www.elsevier.com/locate/ynimg

NeuroImage 32 (2006) 1243–1249

Spatial resolution dependence of DTI tractography in humanoccipito-callosal region

Mina Kim,a,1 Itamar Ronen,b Kamil Ugurbil,a and Dae-Shik Kimb,⁎

aCenter for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN 55454, USAbCenter for Biomedical Imaging, Department of Anatomy and Neurobiology, 715 Albany Street, L-1004, Boston University School of Medicine, Boston,MA 02118, USA

Received 6 March 2006; revised 7 June 2006; accepted 9 June 2006Available online 24 July 2006

Diffusion tensor imaging (DTI) and fiber tracking have been used tomeasure the fiber structural connectivity in humans in a non-invasivemanner. However, low sensitivity is a principal limitation of thesemethods, causing a large number of possibly missing fiber tracts (FTs).Here we studied how the spatial resolution affects the sensitivity of thefiber tracing by rescaling data to different resolutions. Our datasuggest that the spatial resolution can change the degree of theasymmetric cross-callosal connections in the lower visual field (loVF)compared to the upper visual field (upVF). Among connections fromloVF, a larger voxel size resulted in a smaller number of FTs that wasnot commensurate to the number of seed points, while the number ofconnections from upVF was not significantly affected by variation inseeding point numbers. We conclude from our study that the spatialresolution of the acquired data will have to be taken into considerationin interpreting DTI fiber tracking data. Our results further suggestthat the acquisition resolution of around 2 mm iso-voxel in theconventional DTI scheme can reconstruct the asymmetric upper andlower white matter structure in occipito-callosal region.© 2006 Elsevier Inc. All rights reserved.

Keywords: DTI; Fiber tracking; Spatial resolution; Cross-Callosal connec-tions; Occipital lobe

Introduction

Diffusion tensor magnetic resonance imaging (DT-MRI),coupled with fiber-tracking algorithms, promises to provideimportant information about both animal and human fiber tracts(FTs). DTI is a technique in which the presence and the orientation

⁎ Corresponding author. Fax: +1 617 414 2362.E-mail addresses: mina@mri.jhu.edu (M. Kim), dskim@bu.edu

(D.-S. Kim).1 Current address: F.M. Kirby Center for Functional Brain Imaging,

Kennedy Krieger Institute, 707 N. Broadway, Baltimore, MD 21205, USA.Fax: +1 443 923 9505.

Available online on ScienceDirect (www.sciencedirect.com).

1053-8119/$ - see front matter © 2006 Elsevier Inc. All rights reserved.doi:10.1016/j.neuroimage.2006.06.006

of axonal fibers can be reconstructed through a careful measure-ment of the intravoxel motion of water molecules in the brain(Basser et al., 1994). To study the dynamic three-dimensional (3D)process of the formation of brain structures, the white mattertractography based on DTI exploits the increased diffusivity ofwater molecules along the direction of fibers (Basser et al., 2000;Catani et al., 2002). DTI techniques in combination with 3D fiberreconstruction algorithm have been used to generate robust imagesof the axonal connectivity pattern in vivo in humans (Conturo etal., 1999), rodents (Mori et al., 1999) and cats (Kim et al., 2003).

There are, however, technical limitations in DTI that must beconsidered when inferring connectivity. In particular, a detailedknowledge about the precise correlation for spatial resolutionremains elusive. This is of particular concern, as most DT-MRIstudies in humans are being performed at relatively coarse voxelsizes in order to maximize the SNR gain (~1.5–5.0 mm in-plane).Such a coarse resolution may yield ambiguous results however, asDougherty et al. have demonstrated based on cross-callosalconnections between the bi-hemispheric visual cortices (Doughertyet al., 2005). Especially, it was reported that a diffusion-tensor MRIimaging experiment on a normal human brain revealed significantpartial volume effects between oblique white matter regions whenusing very large voxels and large diffusion-weighting, whereas theapparent partial volume effects in white matter decreasedsignificantly when smaller voxel dimensions were used (Alexanderet al., 2001).

Here we studied the effect of differential spatial resolutions ofDTI data by systematically re-sampling the acquired data into avariety of nominal resolutions. To quantitatively measure thevariation from DTI data, we computed the pattern of humanoccipito-callsosal connections. The corpus callosum is a goodexample of anatomically well known structure, as well as withsmall uncertainty level (Jones, 2003). To identify functionallyoriented visual field areas, we used functional MRI (fMRI) incombination with DTI. The upper and lower visual areas in thehuman occipital cortex were used to label the precise pattern ofcross callosal connectivities with a non-invasive manner.

1244 M. Kim et al. / NeuroImage 32 (2006) 1243–1249

Methods

All MR studies were performed on a Philips 3T Interaclinical scanner (Philips system, The Netherlands) using a sen-sitivity encoding (SENSE) phased-array head coil. Subject'sheads were fixed throughout the measurement period by snug-fitting pads.

Eight volunteers (3 female, 5 male, ages 20–30) were scannedusing fMRI, DTI and anatomical sequences on both hemispheres(16 date sets). Written informed consent was obtained from allsubjects, and approved by Boston University School of MedicineInstitutional Review Board participated in the study. All subjectswere given detailed instructions of the experiment both inside andoutside the scanner before actual scan. All experimental procedureswere performed in compliance with guidelines published by theNational Institutes of Health (USA).

fMRI

Functional imaging scans were used to localize retinotopicareas in each subject. Visual stimulus was generated by stimulusdescription language (SDL) on a Visual Stimulus Generator (VSG)(Cambridge Research Systems, Rochester, UK) equipped on PC.The transistor–transistor logic (TTL) signal from the console wasamplified by using Master-8 simulator (A.M.P.I., Israel). It waspresented binocularly with the aid of a video projector and rearprojection screen. Conventional checkerboard type of rotatingwedges were used and consisted of four triangular wedges forupper/lower and left/right visual field (Fig. 2a). Subjects sawrotating wedges for 16 s for each epoch, and stimulus was repeated5 times without baseline.

Each subject was scanned 1–2 times for fMRI with rotatingwedge stimulus. Gradient-echo echo-planar imaging (GE-EPI) wasused with typical imaging parameters (TR, 2 s; TE, 40 ms;128×128 in a FOV of 230.4×230.4 mm2; native resolution,1.8×1.8 mm2 in-plane resolution; slice thickness, 2.0 mm). Thirty2-mm-thick axial slices covered the occipital lobes. Each fMRIscan lasted 7 min 01 s.

All fMRI data were analyzed using BrainVoyager QX (BrainInnovation, Maastricht, Netherlands). The general linear model(Friston et al., 1995) was implemented to determine the strongestactivation and four predictors were used for the model. Eachcorrelated area was mapped in 2D images, and fMRI activationmaps were exported as 2D map which contains thirty slices along zdirection. And then those were imported on top of DTI data forthree-dimensional space.

Diffusion tensor imaging

Diffusion Tensor Imaging (DTI) study was performed bysingle-shot spin-echo EPI (SE-EPI) with conventional parameters(TR, 15.5 s; TE, 91 ms; acquisition resolution, 112×89; imageresolution with 80% of scan percentage, 256×256; 0.9×0.9 mm2

of in-plane image resolution in a FOV of 230.4×230.4 mm2; slicethickness, 1.5 mm). Total acquisition time was 4 min and 55 s withone average. The acquisition time was substantially shortened byusing parallel imaging with the geometric reduction factor 2. It wasreported that parallel acquisition permits both controlling geo-metric distortions and increasing signal-to-noise ratio (SNR)significantly by enhanced encoding speed for echo time reduction(Jaermann et al., 2004). A further improvement on maps of the

trace of the diffusion tensor and of fractional anisotropy (FA) couldbe also accomplished with improved spatial resolution and lessgeometric distortion with single-shot SENSE-EPI (Bammer etal., 2002). Fifteen directions of gradient encoded diffusion-weighted images were used with b=1000 s/mm2, and onereference image with no diffusion weighting was also acquired.For each subject, two raw diffusion-weighted data sets wereacquired and averaged with head motion correction. Co-regis-tration between inter- and intra-raw data of the same subject wasperformed by using a PRIDE registration program (Philips sys-tem, The Netherlands).

DT—data sampling

High-resolution raw data of diffusion weighted image (DWI) inimage space with voxel size of 1.2 (resolution of 0.9×0.9×1.5) mm3

was first rescaled into four different data sets (see details in Fig. 1) byusing bi-cubic interpolation. The down-sampled DT data sets wereused for fiber tracing, and the separate FTs were identified in adifferent spatial resolution.

Fiber-tracking algorithm

Diffusion tensors, FA, and FTs were calculated using custom-written MATLAB (The Mathworks Inc., Natick, MA, USA)software. The tracking algorithm employed is based on themethod described in Basser et al. (2000) and 1-mm fixed stepsize. At any position along the track trajectory, a diffusion tensorwas trilinear-interpolated and eigenvectors were computed. Theeigenvector associated with the greatest eigenvalue indicatedalong this direction over a small distance (<0.5 mm) to the nextpoint where a new diffusion tensor was interpolated. Path tracingproceeded until the FA fell below 0.2 or until the maximumangle between the current and previous path segments was largerthan 60 degrees. Both directions from the initial seed pointprinciple diffusion axis were traced. Each seed point was used toproduce only one fiber trajectory (seed point density equals toone).

Functional localization of tracks

Region of interests (ROIs) were determined using functionalmaps that were obtained from fMRI session, and chosen by strongactivation with p<0.0001 (uncorrected) and then reconstructed as3D volume. All fMRI voxels (above the threshold) were treated ashomologous DTI seeding volume of interests (VOIs) regardless oftheir p-values and/or cross correlation coefficients.

DTI-fMRI images co-registration was confirmed by matchingreproducible landmarks (such as the anterior commissure, thecorpus callosum, the ventricles, and the calcarine circus) betweenthe two types of images superimposed from alignment betweenDTI-anatomy and fMRI-anatomy co-registration.

To investigate the connectivity pattern, tracing was performedfrom four each VOI in loVF and upVF of both hemisphere. Theseed points were laid down in a grid that filled the left or rightoccipital lobe white matter with 1 mm spacing.

Fig. 2b shows the surface tangents of the individual areaswhich were volume rendered to yield three-dimensional recon-struction of the respective polar visual areas for one of the subjects.The panels display the positions of the volume rendered retinotopicvisual areas in a different view. In each hemisphere, around 1000

Fig. 1. The variation of seeding voxel size. The five different resolutions are manipulated from identical DTI data set, and then used as seed VOIs in DTI fibertracing computation. Each resolution is 0.9×0.9×1.5 (voxel volume size of 1.2), 1.2×1.2×2.0 (2.9), 1.8×1.8×1.5 (4.9), 1.8×1.8×3.0 (9.7), 3.6×3.6×1.5(19.4) (unit, mm3).

1245M. Kim et al. / NeuroImage 32 (2006) 1243–1249

estimated seed points were identified by using these parameters. Toevaluate the number of fiber tracts in occipito-callosal region only,additional ROI was placed in the splenium (see Fig. 2c).

To decide on an association with a cortical ROI and the FTswe used the location of the fiber path endpoints. The endpointsare the last point tracked in each of the two initial directionsbefore tracking algorithm terminates. All fibers with an endpointwithin 2 mm of the cortical ROI were found. These fibers wereassociated with the cortical points closest to their endpoints.Fibers with endpoints further than 2 mm from cortical ROIs werenot assigned any functional properties. Because we focused oncallosal projections for these analyses, only those fiber paths thatpassed through the corpus callosum were retained. The amount of

Fig. 2. The functional localization of four cortical visual field maps and estimatewedges was used and consisted of four triangular wedges for upper/lower and left/rthat are volume rendered to yield three-dimensional reconstruction of the respectivvisual field (loVF) (blue) in LH (top image) and upVF (green), loVF (magenta) in Rof T1-weighted (T1W) image at one volunteer. In panel (b), the lower right Inset (corresponding to the occipital visual field in subject. Panel (c) visualizes the cross cfiber tracts in occipito-callosal region, the second ROI was placed in the splenium (to the posterior view of T1W images represent occipito-callosal fiber tracts passiinterpretation of the references to colour in this figure legend, the reader is referre

those specific paths was around 10 to 20% of the total left orright occipital lobe fibers.

Anatomical Data and co-registration

T1-weighted high resolution (0.9×0.9×1.0 mm3) anatomicalimages were obtained for each subject to allow accurate corticalsegmentation. Each diffusion-weighted slice was aligned with thecorresponding anatomical image by a rigid body transformationand fMRI-anatomy images were co-registered by full affinetransformation on BrainVoyager QX (Brain Innovation, Maas-tricht, Netherlands). Corresponding fMRI and DTI tracing data

d cross-callosal fiber tracts. (a) Conventional checkerboard type of rotatingight visual field. Panel (b) shows the surface tangents of the individual VOIse visual areas for one of the subjects. Upper Visual field (upVF) (red), lowerH (bottom image) are superimposed on the three dimensional sagittal viewswhite box) shows the colored localization of polar visual stimulus which isallosal FTs in high resolution DTI data (1.2 mm3). To evaluate the number ofindicated by white arrows). Partially zoomed areas in Insets (white box) nextng through the splenium. The directional information is not included. (Ford to the web version of this article.)

1246 M. Kim et al. / NeuroImage 32 (2006) 1243–1249

were superimposed on three-dimensional anatomical images forvisualization, subsequently.

Results

Here we describe the estimated FTs in the occipital–callosaltracts by combining fMRI seed VOIs and diffusion tensormodalities as it was shown in our previous study that using fMRIseeds is a unique method providing an estimate betweenfunctionally evoked landmarks for the fiber tracking (Kim et al.,2006). Our first experimental goal was to identify the occipitallobe FTs with a high resolution DTI data. We defined four VOIs(upper/lower/left/right fields) based on functional structure inoccipital lobe (see Fig. 2b), and then used fiber trackingalgorithms (see Fiber-tracking algorithm) to estimate cross callosalFTs. To assess the reliability of the DTI-FT measurements,occipital–callosal projections were estimated from each subject'sleft and right hemispheres (LH and RH) independently.

Pattern of the estimated occipital–callosal FTs

The Fig. 2c shows that most of the estimated FTs tend to projectto the frontal lobe/ the contralateral occipital lobe and widelydistributed occipital fibers converge to cross the callosum as acompact fiber bundle confined to the splenium from each VOI with

Fig. 3. The number of seed points at each VOI and the absolute number of estimatshow the intersubject seed coordinate variability in subject 1 (S1) and subject 2 (Svaries. Panels b and d, the estimated number of FTs decreases significantly as the spin number of FTs seeded between upVF and loVF exists. The clear difference candecreases, the number of seed points drops, however, not necessarily proportionat

image voxel size of 1.2 mm3. It also represents the visualization ofthe asymmetric cross callosal FTs between upVF and loVF. Panelsshow that occipito-callosal FTs originated from seed points loVF(blue) and upVF (red) in LH, loVF (magenta) and upVF (green) inRH project to the corresponding contralateral regions respectively,passing through the splenium.

Voxel-based assessment of estimated number of FTs

For computing fiber tracing with the variation of image voxelsize, five different resolutions were manipulated from identicalDTI data set. The image voxel size of 1.2, 2.9, 4.9, 9.7 and19.4 mm3 were used (see Fig. 1). For the comparison to be fairlyperformed, we used the same functional maps and spatialcoordinates of the VOIs in all of our calculations. The onlydifference was in the resolution of the various DTI data sets. In thisway, DTI data sets differing in spatial resolution had the same VOIcoverage with a different number of seed points per each VOI.

The graphs in Figs. 3a and c show the number of seed VOIs withrespect to the spatial sampling resolution of DTI data that is used forthe fiber tracing from subject 1 (S1) and 2 (S2), respectively. Thenumber of seed points varies gradually with the spatial resolution asexpected.

The graphs in Figs. 3b (S1) and d (S2) show the absolutenumber of estimated FTs propagated from VOIs with respect to

ed FTs with respect to each resolution. Panels a and c, the graphs in the left2), respectively, as the spatial sampling resolution of the DTI measurementatial density of the seed points goes down. It is observed that the asymmetrybe seen between seed VOIs originating from LH and RH. As the resolutione with the estimated number of FTs as indicated by arrows.

1247M. Kim et al. / NeuroImage 32 (2006) 1243–1249

each voxel resolution. Our data show that the estimated numberof FTs varies with the spatial density of the seed points used tocompute the algorithm from all subjects (see details of the entiresubjects in the Supplemental Fig. 1 and 2). It is, however,observed that the number of fibers beyond image voxel size of4.9 (1.8×1.8×1.5) mm3 drops dramatically at loVF of bothhemispheres, which results in significantly smaller difference inasymmetric occipital–callosal FTs between upVF and loVF (seedetails in Supplemental Fig. 2). It is also observed that under-sampling of voxel size affects widely distributed FTs to the frontallobe (figure omitted here).

Comparison of right and left occipital–callosal FTs

The clear difference can be seen between seed VOIsoriginating from LH and RH at each resolution (see Figs. 3aand c) which does not necessarily correspond to the number ofFTs seeded between upVF and loVF (see Figs. 3b and d). Fiveout of eight subjects, the number of FTs of the right occipital lobeis slightly larger than the left (Supplemental Fig. 2), whereas onlyone subject out of eight has a larger number of seed points in theRH (Supplemental Fig. 1).

FA measurement

Fig. 4 shows the average and the standard error of FA on thecortico-splenium FTs through all subjects. All FA values of cortico-splenium FTs are high around in between 0.6 and 0.7.

The average FAvalues through all different spatial resolution arenot significantly different at the 0.05 level with one-way ANOVAanalysis (FTs connecting splenium with LH-loVF (blue), p=0.25;LH-upVF (red), p=0.74; RH-loVF (magenta), p=0.52; RH-upVF(green), p=0.72). However, an increase in standard error isobserved as well as a reduction on the variation of mean FA fromdifferent VOIs when the spatial voxel resolution increases.

Fig. 4. The average and the standard error of FA on the cortico-splenium FTsthrough all subjects. All FA values are high around in between 0.6 and 0.7,and the average FA values through all different spatial resolution were notsignificantly different at the 0.05 level with one-way ANOVA. However, thevariation of each FAvalue at different VOI gets smaller while standard errorbecomes larger, as the spatial resolution increases (color labeled as blue forLH-loVF, red for LH-upVF, magenta for RH-loVF, and green for RH-upVF). (For interpretation of the references to colour in this figure legend,the reader is referred to the web version of this article.)

Discussion

In the present study, the dependence of fiber tracking results onthe spatial resolution of DTI data was studied in human occipito-callosal region, using structure–function relationships that wereassessed with the non-invasive techniques of fMRI and DT-MRI.Since this study is mostly concerned with comparing fibertracking results under different spatial resolution conditions, thefiber tracking method chosen for this study was the straightfor-ward stream-line method.

Our results suggest that the intersubject seed coordinate varieswith respect to the spatial sampling resolution of the DTI data,while an apparent difference in the estimated occipital FTs betweenupVF and loVF exists which is not proportionately matched to thenumber of seeds. This implies that the number of seed points is notthe only factor lowering the number of FTs.

Correlation between asymmetric FTs estimation and spatialresolution

While a recent study by Dougherty et al. reported thatasymmetric representation of fibers between loVF and upVF inone hemisphere can be attributed to sensitivity problems associatedwith the native DTI data (Dougherty et al., 2005), it is not clearwhether the asymmetry is indeed an artifact. As some of studies, infact, support the asymmetric population between two VFs, there isevidence to suggest that differences in visual field representationbetween dorsal and ventral streams may also be apparent across theupper and lower portions of the visual field. In both humans andnon-human primates, the superior hemiretina has a greater densityof ganglion cells than the inferior hemiretina, suggesting a biastowards processing of visual stimuli in the loVF (Curcio and Allen,1990). Especially, in humans, visual evoked potential andmagnetoencephalographic studies have found stronger signals inoccipital cortex from loVF stimulation (Fioretto et al., 1995; Portinet al., 1999). This is in agreement with our results presented here.As can be observed from our data (see Results: Voxel-basedassessment of estimated number of FTs), the asymmetry ofoccipital-callosal FTs between upVF and loVF tend to appearclearly at higher spatial resolution, however it becomes tosignificantly decrease when the number of FTs from loVF dropswith the fibers from upVFs stayed roughly the same at lowerresolution. However, using only the streamline tracking algorithm,it is not sufficient to conclude whether this asymmetry is a truefinding or a fiber-tracking related artifact. One way to elucidate thisspecific question is to perform a comparative study usingprobabilistic mapping (Parker et al., 2003).

Contradiction on SNR and spatial resolution

Two important factors for the accuracy of the quantitativemeasurement are the SNR and the spatial resolution, whichgenerally conflict each other. The improved spatial resolution isalso crucial in resolving crossing fiber. In general, signal-to-noiseratio (SNR) is proportional to the square-root of the number ofsignal averaging:

SNR ¼ Dx0dDy0dDz0d

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

BWdNxdNydNzdNavg

rð1Þd f (TR,TE,: : :)

Here, BW, Nx, Ny, Navg, and f (TR, TE, …) are the receiverbandwidth and imaging matrixes for read-out, phase-encoding,

1248 M. Kim et al. / NeuroImage 32 (2006) 1243–1249

number of averages, and signal weighting function, respectively. Inconventional MR imaging techniques such as spin-echo imaging,the higher spatial resolution by increasing the imaging matrixdecreases the number of signal averaging for a given imaging time.However, because DTI using single-shot 2D diffusion weightedEPI acquires the complete k-space in a single shot with therepetition time of typically 7–8 s for the whole brain, the number ofaveraging does not significantly change with different matrix size.Therefore, the effect of the number of signal averaging on SNR canbe neglected for the current study. Upon re-sampling of 256×256imaging matrix down to 128×128 with twice thickness and thesame number of averages, Eq. (1) becomes,

SNR ¼ 2Dx0d2Dy0d2Dz0d

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

BWdNx

2dNy

2dNz

2dNavg

rd f (TR,TE,: : :)

(2)

Eq. (2) for the SNR holds for both acquired and resampled images.

FA variability and partial voluming

FA plays an important role as a threshold parameter that isused to terminate the generation of a fiber. Thus, overestimationof FA in white matter tractography results in false, artificialgeneration of fibers and increased noise contamination. FA is anintravoxel measure that yields values between 0 (purely isotropicdiffusion) and 1 (purely anisotropic diffusion) (Basser andPierpaoli, 1996; Papadakis et al., 1999). Recent report suggestedthat increasing resolution results in a reduced SNR ratio, whichmay affect FA estimation since there is a systematic over-estimation of diffusion anisotropy as noise increases (Werring et al.,1999). The effect of rician noise (noise average>0) tends todecrease the apparent diffusion coefficient (ADC) calculatedwhen the gradient is parallel to the fiber, and thus cause anunderestimation of FA (Jones and Basser, 2004). Our result,however, shows that larger spatial resolution results in an increaseof the standard error of the FA estimation (see Fig. 4). Theincrease in standard error of the FA estimation with lowerresolution along the tract may correlate such that partial volumeeffect with surrounding gray matter becomes more significant.Alexander et al. (2001) suggested in their report that those partialvolume effects may have led to FA variations on the order of 10–20% for the large voxel case (spatial resolution of 64×64) whichis quite consistent with our result (see Fig. 4). That may also leadto a general lowering of the FA with increased voxel size.Additionally, the overestimation of FA as described earlier ismore pronounced in areas with low FA, where the noise can beinterpreted as a random addition of directionality to a generallylow FA value. This implies that FA value does not significantlydecrease along the fiber path due to the low spatial resolution. Itmight be possible that number of diffusion tensor ellipsoid turnsto smaller, as the voxel size becomes larger in a limited space(brain and/or assigned VOI). And that presumably can cause theerroneous or loose fiber connections.

Diffusion tensor partial voluming can also significantlyinfluence measurements of diffusion anisotropy and tensor shape(Alexander et al., 2001). For the large voxel case investigatedhere, missing FTs from lower resolution data presumablyimplicate that two or more distinct tissues with different diffusion

tensors may have occupied the same voxel. The most obviousapproach for the minimization of partial volume effects is todecrease the voxel size as suggested in Alexander et al., whereasthis will decrease the image SNR. However, averaging DT signalscould improve SNR, thereby compromising loss of signal inducedby high resolution data. This could also lead to improve theaccuracy of measurement of the number of possible fibers and thegoodness of these fibers.

In conclusion, our results demonstrate the impact of variousspatial sampling resolutions on DTI and fiber tracking in humanoccipito-callosal region. In general, it is expected that increasingvoxel size results in higher SNR as well as larger partial volumeeffects, and finally results in a reduction on the number of fibertracts. Partial volume effect particularly corresponds to theaccuracy of FA estimation, resulting increased FA variation anderroneous fiber connections. In this report, we propose that theasymmetric upper and lower fiber trajectories can be recon-structed with an acquisition resolution of up to 2 mm iso-voxel inoccipito-callosal region. However, it should be noted that theoptimal native resolution will depend on several additionalfactors, such as fiber geometry, SNR of the data sets, the size ofthe VOIs used as starting regions for fiber tracking, and thenumber of seeding point used in each VOI. Furthermore, theapplication of a given imaging resolution may further improvetissue characterization through the quantification of FA andapparent diffusion coefficient indices.

Acknowledgments

Authors thank Dr. Susumu Mori for helpful comments on themanuscript, Harish Sharma and Dr. Mathieu Ducros for theirtechnological assistance. This work was supported by NIH(RR08079, NS44825 and EB00331), the Keck Foundation, andthe Human Frontiers Science Program.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.neuroimage.2006.06.006.

References

Alexander, A.L., Hasan, K.M., Lazar, M., Tsuruda, J.S., Parker, D.L., 2001.Analysis of partial volume effects in diffusion-tensor MRI. Magn.Reson. Med. 45 (5), 770–780.

Bammer, R., Auer, M., Keeling, S.L., Augustin, M., Stables, L.A., Prokesch,R.W., Stollberger, R., Moseley, M.E., Fazekas, F., 2002. Diffusion tensorimaging using single-shot SENSE-EPI. Magn. Reson. Med. 48 (1),128–136.

Basser, P.J., Pierpaoli, C., 1996. Microstructural and physiological featuresof tissues elucidated by quantitative-diffusion-tensor MRI. J. Magn.Reson., B 111 (3), 209–219.

Basser, P.J., Mattiello, J., LeBihan, D., 1994. MR diffusion tensorspectroscopy and imaging. Biophys. J. 66 (1), 259–267.

Basser, P.J., Pajevic, S., Pierpaoli, C., Duda, J., Aldroubi, A., 2000. In vivo fibertractography using DT-MRI data. Magn. Reson. Med. 44 (4), 625–632.

Catani, M., Howard, R.J., Pajevic, S., Jones, D.K., 2002. Virtual in vivointeractive dissection of white matter fasciculi in the human brain.NeuroImage 17 (1), 77–94.

Conturo, T.E., Lori, N.F., Cull, T.S., Akbudak, E., Snyder, A.Z., Shimony,J.S., McKinstry, R.C., Burton, H., Raichle, M.E., 1999. Tracking

1249M. Kim et al. / NeuroImage 32 (2006) 1243–1249

neuronal fiber pathways in the living human brain. Proc. Natl. Acad.Sci. U. S. A. 96 (18), 10422–10427.

Curcio, C.A., Allen, K.A., 1990. Topography of ganglion cells in humanretina. J. Comp. Neurol. 300 (1), 5–25.

Dougherty, R.F., Ben-Shachar, M., Bammer, R., Brewer, A.A., Wandell, B.A.,2005. Functional organization of human occipital-callosal fiber tracts. Proc.Natl. Acad. Sci. U. S. A. 102 (20), 7350–7355.

Fioretto, M., Gandolfo, E., Orione, C., Fatone, M., Rela, S., Sannita, W.G.,1995. Automatic perimetry and visual P300: differences between upperand lower visual fields stimulation in healthy subjects. J. Med. Eng.Technol. 19 (2–3), 80–83.

Friston, K.J., Holmes, A.P., Worsley, K.J., Poline, J.B., Frith, C.D.,Frackowiak, R.S.J., 1995. Statistical parametric maps in functionalimaging: a general linear approach. Hum. Brain Mapp. 2, 189–210.

Jaermann, T., Crelier, G., Pruessmann, K.P., Golay, X., Netsch, T., vanMuiswinkel, A.M., Mori, S., van Zijl, P.C., Valavanis, A., Kollias, S., et al.,2004. SENSE-DTI at 3 T. Magn. Reson. Med. 51 (2), 230–236.

Jones, D.K., 2003. Determining and visualizing uncertainty in estimates offiber orientation from diffusion tensor MRI. Magn. Reson. Med. 49 (1),7–12.

Jones, D.K., Basser, P.J., 2004. “Squashing peanuts and smashingpumpkins”: how noise distorts diffusion-weighted MR data. Magn.Reson. Med. 52 (5), 979–993.

Kim, D.S., Kim, M., Ronen, I., Formisano, E., Kim, K.H., Ugurbil, K.,

Mori, S., Goebel, R., 2003. In vivo mapping of functional domains andaxonal connectivity in cat visual cortex using magnetic resonanceimaging. Magn. Reson. Imaging 21 (10), 1131–1140.

Kim, M., Ducros, M., Carlson, T., Ronen, I., He, S., Ugurbil, K., Kim, D.S.,2006. Anatomical correlates of the functional organization in the humanoccipitotemporal cortex. Magn. Reson. Imaging 24 (5), 583–590.

Mori, S., Crain, B.J., Chacko, V.P., van Zijl, P.C., 1999. Three-dimensionaltracking of axonal projections in the brain by magnetic resonance imaging.Ann. Neurol. 45 (2), 265–269.

Papadakis, N.G., Xing, D., Houston, G.C., Smith, J.M., Smith, M.I., James,M.F., Parsons, A.A., Huang, C.L., Hall, L.D., Carpenter, T.A., 1999. Astudy of rotationally invariant and symmetric indices of diffusionanisotropy. Magn. Reson. Imaging 17 (6), 881–892.

Parker, G.J., Haroon, H.A., Wheeler-Kingshott, C.A., 2003. A frameworkfor a streamline-based probabilistic index of connectivity (PICo) using astructural interpretation of MRI diffusion measurements. J Magn. Reson.Imaging 18 (2), 242–254.

Portin, K., Vanni, S., Virsu, V., Hari, R., 1999. Stronger occipital corticalactivation to lower than upper visual field stimuli. Neuromagneticrecordings. Exp. Brain Res. 124 (3), 287–294.

Werring, D.J., Clark, C.A., Parker, G.J., Miller, D.H., Thompson, A.J.,Barker, G.J., 1999. A direct demonstration of both structure and functionin the visual system: combining diffusion tensor imaging with functionalmagnetic resonance imaging. NeuroImage 9 (3), 352–361.