Diffusion tensor imaging the medial longitudinal fasciculus in INO: opportunities and challenges

The medial longitudinal fasciculus (MLF) is a white matter pathway in the brainstem that plays a key role in coordinating eye movements. Injury to the MLF leads to abnormalities in eye movements that can be measured with high precision by oculography, making it an ideal eloquent pathway to study imaging/function correlates. Tractography is an emerging method for identifying white matter pathways and offers the tantalizing promise of noninvasive, quantitative characterization of tissue integrity underlying functional deficits. However, the small caliber of the MLF and partial volume averaging with signal from nearby cerebrospinal fluid pose severe technical challenges to tractography‐based delineation of the MLF. We discuss progress toward the goal of imaging the MLF and potential benefits of achieving this goal. Initial work suggests that ultra‐high field (7 tesla) may complement tractography for characterizing the MLF.

Keywords: tractography; medial longitudinal fasciculus; ultra high field; diffusion tensor imaging; brainstem; internuclear ophthalmoparesis

The medial longitudinal fasciculus (MLF) consists of a pair of nerve pathways in the periventricular brainstem tegmentum and plays a central role in carrying signals associated with the coordination of all classes of eye movements. Injury to the MLF frequently occurs as a consequence of multiple sclerosis (MS), leading to clinical deficits in eye movements such as internuclear ophthalmoparesis (INO).[1] These deficits are a derivative of injury to the MLF and can be measured with high precision by infrared oculography.[2] Therefore, injury to the MLF may serve as a model system for understanding injury and recovery of white matter pathways.

While clearly some MS INO patients are asymptomatic, the clinical concomitants of this syndrome are manifold and can include diplopia (typically horizontal binocular but occasionally vertical; especially when associated with skew deviation from otolithic dysfunction within the MLF), visual confusion, the illusion of environmental movement during horizontal saccades, vertigo, and "blurring" of visual image acuity, particularly with reading. A less conspicuous, but potentially dangerous manifestation of INO includes worsening dysconjugacy and a break in binocular fusion during active head turning (which produces a contraversive slow phase, punctuated by saccades in the direction of head movement), while head turning during driving (e.g., with changing lanes), and while body/head turning during ambulation. In the latter circumstances, patients may be more prone to traffic accidents and falls respectively.[3] In fact, we have interviewed many MS patients with INO who initially believed they were asymptomatic in terms of ocular motor dysfunction until we queried them about problems with reading, driving, and turning while walking. In many circumstances, we found that the particular etiological rudiments of their dysfunction was in fact related to their INO, and secondary to appendicular weakness, motor integration, proprioceptive deficits, or other neurological dysfunction.

Several pathological studies have confirmed the anatomical relationship between the presence of lesions along the ipsilateral MLF and the presence of INO.[[4]] However, the in vivo investigation of structure‐function relationships of this system presents some technical challenges, which are related to the small sizes of the structures involved, and a series of factors that will be characterized herein. Due to its high spatial resolution, magnetic resonance imaging (MRI) has allowed us to depict in vivo, the anatomic organization of the human oculomotor nerve complex, the MLF, and related structures in the brainstem. In one study involving 58 MS patients with INO, MRI with proton‐density imaging detected a higher percentage of MLF involvement (100%) than T2‐ (88%), fast‐FLAIR (48%) and T1‐weighted (2%) sequences.[9]

Despite the frequency and conspicuity of MLF lesions on conventional MRI in MS patients with INO, lesions appear similar, despite their heterogeneous histopathological substrates (demyelination, remyelination, gliosis, etc). In order to better understand the relationship between the type of tissue injury, repair, and neuroprotection, we require non‐conventional imaging methods that can more precisely yield signatures of tissue injury that relate to the functional consequences of the clinical disability.

Diffusion tensor imaging (DTI) is an MRI based methodology that noninvasively characterizes the anisotropy of water diffusion.[10] As the degree of anisotropy correlates with axonal integrity and myelin status, DTI has demonstrated promise for quantifying tissue injury and response to treatment.[11] Furthermore, tractography exploits the directionality of this anisotropy to map white matter fascicles,[12] which are otherwise indistinct on imaging.[13] Close study of the MLF by DTI and oculography provides a context for in vivo validation of DTI in humans, thereby complementing previous studies performed in animal models.[11] A small, pilot study by our group suggested that DTI measures within the MLF correlate with functional impairment as measured by infrared oculography, but these correlations were limited by small sample size and limitations in DTI acquisition and imaging coregistration.[14]

A number of technical challenges impede progress toward reliable DTI and tractography of the MLF. First, the caliber of the MLF, on the order of 2 to 2.5 mm in cross‐sectional diameter, is small compared to the typical spatial resolution in DTI, typically 2 mm isotropic voxels at 3 tesla. Second, large signals from nearby cerebrospinal fluid (CSF) in the fourth ventricle leads to a partial volume averaging artifact and concomitantly low fractional anisotropy values. In addition, the large number of parallel tracts in this region often leads to "track jumping" within the tractography analysis. In this contribution, we detail progress toward clearly identifying and eventually overcoming these principal obstacles.

One healthy subject was examined under an Internal Review Board‐approved protocol. Imaging was performed on a 3 tesla Phillips Achieva (Phillips Medical Systems, Best, the Netherlands) with a standard 8‐channel head coil. Anatomical T2‐weighted images (512 × 512 × 32 matrix, 230 × 230 × 64 mm FOV, TE/TR = 80/3000 ms) and DTI with cardiac triggering to minimize pulsatile artifacts (128 × 128 × 20 matrix, 256 × 256 × 40 mm FOV, interpolated to 1 × 1 mm in‐plane resolution, TE = 65 ms. TR ≈ 18 ms, varying according to heart rate, 30 non‐collinear diffusion gradients with b = 700 sec/mm2 and one b = 0 acquisition, NEX = 2) was acquired with scan planes perpendicular to the posterior edge of the brainstem in the midsagittal plane, covering the entire length of the brainstem. Triggering used a fingertip pulse monitor, limiting acquisitions to 1 slice per heartbeat, 300 ms after detection of the peak of the pulse.[15] Gradient echo images were acquired on a 7 tesla Phillips Achieva whole‐body scanner (480 × 381 × 20 matrix, 240 × 240 × 60 mm FOV, interpolated to 0.5 × 0.5 mm in‐plane resolution, TE/TR = 424/12 ms) to determine the potential of ultra high field imaging for identifying the MLF.

Initial postprocessing of the DTI data consisted of motion correction, necessitated by the relatively long duration of the DTI scans, tensor calculation, and streamline tractography. Motion correction of the DTI was performed with an iterative algorithm.[16] The diffusion tensor and streamline tractography was calculated using the Diffusion Toolkit[17] generating streamlines throughout the entire brain, with one seed in each imaging voxel. Correction for partial volume effects from CSF was performed using a constrained bi‐tensor model developed by Pasternak et al.[18]

Selection of tracks representing the MLF required constraints informed by anatomical priors. A subset of tracks was selected from the whole‐brain set of streamlines by choosing only those that intersected hand‐drawn regions of interest (ROIs). Placement of the ROIs was guided by a model of the MLF generated from the Schaltenbrand‐Wahren stereotactic atlas.[19] The model was generated by hand‐drawing the MLF on scanned images from the atlas, affine transformation of the scanned images to the anatomical T2‐weighted image in‐plane using MATLAB (The Mathworks, Natick, MA, USA), and subsequent coregistration using the FLIRT tool of FSL.[20] ROI placement was performed with Trackvis.[17] Visualization was performed using Trackvis[17] and FSL.[21]

Tracking of the MLF is particularly sensitive to constraints. For example, Figure 1 shows that fiber tracks (yellow) intersecting an ROI at the inferior tip of a volume derived from the Schaltenbrand‐Wahren atlas (magenta) bend away from the MLF, following the tectospinal tract. Fiber tracks (orange) intersecting an ROI at the superior tip are more consistent with the overall course of the MLF, but include outliers that extend out of the brainstem. The background is a sagittal view of a colorized fractional anisotropy map.[22] Red, green, and blue correspond to fiber orientations in the left/right, anterior/posterior, and inferior/superior directions, respectively. Also of note is the irregular appearance of the atlas‐derived volume, which partly results from the coarse and irregular spacing of image slices and difficulty in accurate coregistration of the atlas images to anatomical MR images. Large differences in slice plane orientation and shrinkage of tissue during histological preparation when generating the Schaltenbrand‐Wahren atlas may also be contributing confounds.

MAP: 1 Sagittal view of a colorized fractional anisotropy map and a volume representing the medial longitudinal fasciculus (MLF) as determined from the Schaltenbrand‐Wahren atlas (magenta). Two sets of tracks have been selected using regions of interest (ROI) placed at the inferior and superior tips of the atlas‐derived regions (yellow and orange, respectively).

A close examination of the fiber modeling on a voxel‐by‐voxel basis provides some insight into the source of the sensitivity of tracking to the constraints. Figure 2 shows a rendition of the modeled fiber orientations (short line segments) on a voxel‐by‐voxel basis in the midsagittal plane. Tractography essentially works by interpolating between the fiber orientations indicated at neighboring voxels. The red region indicates the MLF derived from the Schaltenbrand‐Wahren atlas while the yellow region indicates the location of the fiber tracks intersecting an ROI at the inferior tip of the atlas‐derived region. Orange indicates the region of intersection. One can see an overall "forking" organization of the voxel‐by‐voxel fiber orientation that drives fiber tracks anterior if constrained by the inferior ROI while allowing more or less vertical tracks if constrained by superior ROI.

Graph: 2 Midsagittal rendering of fiber orientations on a voxel‐by‐voxel basis (line segments) with the MLF region coregistered with the Schaltenbrand‐Wahren atlas (red) and volume occupied by a fiber track intersecting the inferior tip of the atlas region (yellow).

Accounting for partial volume effects due to CSF in neighboring ventricles can have a quantifiable effect on measures of water diffusivity and anisotropy. These measures are of interest because they are quantitative surrogate markers for aspects of tissue injury such as axonal fragmentation and demyelination. The constrained bi‐tensor model accounts for a partial volume of CSF, leading to an increase in FA in the range of several percent in periventricular spaces (arrow, Fig. 3). As the MLF is adjacent to the fourth ventricle along much of its length, and the number of voxels occupied by the MLF is small, even subtle changes in measured anisotropy may have a significant impact on DTI‐based assessment.

Graph: 3 Fractional anisotropy (FA) maps without (center) and with (right) CSF correction. Higher FA is found in periventricular regions (arrow) after the correction. Grayscale FA image (left) is shown as a guide to the anatomy.

The difficulties associated with tractography and atlas‐based methods for pinpointing the MLF suggest that alternative strategies should be examined. Ultra high fields of 7 tesla and higher offer the opportunity for extremely high spatial resolution. Contrast due to variations in magnetic susceptibility indicate texture consistent with white matter fascicles.[23] Preliminary work at 7 tesla indicates the presence of hypointensities at a location consistent with that expected for the MLF (Fig. 4). However, imaging at ultra high field is subject to signal dropout due to limited coil coverage and magnetic susceptibility effects that are apparent in anterior regions of the brainstem. These artifacts also have a deleterious effect on DTI in general, making a wholesale shift to 7 tesla problematic.

Graph: 4 Gradient echo image acquired at 7 tesla indicate hypointensities consistent with the MLF (red arrow). Inset shows magnified view of brainstem region. Significant signal dropout is a consequence of a magnetic susceptibility artifact that is more severe at ultra high fields than at lower fields.

The MLF is a critically important conduit for many brainstem pathways, and constitutes the final common pathway for all major classes of conjugate eye movements including saccades, smooth pursuit, and vestibulocular reflexes, including semicircular and otolith mediated ocular motor reflexes. The six ocular motor nuclei (pairs of cranial nerve III, IV, VI) are interconnected via the MLF, which transmits vital information for the purpose of achieving synchronization and conjugacy of eye movements to a visual target, resulting in foveation, binocular fusion, and stereopsis. Within this system are both the excitatory as well as reciprocal inhibitory projections that serve to precisely regulate the interplay between agonist and antagonist muscles of the eyes.

The juxtaposition of brainstem eye movement circuitry to the ventricular system (IV ventricle and cerebral aqueduct) and the ability to utilize high precision oculography to objectively characterize changes in eye movement metrics, renders these white matter tracts as potentially ideal candidate targets for the application of neuroprotective and even neurorestoration strategies for MS. In contrast to CNS long tract systems (e.g., corticospinal tract), confirmation of therapeutic benefit derived from protective or restoration therapy may be more rapidly appreciated with smaller targets such as the MLF.

We now have the capability to study the relationship between quantitative measures of dysconjugacy, and nonconventional neuroradiologic abnormalities within the dorsomedial brainstem tegmentum, in the region of the MLF. In particular, INO is an eloquent and common MLF related syndrome in MS that represents a useful model by which to objectively characterize a distinctive neurological disability, with associated imaging measures of brain tissue injury.

Previous tractography‐based work has primarily focused on major pathways such as the corticospinal tract.[[24]] However, the brainstem contains numerous fascicles besides the MLF, with small cross sectional areas that are functionally critical. Even in large pathways, significant user intervention during image analysis may be necessary to eliminate physiologically implausible tracks.[26]

Although the work presented here is promising, we have yet to determine the reliability of the methodology, particularly in the presence of a multifocal disease process such as MS. Atlas‐based methods are a potential alternative.[[14], [27]] However, atrophy associated with diseases such as multiple sclerosis may hinder accurate coregistration between atlas and in vivo images. Ultra high field imaging may also provide important information. Previous work has demonstrated agreement between atlases and ultra high field imaging,[29] but the MLF was not specifically indicated. Notwithstanding the remaining technical challenges, if the MLF can be precisely identified at ultra high field, such images may serve as an important reference point for future DTI investigations aimed at elucidating both the pathobiology of disease, and the formulation of novel treatment strategies. Advances in DTI at ultra high field may also allow measurements of tissue integrity at higher spatial resolution than is currently achievable at 3 tesla.[30]

The challenges and potential solutions outlined here may be applicable to the evaluation of other small fiber pathways, including those adjacent to CSF. An advantage in studying the MLF is its eloquent functional expression in conjugate lateral eye movements and targeted injury in patients with MS. Other small pathways could be similarly interrogated using DTI, and their functional correlates evaluated. Despite its small size, the MLF is relatively straight along its brainstem course, which improves the ease of tractography. Fiber tracts which curve or change direction over their trajectory present significant challenge to tractography and would need additional attention to identify accurately and consistently.

Progress continues toward the development of a methodologic paradigm for precise, localization of the MLF across the caudal‐rostral extent of its trajectory. Achievement of this goal remains a formidable technical challenge. Nevertheless, high precision identification of the MLF, and other small pathways in the brainstem, particularly those associated with corresponding eloquent syndromes (as in the case of INO for the MLF), may provide a crucial test bed for validating DTI metrics of tissue injury, tracking the evolution of specific neurological disorders, and for corroborating the therapeutic effects of neuroprotective and neurorestorative agents.

We gratefully acknowledge funding from NMSS (grant RG 4091A3/1) and the DAD's Foundation.

Ivan Dimitrov is an employee of Philips Medical Systems. Gina Remington has been a speaker for NMSS, MSAA, IOMSN, Biogen and Teva. Teresa Frohman receives speaker fees from Biogen and Teva. Elliot Frohman receives speaker fees from Biogen, Teva, and Acorda and consulting fees from Biogen, Teva, Acorda, Novartis and Abbott. Robert Fox has received consulting fees from Avanir, Biogen Idec, EMD Serono and Novartis. In addition, research support, consultant and advisory committee fees from Biogen Idec were paid to Cleveland Clinic.