HomeRadioGraphicsVol. 40, No. 2 PreviousNext Imaging PhysicsFree AccessInvited Commentary on “Optimizing Diffusion-Tensor Imaging Acquisition for Spinal Cord Assessment,” with Response from Dr Martín Noguerol et alSeth A. SmithSeth A. SmithAuthor AffiliationsVanderbilt University Institute of Imaging Science and Department of Radiology and Radiological Sciences and Department of Biomedical Engineering, Vanderbilt University Medical Center Nashville, TennesseeSeth A. SmithPublished Online:Mar 3 2020https://doi.org/10.1148/rg.2020190214MoreSectionsPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked InEmail MRI is a relatively young yet rapidly maturing diagnostic imaging tool that offers unique contrasts, allowing unprecedented exploration of tissues within the human body. This is especially true when considering the impact that MRI has on the understanding and evaluation of diseases of the central nervous system. However, the bulk of diagnostic MRI relies on the presence of water content for identifying lesions, infarcts, and tumors and imaging the central nervous system, and improvements have mostly targeted the brain. The spinal cord is critically important when evaluating diseases of the central nervous system, yet advanced quantitative MRI pulse sequences and contrast agents have lagged behind its brain MRI counterparts by at least a decade. Improving quantitative spinal cord MRI presents some challenges: the small size of the spinal cord (and the smaller size of its internal structures), constant motion, temporal and spatial field inhomogeneity, similar gray and white matter relaxation times, and long scan times. Recent studies, however, have shown that advanced quantitative MRI of the spinal cord is indeed feasible and approachable from a clinical perspective.When considering the scope of quantitative spinal cord MRI, the first technique that is often discussed is DTI because of the richness of information that diffusion MRI contains, the relationship of the derived contrasts to underlying biologic tissue constituents, its relatively well-understood analyses, and its clinical approachability. Although DTI is often first considered for clinical use, the current MRI literature is somewhat sparse on how to perform DTI in the human spinal cord in the clinic and what the considerations are for applying DTI in the spinal cord rather than the brain. This article by Martín Noguerol et al (1) explores the considerations for applying spinal cord DTI in the clinical evaluation of patients with various diseases and carefully outlines the challenges and the pulse sequences design processes that can improve DTI image quality in the spinal cord.The comprehensive article by Martín Noguerol et al (1) helps the radiologist develop the thought process one needs for optimizing DTI of the spinal cord through discussion of six topics: (a) physics and underlying principles; (b) artifacts; (c) technical adjustments and considerations, (d) alternative approaches, (e) analysis methods, and (f) clinical applications.Physical PrinciplesWhile the physical principles of DTI are the same for imaging the spinal cord as they are for imaging any nervous system structure, the authors nicely recap the basics of DTI and identify the areas where DTI acquisition methods can be modified for improved spinal cord assessment. They demonstrate the importance and biologic relevance of the derived indices such as FA, MD, AD, and RD. It is important to note that the pulse sequence and resulting SNR of the acquisition play a role in the consistency of the derived indices. In fact, the authors state that they “suggest using internal reference values… within the same patient or even acquiring a short reference cohort of DTI values obtained in healthy participants,” as different methods, vendors, and gradient systems can yield a range of resulting values.A challenge in evaluating the DTI-derived indices across sites, vendors, and sequences is the dependence of the diffusion tensor eigenvalues (and ultimately FA, AD, and RD) on the SNR, in which low SNR creates a bias (2,3) in the observed indices and must be carefully considered and potentially statistically disentangled in large studies.ArtifactsIt is often easy to consider that the spinal cord is an extension of the brain, and thus optimized brain acquisitions can be deployed in the spinal cord to generate high-quality data. The authors carefully outline the reasons why this is not the case. In fact, the best first step in optimizing spinal cord DTI is to consider that the spinal cord is not the brain and begin with understanding how to develop a high-fidelity acquisition for a small tubular structure. The authors reference three main artifacts that are encountered in the spinal cord to a greater degree than in the brain: (a) B0 inhomogeneity (4), (b) chemical shift artifact, and (c) motion artifact. With respect to B0 inhomogeneity, the authors focus on intrinsic or extrinsic static field inhomogeneities. However, two other considerations should be mentioned.First, the spinal cord sits inside a periodic spinal canal (composed of bone and disk), and thus static inhomogeneity also periodically varies. Also, especially in the thoracic cord, the B0 field changes with inhaling and exhaling, which can change the amount of geometric distortion (4). Regarding motion, the authors discuss the impact of coregistration of diffusion MR images and anatomic images as a concern, but it should be mentioned that cord motion between and during diffusion weighting can mix signals from CSF and the cord, leading to increased variance in the derived indices. Thus, cardiac gating, while it takes longer to perform, is advocated (5).Technical AdjustmentsWhen one sits at the scanner to implement spinal cord DTI, there are a few targets and technical adjustments that can be used for optimization: the b value, the number of gradient directions used, how the scan is set up (field of view, plane, and resolution), and physiologic gating and/or triggering. The authors carefully explore each of these and provide the reader with caveats for each modification. When considering how the scan is set up, the primary balance is struck with SNR and speed and coverage. It is important to maintain high SNR, as well as minimize the echo train in echo-planar imaging–based DTI sequences to reduce the impact of geometric distortion. In fact, many groups use a reduced field of view acquisition (6) to minimize geometric distortions while using the saved time to improve SNR. Two areas that bear further discussion are the b value and gradient direction sampling.For b values, the authors make a good point that with an increasing b value on clinical-grade systems, SNR is sacrificed. However, microstructural sensitivity may also be decreased with limited b values. Farrell et al (7) showed that higher b values in the spinal cord provided unique information about cord microstructure, although restricting the b value is clinically reasonable. The authors state that a b value of 400 sec/mm2 is sufficient to perform DTI reconstructions and generate reproducible parametric maps, but the reader should note that at such a low b value, sensitivity to pathologic lesions can also be hindered. Their recommendation of 800 sec/mm2 is consistent with the current literature.When considering gradient number and orientations, again we must abandon (to some degree) our lessons in the brain. In the brain, many gradient weightings are often chosen (8,9) to sufficiently characterize the fiber pathways that course in multiple directions. Since the spinal cord runs along a rostral-caudal axis, even considering that its curvature and shape may change in conditions affecting the spinal cord, fewer directions are considered reasonable. A recent study by By et al (10) demonstrates that there are diminishing returns after 15–16 gradient directions at a modest b value, agreeing with the authors’ suggestion of 16 directions as standard.Physiologic triggering and/or gating is an area that remains under discussion, and the authors provide the justifications for this controversy. The balance between scan time and motion is not trivial; it has been shown that cardiac gating can improve the quality of diffusion-derived indices (5). However, as the authors note, the increase in scan time can come at a cost of gross patient motion. In our particular laboratory, we perform cardiac triggering for our spinal cord DTI sequences because the inclusion of CSF into the tensor calculations decreases our confidence in the derived indices.Alternative ApproachesOften when readers examine an article that focuses on optimizing DTI (or any other MRI pulse sequence) for use in the spinal cord, they are quickly met with a list of impediments to spinal cord imaging: the spinal cord is very small, it moves during and between acquisitions, conventional readout schemes may not provide the best image quality, the SNR is low, etc. However, the reader should be encouraged that there have been studies conducted on how to best achieve spinal cord DTI within the given constraints of current software and hardware in a reasonable scan time. Recommendations from Martín Noguerol et al (1) will be provided below and serve as a strong summary.What I found encouraging is the reminder that no MRI method is stagnant and there are constant improvements to methods, hardware, and software that can be harnessed for imaging the spinal cord. Indeed, in this article, we are reminded of the recent improvements in readout schemes that can drastically impact the quality of spinal cord DTI. While we recognize that some are not clinically standard today, they may be tomorrow. Here we can see that mitigation strategies to susceptibility-related artifacts, spinal cord motion, and geometric distortions can be achieved through FSE, PROPELLER, and segmented echo-planar readouts. PROPELLER has the added benefit of being so-called self-navigated, so intrascan motion is minimized. Segmented echo-planar imaging methods offer a unique opportunity to increase the resolution without sacrificing the echo time and SNR as with single-shot echo-planar imaging, as long as the phase errors are corrected.Analysis MethodsAs with any calculated contrast, there are different tools for deriving the contrast or desired outcome. Recently, spinal cord–specific analysis tools have become available, such as the Spinal Cord Toolbox (11), which is freely available and well-used in the community, yet we (and the authors) recognize that there are many tools available. What I found most important to highlight were two comments by the authors (1) which are quoted here and serve as the foundation for any and all analysis methods: “Estimations of the parameters derived from a DTI acquisition may vary significantly depending on… the number of motion-probing gradient directions, the b values used, and the strength of the magnetic field. In addition, variability and bias can be introduced as a result of differences in the measurement method used….” With regard to fiber tracking, the authors remind the user that “Although the final images can often be quite impressive, it is important to understand that these reconstructions are highly variable and depend highly on the preset values, as well as the chosen postprocessing software. In addition, the radiologist must be aware that the number of fibers represented does not reflect the actual number of spinal cord tracts.”Clinical ApplicationsPotentially, the most impressive aspect of this work is demonstrating that when high-quality DTI is used, direct clinical impact can be realized. It is an important reminder that time spent optimizing spinal cord DTI is not in vain, and that, in fact, DTI has a rich application portfolio across various spinal cord pathologic conditions. While this commentary cannot highlight each condition described in the article, the authors have provided an excellent survey of where DTI can be used and when it is usable and useful. We do summarize that the application of DTI in the spinal cord can be considered in myelopathies, degenerative diseases, traumatic and nontraumatic cord compression, and primary and secondary cancers. In short, the authors have shown that if the spinal cord is involved in a pathologic lesion, DTI can (and should) be considered as a method to derive greater biologic information from the tissue, which may be useful for assessing treatment efficacy, progression, or improvements.In summary, “Optimizing Diffusion-Tensor Imaging Acquisition for Spinal Cord Assessment: Physical Basis and Technical Adjustments” (1) is as comprehensive of a spinal cord DTI article as currently exists in the literature. While some of the recommendations are not always agreed on universally, it provides a great starting point for conversation in one’s own laboratory or clinical practice. Axial single-shot echo-planar imaging with a b value of ∼800 sec/mm2, 15–16 gradient directions sampled uniformly about a sphere, higher-order shims, spectral fat saturation, cardiac gating (if time allows), and parallel imaging and/or reduced field of view to reduce echo train length is considered an appropriate starting point for many clinical and research studies for spinal cord DTI and is well highlighted by this article.