MAGNETIC RESONANCE

Assessment of reduced field of view in diffusion tensor imagingof the lumbar nerve roots at 3 T

Jean-François Budzik & Sébastien Verclytte &

Guillaume Lefebvre & Aurélien Monnet & Gerard Forzy &

Anne Cotten

Received: 5 July 2012 /Accepted: 15 October 2012 /Published online: 18 November 2012# European Society of Radiology 2012

AbstractObjectives To assess the value of reduced field of view(rFOV) imaging in diffusion tensor imaging (DTI) andtractography of the lumbar nerve roots at 3 T from theperspective of future clinical trials.Methods DTI images of the lumbar nerves were obtained ineight healthy volunteers, with and without the rFOV tech-nique. Non-coplanar excitation and refocusing pulses asso-ciated with outer volume suppression (OVS) were used toachieve rFOV imaging. Tractography was performed. A

visual evaluation of image quality was made by two observ-ers, both senior musculoskeletal radiologists. Fractional an-isotropy (FA) and apparent diffusion coefficient (ADC)were measured in L5 and S1 roots.Results rFOV images of the L5 and S1 roots wereassessed as being superior to full FOV (fFOV) images.Image quality was rated as good to excellent by bothobservers. Interobserver agreement was good. No signif-icant difference was found in FA and ADC measure-ments of the L5 or S1 roots. On the contrary, only

J.-F. Budzik : S. VerclytteGroupe Hospitalier de l’Institut Catholique de Lille, ImagerieMédicale,Lille, France

J.-F. Budzik : S. Verclytte :G. ForzyFaculté Libre de Médecine,Lille, France

J.-F. Budzik :G. Lefebvre :A. CottenCentre Hospitalier Universitaire de Lille,Imagerie Musculo-squelettique,Lille, France

J.-F. Budzik :A. CottenEA 4490 PMOI (Physiopathologie des Maladies OsseusesInflammatoires) IFR 114 PRES, Université Lille Nord de France,Lille, France

J.-F. Budzik : S. Verclytte :G. ForzyUniversité Catholique de Lille,59000 Lille, France

J.-F. Budzik : S. Verclytte :G. Lefebvre :G. Forzy :A. CottenUniversité Nord de France,59000 Lille, France

A. MonnetService de Neuroradiologie, CHU de Lille,EA 1046–IFR 114_IMPRT,Lille, France

G. ForzyLaboratoire de biologie, département de biostatistiques, Groupehospitalier de l’Institut Catholique de Lille,59000 Lille, France

J.-F. Budzik (*) : S. VerclytteService d’Imagerie Médicale, Hôpital St Vincent de Paul,Boulevard de Belfort, B.P. 387, 59020 Lille Cedex, Francee-mail: budzik.jean-francois@ghicl.net

G. Lefebvre :A. CottenService de Radiologie et Imagerie Musculosquelettique, Centre deConsultations et d’Imagerie de l’Appareil Locomoteur,Rue du Pr. Emile Laine,59037 Lille Cedex, France

A. MonnetService de Neuroradiologie, Hôpital Roger Salengro,Rue du Pr. Emile Laine,59037 Lille Cedex, France

G. ForzyLaboratoire de Biologie, département de biostatistiques, Hôpital StPhilibert,115 Rue du Grand But,59160 Lomme, France

Eur Radiol (2013) 23:1361–1366DOI 10.1007/s00330-012-2710-0

poor-quality images could be obtained with fFOV imag-ing as major artefacts were present.Conclusion The rFOV approach was essential to achievehigh-quality DTI imaging of lumbar nerve roots on 3-TMRI.Key Points• Diffusion tensor 3-T MR imaging of lumbar nerve rootscreates severe artefacts.

• A reduced field of view drastically reduces artefacts,thereby improving image quality.

• Good-quality tractography images can even be obtainedwith rFOV imaging.

• rFOV DTI is better than fFOV DTI for clinical studies.

Keywords Magnetic resonance imaging . Diffusion tensorimaging . Peripheral nerves . Echo planar imaging .

Lumbosacral region

Abbreviations and AcronymsADC apparent diffusion coefficientDTI diffusion tensor imagingEPI echo planar imagingFA fractional anisotropyfFOV full field of viewrFOV reduced field of viewOVS outer volume suppressionROI region of interestSNR signal to noise ratio

Introduction

Diffusion tensor imaging (DTI) associated with fibre track-ing has emerged as a new tool for clinical studies of thespinal cord [1], muscles [2] and peripheral nerves [3]. Thepotential of DTI to identify microstructural changes of bio-logical tissues is highlighted by the demonstration thatparameters issuing from DTI images (fractional anisotro-py, FA, and apparent diffusion coefficient, ADC) aremodified in physiological or pathological conditions, suchas cervical spondylotic myelopathy or carpal tunnel syn-drome [4, 5].

The feasibility of this technique in the study of the lumbarnerve roots has been reported recently [6, 7]. The conclusionssuggest that DTI could become a valuable tool for the assess-ment of radicular pain. As 3-T MRI systems are being in-stalled in more and more clinical centres, DTI of the lumbarnerves should be implemented on such devices. The currentpopularity of higher field MRI is mainly justified by theexpected signal increase, enabling higher resolution and fasteracquisition times. However, this is counterbalanced by highersensitivity to susceptibility-related or chemical shift artefactsas well as T2* blurring consequent to faster relaxation. This is

particularly problematic in the lumbar region: lumbar rootshave small cross-sectional areas and high surface-to-volume ratios, leading to increased partial volume effectsalong the nerve boundaries. High spatial resolution is thusrequired. The loss of signal inherent to smaller voxelsneeds to be counterbalanced as a high signal-to-noise ratio(SNR) is necessary to obtain accurate measurements ofdiffusion parameters.

An approach known as reduced field of view (rFOV) hasbeen introduced in DTI with interesting results [8–13]. Onlythe signal coming from a pre-defined area of interest isconserved. A reduced FOV is explored by a shorter echotrain, allowing a shorter TE. Readout is also shortened as theFOV is limited along the phase-encoding direction. Thepreviously described artefacts are therefore reduced.

A combination of two innovative strategies makes rFOVachievable, as described by Wilm [11, 13, 14]. Non-coplanar excitation and refocusing pulses are applied todefine the FOV. With outer volume suppression (OVS),dephasing gradients are then applied in the outer volumeregions [10].

The aims of this study were thus to assess the value ofrFOV imaging in DTI and tractography of the lumbar nerveroots at 3 T from the perspective of future clinical trials.

Materials and methods

Our study was in accordance with the ethical standards ofthe World Medical Association (Declaration of Helsinki)and was approved by the local ethics committee.

Population

Eight volunteers were included in our study. Their informedconsent was obtained. They were asymptomatic, had nohistory of low back pain, sciatica, or previous back surgery.There were six men and two women, aged 25 to 30 years(mean 27.9, SD 1.96).

MRI examination

Acquisition parameters

Magnetic resonance imaging was performed on a 3-T full-body MR system (Achieva, Philips Medical Systems, Best,The Netherlands). Volunteers underwent imaging in thesupine position. A 32-channel multi-phased “Cardiac” coilcovering the pelvis and the lower abdomen was employed.

The MR protocol included a T2-weighted spin-echo se-quence acquired in the axial plane (used for anatomicalcorrelation) and DTI sequences in the axial plane with andwithout rFOV.

1362 Eur Radiol (2013) 23:1361–1366

The full FOV diffusion tensor imaging (fFOV) and rFOVwere performed in the axial plane with a spin-echo single-shot diffusion-weighted echo-planar imaging (SS-EPI) se-quence, a b-value of 700 s/mm2, and 15 different diffusiongradient orientations. Spectral presaturation with inversionrecovery (SPIR) was chosen to suppress the fat signal. Fourexcitations and partial Fourier acquisition (half scan factor:0.61) were used. Twelve contiguous 3-mm-thick slices wereacquired covering L5 and S1 roots according to the T2images acquired previously. For fFOV, TR was 5,379 ms,TE 54 ms, FOV 200×200 mm, and the matrix 132×132.The in-plane resolution was 1.56×1.56 mm. Total acquisi-tion time was 11 min 39 s.

For rFOV, TR was 5,379 ms, TE 42 ms, FOV 100×53 mm, matrix 68×34, and resolution 1.47×1.52 mm; ac-quisition time was 11 min 39 s.

Extra DTI sequences were performed to calculate theSNR, with and without rFOV.

Post-processing of DTI images

DTI images were processed with MedInria software(Asclepios Research Project–INRIA Sophia Antipolis).The algorithm was based on an extraction of the principaldiffusion direction (PDD) of the tensor field in the regionswhere the diffusivity was highly linear and a vector-basedtracing scheme appeared. FA maps were calculated. Tractog-raphy was performed on the whole volume acquired for eachvolunteer. No ROI was used to induce the tracking. Right andleft L5 and S1 nerve roots were isolated. Mean FA and ADCvalues were calculated in each root for each sequence and foreach volunteer.

SNR measurements

Two consecutive images with identical parameters wereacquired on each subject. Both images were then subtractedat b00, hence creating a third pixel-by-pixel differenceimage. A region of interest was positioned on the dural sacon the b0 image (ROI1) and on the subtracted image(ROI2). Signal (S) was the mean value of the pixel intensityin ROI1. The noise was defined as the standard deviation(SD) derived from ROI2. The SNR was calculated accord-ing to the method described by Price et al. [12] as follows:SNR ¼ S ROI1ð Þp2 SD ROI2ð Þ= .

Qualitative assessment

The visual quality of the images was assessed by two seniorradiologists (observers 1 and 2) according to an analogicalvisual scale, as described in Table 1. The mean scores werecalculated for each item and each observer, as well as theoverall mean score for each observer.

Quantitative assessment

The following measurements were made on each individu-alised L5 and S1 root: number of voxels included in thefibres (volume of fibres), number of fibres, and mean lengthof the fibres.

Statistical analysis

Data comparison was performed using Statistica software9.0 (StatSoft, Inc.). P values less than 0.05 were consideredstatistically significant. Because of the small sample size,statistical analysis was performed with non-parametric tests.Comparisons between groups were performed with the Wil-coxon test. Interobserver agreement and pair-wise compar-isons of observers’ evaluations of the different DTIsequences were assessed using Cochran’s Q test.

Results

The two sequences differed significantly on all ratings, withrFOV having consistently higher ratings than the fFOV, asdetailed below.

Artefacts were present and rated as “major” by bothobservers on FA maps issued from fFOV for six out of eightvolunteers. On the contrary, no distortions were visible onrFOV images in six out of eight subjects (Fig. 1). Minordistortions were identified on the other subjects. The meanevaluation scores were 1.5 for fFOV and 4.5 for rFOV forboth observers. Overall image quality was rated 2.25 forfFOV and 4.4 for rFOV (mean scores for both observers).

With rFOV, L5 and S1 roots could be depicted along theirentire length in the volume studied (Fig. 2) in all of thevolunteers. Image quality was rated good to excellent byboth observers (mean score: 4.625), as was the correlationwith T2 images (mean score: 4.25). On the contrary, imagequality of the L5 and S1 roots obtained with fFOV was poor(mean score: 1.75); correlation with T2 images was rated aspoor (mean score: 1.375).

Interobserver agreement was good as no significant dif-ference was noted in the ratings of the different items. Theobservers preferred the rFOV images by a strong margin forall items: the overall mean score was <2.5 for fFOV and >4for rFOV. Differences between rFOVand fFOV images weresignificant for all items for both observers.

The fibres appearing on the fFOV images displayed thefollowing characteristics: mean number of voxels0410.7;mean number of fibres030.7; mean length of fibres 14.4.For rFOV images, the corresponding values were 1,079.3,111.9, and 25.9 respectively. The comparison of rFOV andfFOV revealed significant differences for each item, both foroverall and root-by-root comparisons.

Eur Radiol (2013) 23:1361–1366 1363

Mean FA values were 0.38 for L5 roots and 0.335 for S1roots. Mean ADC values were 2.8 mm2/s for L5 roots and4.03 mm2/s for S1 roots. These measures were made on rFOVsequences as the overall quality of the root images obtainedwith fFOV sequences was poor and clinically pointless.

The mean signal-to-noise ratio values were 14.49 forrFOV and 11.61 for fFOV. The box plot (Fig. 4) showed awider scattering of the SNR values in fFOV (from 2.19 to22.10) compared with rFOV (from 9.11 to 16.94). However,no significant difference was found between SNR values ofrFOV and fFOV sequences.

Discussion

A reduced FOV was compulsory to achieve correct depic-tion of the L5 and S1 lumbar nerve roots. Without the rFOVtechnique, the images were impeded by macroscopic arte-facts leading to severe image distortion and low-quality orimpossible fibre tracking (Fig. 3).

Reducing the FOVon the x and y axes makes it possibleto obtain images at a given spatial resolution with a reducednumber of k-space lines. For single-shot EPI imaging, thisalso leads to a reduced echo train and consequently a shorterTE. Thus the effects of T2* decay are less important, result-ing in a smaller loss of signal. As the readout time along thephase-encode dimension is also shorter, sensitivity to chem-ical shift and blurring artefacts is lessened [9, 11]. However,one must keep in mind that the single-shot EPI base gives alower limit for TE as the whole k-space has to be read by asingle echo.

Mean signal-to-noise ratio values were higher with rFOVthan with fFOV, and the SNR values were more homoge-neous with rFOV, whereas they were scattered from verylow to very high values with fFOV. The small size of our

Table 1 Analogical visual scale used by the observers for the image quality assessment

1 2 3 4 5

Distortions on FA maps Major distortions X Minor distortions X No distortion

Tractography Impossible Many aberrant fibres Few aberrant fibres rare aberrant fibres No aberrant fibres

Visualisation of L5 and S1 roots No root poor satisfactory good excellent

Correlation with T2 images No root poor satisfactory good excellent

FA fractional anisotropy

Fig. 2 High-quality tractography images with anatomical fusionobtained with rFOV technique; posterior (a) and superior (b) views.In this example, S2 and S3 roots are also visible within the dural sac.The colour of the roots is consistent with their orientation. The L5 rootsare blue in their dural and extra-foraminal segments, whereas they turnpurple because of their oblique direction within the foramen. VerticalS1 roots are blue as they are about to enter the L5–S1 foramens.Inferior sacral roots turn green as they run down obliquely in anantero-posterior direction to reach the sacral foramens

Fig. 1 a FA map freed from artefacts obtained with reduced field ofview (rFOV) imaging. b Same image from the same volunteer obtainedwith fFOV imaging. The distortion of the dural sac is obvious

1364 Eur Radiol (2013) 23:1361–1366

sample probably explains the lack of significant differencein the statistical analysis.

Cardiac and/or respiratory gating or image registrationcould have been applied to increase the SNR. But consideringthe prospect of further clinical use, these time-consumingsteps were ruled out. Moreover, although some artefacts mayhave been reduced, it is important to note that gating increasesacquisition time and does not prevent subject motion.

The agreement between the two radiologists was good. Inour opinion, the use of a semi-automated method contribut-ed to this concordance. Our tractography software performsthe calculations on the whole volume. The role of theoperator is then to crop the image to focus on the structuresof interest. This method lessens the subjectivity induced byROI positioning, a method used in many DTI studies: therisk of including voxels that do not belong to the nerve inthe ROI is minimised with our method. The risk of includingor excluding voxels because of faulty threshold settingsremains, but is the same with both methods.

A similar work was recently published by Zaharchuk etal. [13], although it examined diffusion-weighting imaging(DWI) instead of DTI. The authors compared the visualquality of DWI images of the spinal cord acquired witheither the rFOV or fFOV technique. rFOV proved superiorto fFOV imaging for all criteria studied. The main technicaldifferences with our study are the absence of DTI, the use ofa 1.5-T magnet, and a different method of obtaining areduced FOV, as described by Saritas [10].

Tractography of the lumbar nerves at 3 T was onlyrecently reported by Eguchi et al. [7]. The authors did notreport using any reduced FOV technique. Compared withour study, the b-value and number of directions were close;the slice thickness and number of excitations were identical;TE was longer (76 ms versus 42 ms) and the voxel wasbigger (3.3×1.6×3 versus 1.52×1.47×3). However, 50 ax-ial slices were acquired in less than 5 min versus 12 in11 min 39 s in our study. This important difference may beexplained by the time needed to apply the OVS. Yet thishypothesis cannot be confirmed as the authors did not provideexhaustive information concerning the sequence developed intheir study. Moreover, no assessment of the SNR, imagequality, or interobserver agreement was presented.

This long acquisition time is the main criticism that canbe made about OVS. This technique requires an excitationof the full FOV, followed by a suppression of the signalcoming from the outer regions. However, in our study, thesame time was necessary to obtain reduced and full FOVimages. As the quality of the rFOV images was far betterthan the that of the fFOV ones, this notion of loss of timemust be counterbalanced: the reduced number of k-spacelines to be coded saves time and allows modifications of thesequence’s parameters to achieve better image quality.

Fig. 3 a Tractography images of L5 and S1 roots obtained using therFOVapproach. b The same view is shown in the same volunteer usingthe fFOV technique. In the latter the fibres are scarce and shorter: thisis particularly demonstrative for left L5 and S1 roots

rFOV fFOV0

2

4

6

8

10

12

14

16

18

20

22

24 median 25%-75% Min-Max

Fig. 4 Box plot with signal-to-noise ratio (SNR) measures (on y axis)in rFOV and fFOV sequences. Although SNR values are rather homo-geneous in rFOV imaging, they range from very low to very highvalues in fFOV images and are much more scattered

Eur Radiol (2013) 23:1361–1366 1365

There are alternatives to achieving rFOV imaging. InnerFOV imaging has been used in cervical cord imaging [8,14]; its main difference compared with our technique is aspecific excitation of a rectangular volume of interest donewith a 2D spatially selective echo-planar radiofrequencyexcitation pulse. One might suppose that this techniqueis faster than ours as it does not take into considerationthe voxels that do not belong to the final volume, unlikeOVS in which the molecules from these voxels areexcited and then dephased. Comparison of different tech-niques would be interesting to implement time-efficientrFOV sequences on 3-T MRI.

The FA values of L5 and S1 roots differed from thoseobtained in the study of the lumbar nerve roots at 1.5 T [6].For example, mean FAwas 0.38 in the L5 roots at 3 T versus0.22 at 1.5 T. These differences are frequent in DTI studies.Santarelli et al. demonstrated that FA values were dependenton the number of diffusion gradients, repeated acquisi-tion, and voxel size [15]. Farrell et al. showed that anFA measurement bias existed as SNR varied [16]. Thesediscrepancies in the measures seem to limit FA compar-isons to subjects studied on the same MR device withthe same sequence. Reproducibility evaluation shouldthen be mandatory before initiating a DTI multicentrestudy [10, 19, 16, 11].

A recent review of research studies comparing 1.5 Tversus 3 T in the field of neuroradiology [17] reports thatno convincing evidence of improved diagnostic accuracy orreduced total examination times was found. However, therelevance of such comparisons can be discussed [18], andthis does not seem essential for DTI imaging of the lumbarnerves. The fact that this technique can be implemented atboth fields is enough to perform clinical studies in anyimaging centre [9, 12].

Our study is mainly limited by the small number ofsubjects included. However, the quality of tractographyimages was far better with rFOV than with fFOV inevery subject. Comparison with other rFOV techniqueswas not made, but rFOV was implemented in a singlemanner on our MRI.

In conclusion, reduced FOV imaging was essential toachieve DTI imaging and tractography of the lumbar nerveroots at 3 T. This technique enabled a drastic reduction ofthe susceptibility and chemical shift artefacts that impededaccurate tractography on full FOV imaging. However, asthis work suggests that this technique is time-consuming,further studies are needed to assess the pros and cons ofother reduced FOV imaging techniques (Fig. 4).

Acknowledgements The authors would like to thank Prof. XavierLeclerc, PhD, from the Neuroradiology Department of the CHRU, whogave us access to the imaging platform of IFR 114– IMPRT, and Mr.David Chechin, PhD, Philips Healthcare, for his technical advice.

References

1. Budzik JF, Balbi V, Le Thuc V, Duhamel A, Assaker R, Cotten A(2011) Diffusion tensor imaging and fibre tracking in cervicalspondylotic myelopathy. Eur Radiol 21:426–433

2. Budzik JF, Le Thuc V, Demondion X, Morel M, Chechin D,Cotten A (2007) In vivo MR tractography of thigh muscles usingdiffusion imaging: initial results. Eur Radiol 17:3079–3085

3. Khalil C, Budzik JF, Kermarrec E, Balbi V, Le Thuc V, Cotten A(2010) Tractography of peripheral nerves and skeletal muscles. EurJ Radiol 76:391–397

4. Hiltunen J, Kirveskari E, Numminen J, Lindfors N, Goransson H,Hari R (2012) Pre- and post-operative diffusion tensor imaging of themedian nerve in carpal tunnel syndrome. Eur Radiol 22:1310–1319

5. Khalil C, Hancart C, Le Thuc V, Chantelot C, Chechin D, Cotten A(2008) Diffusion tensor imaging and tractography of the mediannerve in carpal tunnel syndrome: preliminary results. Eur Radiol18:2283–2291

6. Balbi V, Budzik JF, Duhamel A, Bera-Louville A, Le Thuc V,Cotten A (2011) Tractography of lumbar nerve roots: initial results.Eur Radiol 21:1153–1159

7. Eguchi Y, Ohtori S, Orita S et al (2011) Quantitative evaluationand visualization of lumbar foraminal nerve root entrapment byusing diffusion tensor imaging: preliminary results. AJNR Am JNeuroradiol 32:1824–1829

8. Barakat N, Mohamed FB, Hunter LN et al (2012) Diffusion tensorimaging of the normal pediatric spinal cord using an inner field ofview echo-planar imaging sequence. AJNR Am J Neuroradiol33:1127–1133

9. Karampinos DC, Van AT, Olivero WC, Georgiadis JG, Sutton BP(2009) High-resolution diffusion tensor imaging of the humanpons with a reduced field-of-view, multishot, variable-density,spiral acquisition at 3 T. Magn Reson Med 62:1007–1016

10. Saritas EU, Cunningham CH, Lee JH, Han ET, Nishimura DG(2008) DWI of the spinal cord with reduced FOV single-shot EPI.Magn Reson Med 60:468–473

11. Wilm BJ, Svensson J, Henning A, Pruessmann KP, Boesiger P, KolliasSS (2007) Reduced field-of-viewMRI using outer volume suppressionfor spinal cord diffusion imaging. Magn Reson Med 57:625–630

12. Price RR, Axel L, Morgan T et al (1990) Quality assurancemethods and phantoms for magnetic resonance imaging: reportof AAPM nuclear magnetic resonance Task Group No. 1. MedPhys 17:287–295

13. Zaharchuk G, Saritas EU, Andre JB et al (2011) Reduced field-of-view diffusion imaging of the human spinal cord: comparison withconventional single-shot echo-planar imaging. AJNR Am J Neuro-radiol 32:813–820

14. Finsterbusch J (2009) High-resolution diffusion tensor imagingwith inner field-of-view EPI. J Magn Reson Imaging 29:987–993

15. Santarelli X, Garbin G, Ukmar M, Longo R (2010) Dependence ofthe fractional anisotropy in cervical spine from the number ofdiffusion gradients, repeated acquisition and voxel size. MagnReson Imaging 28:70–76

16. Farrell JA, Landman BA, Jones CK et al (2007) Effects of signal-to-noise ratio on the accuracy and reproducibility of diffusiontensor imaging-derived fractional anisotropy, mean diffusivity,and principal eigenvector measurements at 1.5 T. J Magn ResonImaging 26:756–767

17. Wardlaw JM, Brindle W, Casado AM et al (2012) A systematicreview of the utility of 1.5 versus 3 Tesla magnetic resonance brainimaging in clinical practice and research. Eur Radiol. doi:10.1007/s00330-012-2500-8

18. Wattjes MP, Barkhof F (2012) Diagnostic relevance of high fieldMRI in clinical neuroradiology: the advantages and challenges ofdriving a sports car. Eur Radiol. doi:10.1007/s00330-012-2552-9

1366 Eur Radiol (2013) 23:1361–1366