Magnetic Resonance in Medicine 69:144–149 (2013)

Fast Diffusion Tensor Imaging and Tractography of theWhole Cervical Spinal Cord Using Point Spread FunctionCorrected Echo Planar Imaging

Henrik Lundell,1,2,3* Dorothy Barthelemy,1,2,4 Fin Biering-Sørensen,5 Julien Cohen-Adad,6

Jens Bo Nielsen,1,2 and Tim B. Dyrby3

Diffusion tensor imaging has been used in a number of spinalcord studies, but severe distortions caused by susceptibilityinduced field inhomogeneities limit its applicability to investi-gate small volumes within acceptable acquisition times. A wayto evaluate image distortions is to map the point spread func-tion of the voxel intensity in a reference scan. In this study, thepoint spread function was mapped for an echo-planar imagingsequence in the human cervical spinal cord with isotropic res-olution and large field of view. Correction with the point spreadfunction map improved anatomical consistency, and full cervi-cal tractography was thereby possible from a C1 seed region inhealthy controls and one individual with spinal cord injury. It issuggested that point spread function mapping of the spinal cordcan be used in combination with sequence-based methods forreduction of susceptibility artifacts or in high-field imaging set-tings where off-resonance effects are pronounced. Magn ResonMed 69:144–149, 2013. © 2012 Wiley Periodicals, Inc.

Key words: spinal cord; diffusion tensor imaging, tractography;point spread function

Over the last decade, the value of Diffusion Tensor Imag-ing (DTI) as a method for detecting changes in spinal cordwhite matter has been manifested in a number of studies(1). DTI provides both sensitive markers for microstruc-tural changes and estimates of local axonal fiber direction,which can be used to map global connectivity by meansof tractography (2,3). This has become an important toolfor evaluation of brain networks, but applying the sametechniques on the spinal cord is less straight forward. Due

1Department of Exercise and Sport Sciences, University of Copenhagen,Copenhagen, Denmark2Department of Neuroscience and Pharmacology, University of Copenhagen,Copenhagen, Denmark3Danish Research Centre for Magnetic Resonance, Copenhagen UniversityHospital Hvidovre, Hvidovre, Denmark4School of Rehabilitation, Faculty of Medicine, Université de Montréal,Montreal, Québec, Canada5Clinic for Spinal Cord Injuries, Rigshospitalet and Faculty of Health Sciences,University of Copenhagen, Denmark6A. A. Martinos Center for Biomedical Imaging, Massachusetts GeneralHospital, Harvard Medical School, Charlestown, Massachusetts, USAGrant sponsors: Elsass Foundation, Denmark, Canadian Institutes of HealthResearch (CIHR), Lundbeck Foundation, The Future and Emerging Technolo-gies of the EU FP7*Correspondence to: Henrik Lundell, Ph.D., DRCMR, 340B, Copenhagen Uni-versity Hospital Hvidovre, Kettegaards Allé 30, Hvidovre DK-2650, Denmark.E-mail: lundell@drcmr.dkReceived 27 June 2011; revised 9 January 2012; accepted 3 February 2012.DOI 10.1002/mrm.24235Published online 6 March 2012 in Wiley Online Library (wileyonlinelibrary.com).

to severe susceptibility-induced geometrical distortions inthe acquired diffusion-weighted MRI data, previous humanstudies have only traced fibers over few vertebral lev-els (4–7). Longer segments are sometimes rendered usingseed regions covering several levels or the full length ofthe spinal cord, but individual reconstructed tracts usingthis approach rather reflect local anisotropy in the seedregion than the true trajectories of the anatomical pathways(8–12).

Point spread function (PSF) mapping is a technique forcorrecting such distortions by measuring the displacementand dispersion of signal intensities in each voxel (i.e., thePSF) in a separate reference scan applied prior to the MRIsequence (13–15). The echo-planar imaging (EPI)-PSF map-ping sequence adds an additional phase-prewinder priorto the conventional EPI readout, which is incrementedover a number of repetitions. Effectively, this introducesan additional k-space dimension that samples each phase-encoding line at different stages of phase accumulation.This independent k-space dimension can be transformedinto the voxel PSF (14).

The aim of this study was to investigate, if the PSF couldbe effectively measured in the cervical spinal cord andused for correction of diffusion-weighted EPI. To demon-strate the usability of this technique, sequence settingswere adjusted to provide DTI data with full cervical cov-erage and isotropic voxel size within a short acquisitionof about 5 min that would be suitable for routine clinicalstudies.

METHODS

Subjects

Six healthy male controls (age: 22, 27, 28, 44, 45, and61 years) and one individual with chronic incompletetraumatic spinal cord injury (SCI) (age: 46 years) volun-teered for the experiment. All experimental procedureswere approved by the Ethics Committee of the CapitalRegion of Copenhagen, Denmark, and in compliance withthe Declaration of Helsinki, and undertaken with eachparticipants informed consent.

MRI Protocols

The scans were performed on a Simens Trio 3T scanner(Siemens, Erlangen, Germany) using the eight cervical coilelements of a 16-channel neck and spine phased arraycoil. Diffusion-weighted EPI: a sagittal 128 × 128 × 15matrix was reconstructed from an accelerated acquisition

© 2012 Wiley Periodicals, Inc. 144

Spinal Cord PSF Mapping and DTI 145

with GRAPPA acceleration factor of 2 and 32 autocalibra-tion signal lines along the anterioposterior phase encodingdirection (16). Voxel size was set to isotropic 2×2×2 mm3.The EPI readout train was 51 ms, and partial Fourier encod-ing was not applied. Diffusion weighting was applied witha twice refocused spin-echo with bi-polar gradients toreduce eddy current effects (17) with pulse repetition time= 3000 ms and echo time = 90 ms. Sixty-one uniformlyspaced gradient vector directions were used with diffu-sion weighting factor b = 500 s/mm2 (18). Five additionalb = 0 images were recorded, and the total acquisition timewas 3 min 18 s. PSF mapping sequence: EPI parameters andfield of view (FoV) were identical to the b = 0 diffusion-weighted EPI, and the PSF encoding was performed with32 increments of the additional phase prewinder with areduced PSF FoV factor of 4 to speed up the acquisition(14). A larger set of increments, i.e., a smaller PSF FoV fac-tor, could improve SNR and would allow for broader ormore displaced PSF peaks over the full length of the phaseencoding FoV dimension, but this was not needed for thisapplication. The acquisition time for the PSF-EPI sequencewas 1 min and 36 s. The EPI and PSF FoV covered from thebrain stem to ∼ T4/T5. Structural T2-weighted (T2W) 2Dspin-echo sequence: a distortion-free structural image wasrecorded as reference with sagittal orientation, in plane res-olution of 0.88 × 0.88 mm2 and slice thickness of 3.3 mm.pulse repetition time = 4000 ms, echo time = 113 ms, andflip angle = 90◦.

Model-Based Registration

The diffusion-weighted dataset was coregistered in twosteps: first to assure alignment within the diffusion datasetand second to assure alignment between the diffusiondataset and the PSF dataset for the subsequent distortioncorrection. The first coregistration step was done using amodel-based registration algorithm (19). Each diffusion-weighted image was registered to a corresponding syn-thetic diffusion-weighted image inversely calculated fromthe diffusion tensor of the unregistered dataset using theCamino toolbox (20). The diffusion tensor thereby indi-rectly worked as a common reference for all images, butthe individual images were registered to a reference witha similar contrast. 3D rigid body coregistration was per-formed in SPM5 (http://www.fil.ion.ucl.ac.uk/spm) using anormalized mutual information cost function. The processwas iterated with the registered dataset to assure conver-gence. The second step was done by first coregisteringthe synthetic b = 0 image to a distorted EPI recon-structed from the PSF mapping sequence and then applyingthe transformation matrix to the whole diffusion-weighteddataset.

PSF Distortion Correction

The PSF was reconstructed, and the peak displacement,the standard deviation (SD) across the phase prewinderincrements (which relates to the width of the PSF) and dis-torted and undistorted EPI references were estimated aspresented earlier (14). Regions outside the spinal cord and

CSF were masked out of the PSF peak displacement map toexclude noisy background tissue. To achieve a continuousmap, the displacement at the mask border was extrapo-lated to zero displacement at the image border along eachphase encoding line. To reduce the impact of noise, the mapwas then smoothed using a 3D gaussian kernel with 5-mmFWHM. The distorted diffusion dataset was unwarped andresliced with the masked PSF-displacement map and thepreviously estimated coregistration parameters using thirddegree B-spline interpolation in one dimension along thephase encoding direction. The correction algorithm wasimplemented in Matlab (Mathworks, Inc., Navick, MA).

DTI Analysis

The diffusion tensor was fitted using linear regression(21). Deterministic streamline tractography was performedusing a fixed step size of 0.1 mm, anisotropy threshold wasset to fractional anisotropy (FA) = 0.2, stepwise angularthreshold was 45◦, and streamlines shorter than 100 mmwere subsequently excluded (22). These settings were cho-sen to capture long white matter pathways and reduce theimpact of nerve roots or other anisotropic structures inthe axial plane such as the collateral fibers in white mat-ter or commissural fibers in gray matter. DTI metrics wereextracted from these remaining tracts. Seed regions weredrawn in the b = 0 images over the full cross-section ofthe spinal cord in one plane at C1. The seed region ofthe SCI participant was divided into a left and right partover the medial line, and tractography was performed fromthese two regions individually. Streamline data and seedregion coordinates were exported to Paraview (Kitware,Inc., Clifton Park, NY) for 3D rendering (Figs. 2 and 3). Thesuperior–inferior coordinates for the cervical interverte-bral disks were manually detected in each participant, andtracts were warped onto a common rostro-caudal referenceframe as suggested for spinal fMRI data (23).

RESULTS

Model-Based Registration

Visual inspection was performed to check alignment beforeand after correction, and improved correspondence wasachieved in all datasets. Registration corrected for dis-placements up to ∼1.5 mm. In each dataset, convergencewas reached after one iteration only at a residual level of0.2 mm.

PSF Mapping and Correction

Small scale distortions around intervertebral disks as wellas the larger scale distortions toward the end of the cervicalregion were visible in the PSF peak data with shifts up to20 mm (Fig. 1a). Corrected images showed consistency withthe distortion-free structural image, both for short scale dis-tortions around intervertebral disks (arrows in Fig. 1a,d)and for larger scale displacements (crosses in Fig. 1b–d).Distortion-free EPI references reconstructed from the mul-tishot PSF data showed some minor artifacts especiallyproximal to the brainstem probably due to CSF-flow (seewhite arrows in Fig. 1e). Level-wise variations were found

146 Lundell et al.

FIG. 1. a: The mean PSF peak displacement of A–P coordinates (phase encoding direction) in a line through the spinal cord in one participant.The vertical axis correspond to the same superior–inferior positions in (a)–(d). The C1–T4 vertebral levels are marked along the vertical axisof (a). b: Structural undistorted T2W reference. c: Original uncorrected EPI. d: Same slice as in (c) but corrected with the PSF map. Thecrosses correspond to the same location in space in the scanner coordinate system for the three images. Arrows indicate intravertebraldistortions corrected in (d). e: Undistorted multishot images of the PSF sequence from one healthy and the SCI participant. Minor flowartifacts are visible at the level of the brain stem (white arrows). f: SD images show increased PSF variation at intravertebral levels (blackarrows).

in the PSF SD images data with increasing values in bandsat intravertebral levels.

Tractography

Data corrected using the PSF produced consistent tractsthroughout the whole cervical spinal cord in all healthycontrols and extended in some cases throughout the FoV,i.e., approximately to level T5 (Fig. 2a,c). In contrast, recon-structed tracts from uncorrected data were truncated atlevels C3–C4 in all participants due to large image distor-tions (Fig. 2b). Below the cervical region, inconsistencieswere observed around T2 in all corrected tracts (whitearrow, Fig. 2c). The participant with a traumatic SCI hada hyperintense lesion in the T2W image at C5/C6 (Fig.3a). Tracts went through C5/C6 levels only on the rightside, which also corresponded to the participants func-tionally least impaired side (Fig. 3b). Tracts on the leftside terminated because of FA values below the chosenthreshold. Figure 4 shows the mean FA for the healthy con-trols overlaid with the FA for the SCI participant with theleft and right side plotted separately. The mean FA val-ues of the healthy controls were constant across cervical

levels, but small variations of approximately ±5% wereobserved throughout the cervical cord. In the SCI partici-pant, decreased FA values were observed from mid bodylevel of C3, i.e., a larger extent than the visible lesion in theT2W image.

DISCUSSION

In this study, we showed the feasibility of using PSFmapping for correction of large distortions in spinal corddiffusion-weighted EPI data in combination with a model-based registration framework for correction of subjectmotion. Corrected images showed improved correspon-dence to structural images with low distortions, and wewere able to perform deterministic tractography from C1throughout the cervical spinal cord of healthy controls. Wealso demonstrate the method in one individual with SCI.

Correction and Acquisition Strategies

The EPI settings used in this study, i.e., large FoV andlong readout time, were set to provide large volume cov-erage and high SNR within a short acquisition time. This

Spinal Cord PSF Mapping and DTI 147

FIG. 2. Deterministic tractography with seed region in C1 for theparticipant shown in Fig. 1b b: Uncorrected dataset and c: correcteddataset in 3D view overlaid with a mid-sagittal FA map. The tractsare seeded in the red plane. Typical disruption at ∼T2 is marked withwhite arrow. a: Corresponding structural T2W image.

results in a high level of geometric distortion that could bereduced using an alternative acquisition strategy to reducethe amount of distortion correction that is required. Severalmethods for reducing distortions at the acquisition stage,mainly based on readout time reduction, may be used incombination with a PSF based correction (24–29).

Some artifacts were visible in the distortion-free EPI ref-erence reconstructed from the multishot PSF data probablyrelated to CSF flow or motion, see Fig. 1e. Those were dom-inant in the brainstem region where flow patterns might bemore complex compared to lower levels where flow is morelaminar and perpendicular to the anterior–posterior (A–P) phase encoding. Similarly, more turbulent flow mightoccur in lesioned areas, and this could also give rise to arti-fact. Cardiac gating may potentially solve this problem andcould reduce signal attenuation because of dispersive flowin the same regions. The SD image, shown in Fig. 1f, had ingeneral low values in the spinal cord, but striking patterns

FIG. 3. Data from the SCI participant. a: Structural undistorted T2Wreference. Lesion visible at C5/C6. b: 3D view of tracts from the seedplanes on the left (red) and right (green) sides. Pathways bypassingthe lesion are found on the right part of the spinal cord [green arrowin (b)].

FIG. 4. Mean tract FA for healthy control group (blue solid lines ±SD,blue dotted lines min/max) and SCI participant mean tract FA for left(red) and right (green) portions of the spinal cord shown in Fig. 3b.Left tract terminates at C5.

of bands with higher values were seen across the levels ofthe intravertebral disks (black arrows in Fig. 1f). This mir-rors large field inhomogeneities over the voxel size causingdecreased T2*-relaxation and thus lower the effective res-olution, i.e., a broader PSF peak, which could explain thesomewhat uneven intensity seen in the corrected images(Fig. 1d).

For future spinal cord applications in ultra high-fieldsettings where distortions and relaxation effects are moresevere, PSF mapping could be of particular interest. In ourcorrection, we took the displacement of the PSF peak intoaccount. However, the full shape of the PSF may be usefulfor deconvolution of voxel dispersion because of relaxationeffects or for a general assessment of image quality (30,31).Image blurring is also introduced by the reslicing proce-dures and could be further suppressed by higher orderinterpolation methods or, if possible, increased originalresolution. Furthermore, the PSF method is not limited tothe phase encoding direction but could be applied to anydimension in sequences with cartesian encoding (13).

Tractography and DTI

Tractography was used in this study to evaluate the relationbetween the reliability of global geometry after correc-tion in relation to local fiber direction estimates. A simpledeterministic diffusion tensor tractography algorithm wasdeliberately chosen to demonstrate consistent streamlineprogression in the main fiber pathways, which should bea prerequisite for a trustworthy use of more elaborate,for instance, probabilistic methods (32). High streamlineconsistency achieved from simply defined seed regionsalso opens up for standardized and automated analysis forlarger studies and for extraction of quantitative tract dataas shown in Fig. 4. A more general issue in DTI analysis ispartial volume effects induced by FoV position, tract size,or atrophy. Especially, CSF contamination in white mattervoxels is dominant in the spinal cord setting and atrophy in

148 Lundell et al.

the SCI individual could explain the FA decreases observeddistal to the lesion. FLAIR preparation or multiple b-valuescould here be used for suppression or estimation of theCSF partial volume component (33). Partial volumes couldalso be corrected for in voxel based analysis given relevantmorphological data (34). Morphological methods measur-ing regional atrophy in the spinal cord has recently beenintroduced and has been shown to be sensitive to specificfunctional qualities in individuals with SCI (35,36). Thevalue of combining multiple MRI-based methods, e.g., dif-fusion, magnetization transfer, and atrophy measures hasalso recently been demonstrated (37). In this context, robustDTI tractography methods like the method proposed in thisarticle could provide both diffusion data and anatomicallyrelevant tracts.

We observed a variation in anisotropy over the length ofthe vertebral elements throughout the spinal cord. This wasnot evident on single subject data but apparent after aver-aging the warped tracts data from all subjects. This mayhave anatomical causes, which is supported by findingsin single slices acquired at high resolution in man, cat,and monkey (38–40). Another effect could be decreasedeffective resolution because of local relaxation effects assuggested in Fig. 2f. Regardless of source, this could be arelevant factor to take into account when prescribing thickor noncontiguous slices, which are often applied in axialspinal cord imaging with high in plane resolution.

In our data, tractography was also possible below thecervical region but with disruptions around T1/T2. Thiscould be due to the proximity of the airways in this regioncausing large local field gradients. Large distortions werealso registered in the PSF map in this region, see Fig. 1a.To improve performance in this region, distortions couldbe reduced at the scanning stage by for instance increasedbandwidth and increased number of repetitions to accountfor decreased SNR.

Model-Based Registration

This is to our knowledge the first application of this regis-tration framework to diffusion weighted spinal cord dataand robust performance was achieved. The applicationof a correction map on a series of images demands goodalignment and the possibility of gross subject movementduring acquisition therefore motivated a registration step.The model-based registration is attractive, as it can be usedwith standard registration algorithms and provides optimalregistration references for the individual image contrast ofinterest for high b-value or low SNR data. Robust registra-tion of diffusion data allows extended acquisition times toincrease SNR, angular resolution or the number of diffusionweightings with reduced confounds from subject motion.Eddy current effects, causing different image deformationsacross the different gradient directions, can be a problemespecially with monopolar diffusion encoding gradientsbut could also be addressed with the model-based regis-tration reference combined with higher order registrationalgorithms.

CONCLUSIONS

In this study, we have demonstrated that the PSFcould be recorded from the spinal cord and applied for

distortion correction of diffusion-weighted EPI datasetswith isotropic resolution and large volume coverage. Amodel-based registration method was applied providinggood alignment within the diffusion-weighted dataset andto the PSF reference. As a consequence of the correction,anatomical consistency was improved and made determin-istic tractography possible throughout the whole cervicalspinal cord. The method was demonstrated both on healthycontrols and on an individual with SCI. We suggest that oursetup may have clinical interest providing large volumecoverage and long tract information for early detection oflesions or for monitoring axonal degeneration over time.The imaging settings chosen in this study were set to pro-vide high SNR and large volume coverage in short timeresulting in large distortions, but with longer scan timesand reduced FoV methods a less challenging basis forcorrection could be provided.

ACKNOWLEDGMENTS

The authors thank Maxim Zaitsev at MRDAC, Freiburg,Germany for providing the PSF-EPI sequences. Tim B.Dyrby has been supported by the Lundbeck Foundation.Tim B. Dyrby and Henrik Lundell are partly supported bythe CONNECT consortium project supported by The Futureand Emerging Technologies (FET open) of the EU FP7.

REFERENCES

1. Maier S. Examination of spinal cord tissue architecture with magneticresonance diffusion tensor imaging. Neurotherapeutics 2007;4:453–459.

2. Conturo TE, Lori NF, Cull TS, Akbudak E, Snyder AZ, Shimony JS,McKinstry RC, Burton H, Raichle ME. Tracking neuronal fiber pathwaysin the living human brain. Proc Natl Acad Sci 1999;96:10422–10427.

3. Beaulieu C. The basis of anisotropic water diffusion in the nervoussystem — a technical review. NMR Biomed 2002;15:435–455.

4. Ciccarelli O, Wheeler-Kingshott CA, Mclean MA, Cercignani M, Wim-pey K, Miller DH, Thompson AJ. Spinal cord spectroscopy anddiffusion-based tractography to assess acute disability in multiplesclerosis. Brain 2007;130:2220–2231.

5. Filippi CG, Andrews T, Gonyea JV, Linnell G, Cauley KA. Magneticresonance diffusion tensor imaging and tractography of the lower spinalcord: application to diastematomyelia and tethered cord. Eur Radiol201020:2194–2199.

6. Onu M, Gervai P, Cohen-Adad J, Lawrence J, Kornelsen J, Tomanek B,Sboto-Frankenstein UN. Human cervical spinal cord funiculi: investi-gation with magnetic resonance diffusion tensor imaging. J Magn ResonMed 2010;31:829–837.

7. Smith S, Jones C, Gifford A, Belegu V, Chodkowski B, Farrell J, Land-man B, Reich D, Calabresi P, McDonald J, van Zijl P. Reproducibility oftract-specific magnetization transfer and diffusion tensor imaging in thecervical spinal cord at 3 tesla. NMR Biomed 2009;23:207–217.

8. Chang Y, Jung TD, Yoo DS, Hyun JK. Diffusion tensor imaging and fibertractography of patients with cervical spinal cord injury. J Neurotrauma2010;27:2033–2040.

9. Ellingson BM, Ulmer JL, Kurpad SN, Schmit BD. Diffusion tensor MRimaging in chronic spinal cord injury. Am J Neuroradiol 2008;29:1976–1982.

10. Renoux J, Facon D, Fillard P, Huynh I, Lasjaunias P, Ducreux D. MRdiffusion tensor imaging and fiber tracking in inflammatory diseases ofthe spinal cord. Am J Neuroradiol 2006;27:1947–1951.

11. Tsuchiya K, Fujikawa A, Honya K, Nitatori T, Suzuki Y. Diffusion tensortractography of the lower spinal cord. Neuroradiology 2008;50:221–225.

12. Hecke WV, Nagels G, Emonds G, Leemans A, Sijbers J, Goethem JV,Parizel PM. A diffusion tensor imaging group study of the spinal cordin multiple sclerosis patients with and without T2 spinal cord lesions.J Magn Reson Med 2009;30:25–34.

Spinal Cord PSF Mapping and DTI 149

13. Robson MD, Gore JC, Constable RT. Measurement of the point spreadfunction in MRI using constant time imaging. Magn Reson Med1997;38:733–740.

14. Zaitsev M, Hennig J, Speck O. Point spread function mapping with par-allel imaging techniques and high acceleration factors: fast, robust, andflexible method for echo-planar imaging distortion correction. MagnReson Med 2004;52:1156–1166.

15. Zeng H, Constable RT. Image distortion correction in EPI: comparisonof field mapping with point spread function mapping. Magn Reson Med2002;48:137–146.

16. Griswold M, Jakob P, Heidmann R, Nittka M, Jellus V, Wang J, KieferB, Haase A. Generalized autocalibrating partially parallel acquisitions(GRAPPA). Magn Reson Med 2002;47:1202–1210.

17. Reese T, Heid O, Weisskoff R, Wedeen V. Reduction of eddy-current-induced distortion in diffusion MRI using a twice-refocused spin echoMagn Reson Med 2003;49:177–182.

18. Jones DK, Horsfield MA, Simmons A. Optimal strategies for measuringdiffusion in anisotropic systems by magnetic resonance imaging. MagnReson Med 1999;42:515–525.

19. Bai Y, Alexander D. Model-Based registration to Correct for MotionBetween Acquisitions in Diffusion MR Imaging. In: The 5th IEEEInternational Symposium on Biomedical Imaging, 2008. pp 947–950.

20. Cook PA, Bai Y, Nedjati-Gilani S, Seunarine KK, Hall MG, Parker GJ,Alexander DC. Camino: Open-Source Diffusion-MRI Reconstructionand Processing. In: Proceedings of 14th ISMRM, 2006 November, p 2759.

21. Basser P, Mattiello J, Bihan DL. Estimation of the effective self-diffusiontensor from the NMR spin echo. J Magn Reson B 1994;103:247–254.

22. Mori S, Crain B, Chacko V, Zijl P. Three-dimensional tracking of axonalprojections in the brain by magnetic resonance imaging. Ann Neurol1999;45:265–269.

23. Stroman P, Figley C, Cahill C. Spatial normalization, bulk motion cor-rection and coregistration for functional magnetic resonance imagingof the human cervical spinal cord and brainstem. Magn Reson Imaging2008;26:809–814.

24. Dowell NG, Jenkins TM, Ciccarelli O, Miller DH, Wheeler-KingshottCAM. Contiguous-slice zonally oblique multislice (CO-ZOOM) diffu-sion tensor imaging: examples of in vivo spinal cord and optic nerveapplications. J Magn Reson Med 2009;29:454–460.

25. Jeong EK, Kim SE, Kholmovski EG, Parker DL. High-resolution DTI ofa localized volume using 3D single-shot diffusion-weighted stimulatedecho-planar imaging (3D ss-DWSTEPI). Magn Reson Med 2006;56:1173–1181.

26. Wilm BJ, Gamper U, Henning A, Pruessmann KP, Kollias SS, BoesigerP. Diffusion-weighted imaging of the entire spinal cord. NMR Biomed2008;22:174–181.

27. Saritas EU, Cunningham CH, Lee JH, Han ET, Nishimura DG. DWI ofthe spinal cord with reduced FOV single-shot EPI. Magn Reson Med2008;60:468–473.

28. Bammer R, Fazekas F, Augustin M, Simbrunner J, Strasser-Fuchs S,Seifert T, Stollberger R, Hartung HP. Diffusion-weighted MR imagingof the spinal cord. Am J Neuroradiol 2000;21:587–591.

29. Clark CA, Barker GJ, Tofts PS. Magnetic resonance diffusion imaging ofthe human cervical spinal cord in vivo. Magn Reson Med 1999;41:1269–1273.

30. Jaermann T, Pruessmann KP, Valavanis A, Kollias S, Boesiger P. Influ-ence of SENSE on image properties in high-resolution single-shotecho-planar DTI. Magn Reson Med 2006;55:335–342.

31. Callaghan PT. Principles of nuclear magnetic resonance microscopy.Oxford, UK: Oxford University Press; 1991.

32. Parker GJM, Haroon HA, Wheeler-Kingshott CAM. A framework fora streamline-based probabilistic index of connectivity (PICo) using astructural interpretation of MRI diffusion measurements. J Magn ResonImaging 2003;18:242–254.

33. Kwong K, McKinstry R, Chien D, Crawley A, Pearlman J, Rosen B. CSF-Suppressed Quantitative Single-Shot Diffusion Imaging. Magn ResonMed 1991;21:157–163.

34. Vos S, Jones D, Viergaver M, Leemans A. Partial volume effect as ahidden covariate in DTI analyses. NeuroImage 2011;55:1566–1576.

35. Lundell H, Christensen MS, Barthélemy D, Willerslev-Olsen M, Biering-Sørensen F, Nielsen JB. Cerebral activation is correlated to regionalatrophy of the spinal cord and functional motor disability in spinalcord injured individuals. NeuroImage 2011;54:1254–1261.

36. Lundell H, Barthelemy D, Skimminge A, Dyrby TB, Biering-SørensenF, Nielsen JB. Independent spinal cord atrophy measures correlate tomotor and sensory deficits in individuals with spinal cord injury. SpinalCord 2011;49:70–75.

37. Cohen-Adad J, Mendili MME, Lehéricy S, Pradat PF, Blancho S, Rossig-nol S, Benali H. Demyelination in the injured human spinal corddetected with diffusion and magnetization transfer imaging. NeuroIm-age 2011;55:1024–1033.

38. Cohen-Adad J, Leblond H, Ghosh A, Descoteaux M, Deriche R, BenaliH, Rossignol S. Evaluation of q-Ball Metrics for Assessing the Integrityof the Injured Spinal Cord. In: Proceedings of 17th ISMRM, 2009November, p 638.

39. Schneider T, Alexander DC, Wheeler-Kingshott CA. Preliminary Inves-tigation of Position Dependency of Radial Diffusivity in the CervicalSpinal Cord. In: Proceedings of 17th ISMRM, 2009 November, p 3191.

40. Lundell H, Nielsen J, Ptito M, Dyrby T. Distribution of collateral fibersin the monkey cervical spinal cord detected with diffusion-weightedmagnetic resonance imaging. NeuroImage 2011;56:923–929.