropic resolution in which the latter sequence parameters were the following: TR/TE/TI ¼ 3000/2.11/410 ms, nonselective 180°inversion pulse, flip angle ¼ 6°, matrix ¼ 288 Â 288, FOV ¼ 230 mm square, 192 Â 0.8 mm sagittal slices, bandwidth ¼ 460 Hz/pixel, time ¼ 14 minutes 26 seconds per 1 average. The image quality and fine anatomic detail of the 0.8mm isotropic resolution were preferred by both neuroradiologists. Then, a single subject underwent the 0.8-mm isotropic protocol in 2 separate sessions with 4 and 5 individual averages (57 minutes 44 seconds and 72 minutes 10 seconds, respectively). Within-session averages were coregistered using 6-df rigid-body transforms, and data between the 2 scan sessions were aligned using a 12-df affine transform. Individual averages were then combined sequentially to assess image-quality improvement with additional averages. The 9-average dataset was used for annotating the relevant brain stem anatomy in this report; however, 4 averages were more practical for multiple subjects and appeared to provide sufficient contrast resolution for most individual brain stem structures (per both neuroradiologists). Ten right-handed subjects (5 women, 28.8 6 4.8 years of age, mean Edinburgh Handedness Inventory score ¼ 86 6 19) then were each imaged using 4 individual averages of the 0.8-mm FGATIR protocol to assess inter-subject variability in brain stem anatomy. Anatomic assignment of structures was determined by consensus between the 2 neuroradiologists using commonly available references and standard nomenclature. [48][49][50] Note that axial images were created parallel to the commissural plane, while coronal images were parallel to the rhomboid fossa (which is not orthogonal to the commissural plane) (On-line Fig 1). For all figures, please see the Table for the specific labeled structures.
Key summary of labels for multiplanar anatomic images of the brain stem (Figs 1-6 and On-line Fig 2) Label  Finally, to demonstrate clinical feasibility, we performed a single average, 1-mm isotropic resolution FGATIR sequence ($12 minutes) in a patient with hemiplegia after ischemic stroke involving the corticospinal tract.

RESULTS
The FGATIR sequence provided novel MR imaging-derived contrast throughout the in vivo brain stem and allowed the direct identification of many small functionally important structures. Structure identification by 2 neuroradiologists was facilitated by 3plane cross-referencing. Selected serial axial images illustrate contrast resolution of internal brain stem anatomy at canonical levels (Fig 1). Selected coronal and sagittal images also reveal excellent contrast resolution of internal brain stem anatomy (Figs 2 and 3, respectively). Such longitudinal images demonstrate the superiorinferior course of several brain stem pathways and the potential for the 0.8-mm isotropic FGATIR to enhance our understanding of the 3D organization of the brain stem. Videos of axial, coronal, and sagittal images through the brain stem are provided to interested readers on-line (On-line Videos 1-3).
Within the brain stem, the signal intensity or brightness of a specific structure appeared inversely correlated with the intensity of myelin staining in postmortem histology atlases. [48][49][50] Blood in the vertebral arteries and dural venous sinuses appeared the brightest. The next most hyperintense structures were CSF and adjacent gray matter such as the periaqueductal gray and area postrema, the latter a circumventricular organ with an altered extracellular matrix and vascular structure. 51,52 The third most hyperintense structures were more central gray matter within the brain stem such as the substantia nigra and pontine nuclei. Structures with intermediate hyperintensity were less compact or less densely myelinated structures like the inferior colliculi, central tegmental tract, and corticonigral pathways. Dark structures were classic myelinated brain stem pathways such as the medial longitudinal fasciculus, medial lemniscus, and pontocerebellar fibers. The darkest structures were densely myelinated pathways, including the medullary pyramids, superior/middle cerebellar peduncles, and larger cranial nerves. Cortical bone in the central skull base and the tectorial membrane also appeared very dark. Of note, the signal intensity (and shape) of the corticospinal tract varied along its course, appearing dark within the cerebral peduncles and medullary pyramids, but more diffuse and relatively brighter within the basis pontis. This appearance may represent interdigitation of pontine nuclei and pontocerebellar projections between white matter fascicles of the corticospinal tract. [48][49][50]53 Labeling in the figures emphasizes unambiguous structures that can be identified well on the images. Most of these labeled structures are wellknown to clinicians familiar with brain stem anatomy (eg, the medial  longitudinal fasciculus). Directly labeled visible structures also can be used to generate more exact indirect localization for other bordering internal brain stem structures. Figure 4 provides a highly detailed example of this approach for the inferior midbrain where the likely positions of multiple additional structures can be estimated relative to the borders and anatomic spaces formed between the medial longitudinal fasciculus, medial lemniscus, and decussation of the superior cerebellar peduncles. Annotation at this level of detail with descriptions of functional relevance is possible throughout the brain stem, but is beyond the scope of this initial report. Other selected examples of indirect localization that may be of general interest include the oculomotor complex (bright region posteromedial to the central tegmental tract in Fig 1A), pedunculopontine nucleus (bright region medial to the medial lemniscus and lateral to the decussation of the superior cerebellar peduncles, in Fig  1B), locus coeruleus (bright region posterior to the central tegmental tract and medial to the superior cerebellar peduncle in Fig 1C), and facial nucleus (bright region posterolateral to the central tegmental tract and medial to the cerebellar peduncles in Fig 1F). An inferior notch along the medial aspect of the left middle cerebellar peduncle with gray matter signal intensity in Fig 2D is the cochlear nucleus.
There also are subtle features in the images that are difficult to assign with certainty (and sometimes only visualized on 1 side) that likely correspond to specific structures better resolved with postmortem MR microscopy 53 and histology atlases. [48][49][50] Limited visualization may be from partial volume effects due to the small size of the structures relative to 0.8-mm isotropic voxels and/or less contrast relative to surrounding nervous tissue. On the right side of Fig 1B, there is an arc-like medially concave dark structure oriented anterior-posterior, consistent with the expected location of the left oculomotor nerve. In Fig 1C, -D, the tectospinal tract appears as a faint dark dotlike structure anterior to the medial lemniscus, particularly on the left side. In Fig 1E, the left tectospinal tract is more obvious, but there is marked asymmetry compared with the subject's right side. Also, in Fig 1E, the indistinct anterior margin of the medial lemniscus may represent the anteromedial continuation of the trapezoid body. In Fig 2D, an intermediate signal intensity line oriented inferomedial to superolateral, dividing the area postrema (on the right side of the image), may be the sulcus limitans. Some structures also are better recognized when the imaging plane is transverse to the structure. For example, the facial nerve in the lateral midpons is difficult to appreciate in the axial plane (Fig 1F), but better resolved in the coronal plane ( Fig 2D).
The FGATIR images did not discriminate all brain stem structures well. It was not possible to distinguish the corticospinal tract from other white matter in the cerebral peduncle ( Fig 1A). Boundaries of the medial lemniscus, tectospinal tract, and medial longitudinal fasciculus, which are adjacent to one another in an anterior-posterior orientation within the medulla, remain indistinct (Fig 1G). The inverse relationship of MR imaging signal intensity to white matter content described above also was not entirely consistent. For example, the red nucleus appeared isointense to enveloping cerebellorubral and cerebellothalamic fibers on MR imaging, but these structures are more distinct with classic white matter staining on histology. 48,49 Contrast discrimination of structures improved with more signal averages for the FGATIR sequence with 0.8-mm isotropic resolution, but subjectively yielded less improvement beyond 4 averages (On-line Fig 2). Thus, a 4-average protocol was used to image 10 healthy volunteers ($1-hour scan time) to assess individual variability in contrast (Fig 5). We observed subtle differences in morphology (eg, compare the medullary pyramids in A and C) or contrast resolution (eg, compare sensory decussation in A and D), but the overall contrast and the structures observed appeared consistent across individuals. In some subjects, the Direct, unambiguous identification of many myelinated structures in the inferior midbrain (A) also helps improve the spatial accuracy for indirect localization of bordering brain stem nuclei and pathways that are less densely myelinated at this level (B). The ventral trigeminothalamic tract 61 is located between the decussation of the superior cerebellar peduncles and medial lemniscus. Subtle dark bumps on the anterior and posterior aspects of the decussation 15 are consistent with the rubrospinal 17 and tectospinal tracts, 58 respectively. The bright midline structure anterolateral to the medial longitudinal fasciculus 8 should be the substantia nigra pars compacta. 5 The reticular formation 60 is the bright region posterolateral to the medial longitudinal fasciculus and central tegmental tract, 10 and anterolateral to the periaqueductal gray. 11 This region will also contain the locus coeruleus. The trochlear nucleus 59 should be along the posterior border of the medial longitudinal fasciculus. The interpeduncular nucleus 57 is off-midline anterior to the decussation of the superior cerebellar peduncles. Similar indirect localization can be derived at other brain stem levels, but is beyond the scope of this initial report.  Fig 1F) selected from 4 different healthy adult controls (2 men) suggest reproducibility of overall image contrast across individuals and sexes (0.8-mm isotropic resolution, 4 averages, 58-minute total scan time, scale bar ¼ 5 mm). The putative location of the pontomedullary reticular formation, a structure of interest for predicting limb function recovery after stroke, is denoted with an asterisk.
AJNR Am J Neuroradiol : 2020 www.ajnr.org corticospinal tract signal intensity appeared asymmetrically darker on the left within the basis pontis (eg, Fig 1E). A similar observation has been reported in postmortem T2-weighted MR imaging microscopy of brain stems. 54 To demonstrate clinical feasibility, we used the 1-average, 1mm isotropic FGATIR (theoretically increasing available intravoxel signal by $95%) to obtain images in a consented patient with prior right-sided infarct (Fig 6). Here we observed loss of the normal right corticospinal tract signal intensity in the cerebral peduncle and basis pontis, with complete absence of the right medullary pyramid. These findings were spatially remote from the site of ischemic stroke (right basal ganglia and internal capsule) and consistent with trans-synaptic degeneration of the tract.

DISCUSSION
Better discrimination of internal brain stem anatomy is critical for building knowledge about brain stem structure-function relationships in health, normal aging, and disease. 3,7,8 In current clinical practice, however, brain stem structure location can only be inferred indirectly on axial MR images on the basis of a priori knowledge of anatomy derived from histology (or postmortem MR microscopy), the superior-inferior position and surface landmarks, and limited internal contrast on clinical MR imaging. Current research atlases for imaging data like FreeSurfer 22,23 provide limited detail for the internal brain stem anatomy. Here, we identified brain stem anatomy using the FGATIR sequence, which uses a short inversion time to suppress signal from myelin. Previous groups using FGATIR (or similar contrast) emphasized applications for whole-brain segmentation or identification of diencephalic and spinal cord structures. [41][42][43][44][45][46] FGATIR not only enables the direct visualization of brain stem structures, but also the brain stem structures identified can be used for more precise indirect localization of additional bordering brain stem structures (eg, Fig 4). Internal brain stem contrast can be generated directly using FGATIR without the need for postprocessing, modeling assumptions, or ultra-high-field MR imaging systems. Contrast was consistent across multiple subjects and was feasible to obtain in patients with stroke.
Other MR imaging methods have provided some contrast resolution of internal brain stem anatomy, but are clinically or technically limited. Submillimeter isotropic resolution MR microscopy images can be obtained from isolated postmortem brain stem samples; [24][25][26][27] however, these protocols required 8-200 hours of scanning using ultra-high-field magnets and radiofrequency coils smaller than the human head, making this approach impractical for living subjects. Maps of diffusion parameters and tractography provide improved contrast resolution of internal brain stem anatomy. 36,39 However, diffusion MR imaging resolution of submillimeter brain stem pathways is challenging in vivo due to signal-tonoise limitations. Reliance on high-resolution diffusion MR imaging for anatomic localization can also be problematic due to inherent geometric distortion, particularly in structures near the skull base, and results in spatial misregistration with volumetric sequences. Unlike FGATIR, diffusion MR imaging contrast depends on complex mathematic representations or models that have proved challenging to validate in vivo. 40 FGATIR may provide complementary contrast for validating brain stem diffusion tractography or for independently extracting diffusion values from brain stem structures/regions. Joint reconstruction of multiple MR imaging contrasts, including FGATIR, may provide the best contrast resolution of brain stem anatomy. 55 The 1-mm isotropic FGATIR acquisition with 1 average ($12-minute acquisition) is a relatively long acquisition for clinical MR imaging protocols, but appears feasible in patients (Fig 6), particularly in specific clinical circumstances that justify such a long scan. For example, the FGATIR sequence was originally developed for functional neurosurgery to improve deep brain stimulator electrode placement within the subthalamic nucleus. 42 We plan to implement FGATIR in clinical patients to better localize specific brain stem structures affected by focal lesions (eg, infarcts, demyelinating, or inflammatory lesions). This sequence also may prove helpful for planning surgical biopsies or anatomic corridors for tumor resection. 12,13 The 800-mm isotropic data require 3-4 averages for useful contrast resolution, which may be susceptible to motionrelated image degradation in noncompliant individuals. This concern may be addressed by applying several recently developed methods to accelerate FGATIR acquisitions, including sliding window reconstructions of k-space, 56 compressed sensing image reconstruction of undersampled data, 57 and deep learning approaches to improve suboptimal image contrast produced by shortened scan times. 58,59 Clinical feasibility also can be supported through the use of a diffeomorphic template for coregistration of limited contrast There is asymmetric absence of the compact dark corticospinal tract (red arrows) in the middle third of the right cerebral peduncle of the midbrain (B), right central basis pontis (C), and right medullary pyramid (D). These images appear very similar to myelin-stained histology changes of the brain stem after chronic right-sided infarct published in a popular neuroanatomy atlas. 48 This clinical example illustrates the feasibility of detecting brain stem pathology in patients using the faster, but lower 1-mm isotropic spatial resolution.
FGATIR images from individual patients 60 or by training a neural network to recognize small brain stem structures with limited FGATIR contrast. 58 Using FGATIR to depict brain stem anatomy has other potential limitations. Not all brain stem structures were directly resolved by FGATIR. We identified brain stem structures on the basis of correspondences to labeled myelin-stained histology images in published atlases, 48-50 but direct structure-function correlations in the same brain would be helpful. Like diffusion MR imaging, FGATIR image contrast in the brain stem also could change with aging or disease. Such changes occur in other non-diffusionweighted scans of the brain stem such as neuromelanin MR imaging in subjects with Parkinson disease. 61 In our ongoing work, FGATIR-based contrast in older patients with stroke and essential tremor appears similar to that in age-matched healthy controls and younger subjects (unpublished data). However, the effects of aging and disease on FGATIR contrast require systematic investigation.
To date, the teaching, clinical targeting, and research of brain stem structures have largely relied on neuroanatomic atlases that have limitations. Most brain stem plates in the Haines atlas are from a single individual with trans-synaptic degeneration of the corticospinal tract due to internal capsule infarction. 48 The Schaltenbrand brain atlas, widely used for functional neurosurgery in the United States, relies predominately on 3 thalami of different sizes and limited spatial overlap. 62 This atlas also provided annotated images of only 1 brain stem in the axial plane. 17 MR imaging data with FGATIR contrast have some advantages over traditional neuroanatomy atlases that use postmortem histology. First, the 3D imaging datasets enable simultaneous localization and visualization of specific structures in multiple planes. Depiction of brain stem anatomy is not limited to the plane of section like histologic sampling, and specific oblique image reconstructions can be obtained to emphasize certain key spatial relationships. Second, collections of MR imaging data are less expensive and less time-intensive, providing greater opportunity to characterize brain stem structure variations across multiple individual brains. We intend to collect FGATIR data from a large cohort of subjects to create a diffeomorphic template with atlas labels 60,63,64 for the human brain stem. These data could be used to investigate size and shape differences in the brain stem and as an independent method for extracting quantitative data from MR imaging and PET data in future investigations of small brain stem structure changes associated with aging and different diseases. 65,66

CONCLUSIONS
The FGATIR sequence provided excellent novel contrast resolution of internal brain stem pathways and nuclei at 3T in healthy living subjects. The 1-mm isotropic resolution protocol was clinically feasible ($12-minute acquisition) and may immediately help with structure-function correlations or surgical corridor planning in the brain stem for individual patients. The FGATIR sequence also provided image contrast that may be helpful to complement and/or validate other advanced MR imaging methods for brain stem anatomy such as diffusion tractography.