Original Research

Diffusion-Tensor MRI Reveals the Complex MuscleArchitecture of the Human Forearm

Martijn Froeling, MS,1,2* Aart J. Nederveen, PhD,2 Dennis F.R. Heijtel, MS,1,2

Arno Lataster, MS,3 Clemens Bos, PhD,4 Klaas Nicolay, PhD,1 Mario Maas, MD, PhD,2

Maarten R. Drost, PhD,5 and Gustav J. Strijkers, PhD1

Purpose: To design a time-efficient patient-friendly clini-cal diffusion tensor MRI protocol and postprocessing toolto study the complex muscle architecture of the humanforearm.

Materials and Methods: The 15-minute examination wasdone using a 3 T system and consisted of: T1-weightedimaging, dual echo gradient echo imaging, single-shotspin-echo echo-planar imaging (EPI) diffusion tensor MRI.Postprocessing comprised of signal-to-noise improvementby a Rician noise suppression algorithm, image registra-tion to correct for motion and eddy currents, and correc-tion of susceptibility-induced deformations using mag-netic field inhomogeneity maps. Per muscle one to fiveregions of interest were used for fiber tractography seed-ing. To validate our approach, the reconstructions of indi-vidual muscles from the in vivo scans were compared tophotographs of those dissected from a human cadaverforearm.

Results: Postprocessing proved essential to allow musclesegmentation based on combined T1-weighted and diffu-sion tensor data. The protocol can be applied more gener-ally to study human muscle architecture in other parts ofthe body.

Conclusion: The proposed protocol was able to visualizethe muscle architecture of the human forearm in greatdetail and showed excellent agreement with the dissectedcadaver muscles.

Key Words: diffusion tensor imaging; skeletal musclearchitecture; forearm; segmentationJ. Magn. Reson. Imaging 2012; 000:000–000.VC 2012 Wiley Periodicals, Inc.

THE STRUCTURAL ARRANGEMENT of skeletal mus-cle is described by its fiber architecture. The musclearchitecture, which is defined as the arrangement ofmuscle fibers relative to the axis of force generation, isthe main determinant of mechanical muscle function.Parameters that are typically used to describe musclearchitecture are muscle volume, pennation angle, andfiber length (1). Detailed knowledge of muscle architec-ture, obtained in a noninvasive manner, thereforecould have profound functional and clinical signifi-cance (2). In recent years diffusion-tensor magneticresonance (MR) imaging (DTI) has developed into amethod that enables noninvasive in vivo 3D assess-ment and visualization of the muscle fiber architecture.DTI exploits the property that the apparent diffusivityof water is greatest along the dominant muscle fiberdirection. In each imaging voxel a diffusion tensor isreconstructed from a series of diffusion-weighted MRimages along at least six independent diffusion-encod-ing directions. An eigenvalue analysis yields the princi-pal axis of diffusion, which parallels the local musclefiber orientation (3). Principal diffusion directions ofneighboring voxels are combined for 3D muscle fibertractography (4–6). Fiber architectural parameterssuch as muscle fiber length, physiological cross-sec-tional area, and pennation angle, which are tradition-ally obtained by 2D ultrasound (7–9), can be obtainedwith DTI with great accuracy in a 3D setting (10–12).Potential clinical application of the DTI technique canbe found, among others, in the planning of tendontransposition surgery (13), monitoring of degenerativemuscle disease (14,15), evaluation of acute muscleinjury (16,17), and determination of subject specificmuscle parameters for biomechanical models (18).

DTI and muscle-fiber tractography have beenapplied previously with the aim to investigate musclearchitecture in vivo (10–12,19–25). These studiesmainly focused on the large muscles of the calf and

1Biomedical NMR, Department of Biomedical Engineering, EindhovenUniversity of Technology, Eindhoven, The Netherlands.2Academic Medical Center, Department of Radiology, Amsterdam, TheNetherlands.3Maastricht University, Department of Anatomy and Embryology,Maastricht, The Netherlands.4Philips Medical Systems, Best, The Netherlands.5Maastricht University, Department of Human Movement Science,School for Nutrition, Toxicology and Metabolism, Maastricht,Maastricht, The Netherlands.

Additional Supporting Information may be found in the online versionof this article.

*Address reprint requests to: M.F., Department of Radiology, Aca-demic Medical Center, Meibergdreef 9, Z0-178, 1105 AZ Amsterdam,The Netherlands. E-mail: M.froeling@amc.uva.nl

Received September 29, 2011; Accepted January 11, 2012.

DOI 10.1002/jmri.23608View this article online at wileyonlinelibrary.com.

JOURNAL OF MAGNETIC RESONANCE IMAGING 000:000–000 (2012)

CME

VC 2012 Wiley Periodicals, Inc. 1

the thigh, and commonly involved lengthy imagingprotocols. For example, a DTI acquisition of 9 minuteswas used to acquire 20 to 30 slices with 13 diffusion-encoding directions (24,25). The aim of this work wasto design a time-efficient, clinically usable diffusion-tensor MRI protocol to reveal the complex musclearchitecture of the human forearm. The protocol wasdeveloped using a 3 T MRI system and healthy volun-teers. DTI of the human forearm is particularly chal-lenging because the forearm contains 19 smaller andlarger muscles, most of which possess a complicatedcurved and multipennate fiber organization. DTIimages were acquired using a single-shot echo-planarimaging (EPI) sequence. With EPI an image can beacquired within a fraction of a second and thus ‘‘freez-ing’’ macroscopic motion to a large extent. However,diffusion-tensor EPI faces two major obstacles. First,the cross-section of the human forearm is small andanatomy requires off-center positioning of the arm inthe MRI scanner. This leads to susceptibility-inducedtranslation and deformation of images and diffusiontensor, and misregistration with T1- and T2-weightedimages. Second, at 3 T skeletal muscle has a relativelyshort T2 relaxation time of about 30 to 50 msec(26,27), which results in a low signal-to-noise ratio(SNR) with commonly used echo times ranging from42 to 104 msec (10,11,14,16,17,19,28).

We have succeeded in developing a DTI protocolthat enables visualization of the complex musclearchitecture of the human forearm in great detail,without sacrificing scan time, patient comfort, andreliability. The protocol presented in this article con-sists of three sequences; high-resolution T1-weightedimaging for anatomical reference, dual-echo gradient-echo imaging, and single-shot spin-echo (SE-EPI) dif-fusion-tensor MRI. Total acquisition time was �15minutes. SNR is improved by using a Rician-noise

suppression algorithm. Motion and eddy currentsinduced geometric distortions were corrected usingaffine registration. The dual-echo gradient-echoimages are reconstructed to derive magnetic fieldinhomogeneity maps, which serve to correct the EPIimages as well as the diffusion tensor for susceptibil-ity-induced deformations. Fiber tractography isapplied to the whole forearm using a low number ofseeding regions of interest (ROIs) for each individualmuscle. To validate our approach, the fiber recon-structions of individual muscles from the in vivoscans were compared to photographs of those dis-sected from a human cadaver forearm.

MATERIALS AND METHODS

Volunteers

MRI protocol optimization was performed on fivehealthy untrained male volunteers (23 6 2 years, 786 8 kg, 185 6 7 cm) using a 3 T Philips Intera scan-ner (Philips Medical Systems, Best, The Netherlands).The representative images in this article are of a 26-year-old subject. All subjects were screened for MRI-risk factors and provided written consent prior to thestudy. The research was approved by the institutionalEthics Committee Review Board.

MRI Protocol

Subjects were placed in the scanner in a prone positionwith their right arm above the head as schematicallyshown in Fig. 1A. We chose this position, which waswell tolerated by the volunteers, to place the arm asmuch as possible in the iso-center of the magnet andminimize inclusion of other tissues in the field of view(FOV). For signal reception four flexible surface coils

Figure 1. Experimental setupand MRI protocol. A: Sche-matic top view of the subjectpositioning in the MRI scan-ner. B: Top and volar views ofRF-coil placement around theforearm. C: T1-weighted TSEimaging for anatomical refer-ence. D: Dual-echo GE phasemapping for B0-inhomogeneitycorrections. E: Diffusion-ten-sor SE-EPI imaging (b ¼ 0 s/mm2).

2 Froeling et al.

(2� circular diameter ¼ 20 cm and 2� elliptical diame-ter ¼ 14 � 17 cm) were placed below and above theforearm as illustrated in Fig. 1B. The MRI examinationconsisted of three acquisitions; T1: T1-weighted turbo-spin-echo (TSE) imaging for anatomical reference (Fig.1C), GE: dual-echo gradient-echo (GE) imaging withphase reconstruction to derive a B0-field inhomogene-ity map (Fig. 1D), and DTI: diffusion-tensor SE-EPI forfiber tractography (Fig. 1E). The acquisition times were4 minutes 57 seconds, 1 minute 24 seconds, and 7minutes 16 seconds, respectively, resulting in a totalprotocol time of 13 minutes 37 seconds.

In all three acquisitions, 60 slices of 5-mm thick-ness were acquired with an FOV of 192 mm in the fre-quency-encoding direction and 120 mm in the phase-encoding direction. The smaller FOV in the phase-encoding direction allowed for a low number of phase-encoding steps without sacrificing resolution. Thisreduced scan time and minimized the SE-EPI echotrain length, which was beneficial for minimizing sus-ceptibility-induced distortions. Imaging parameterswere; T1: sequence: TSE, voxel size: 0.5 � 0.5 � 5mm3, TR/TE: 550/12 msec, number of signal aver-ages (NSA): 2; GE: sequence: GE, voxel size: 2 � 2 � 5mm3, matrix size 96 � 60, TR/TE1/TE2: 12/4.6/9.6msec, NSA: 2; DTI: sequence: SE-EPI with Stejskaland Tanner pulsed field gradients, voxel size: 2 � 2 �5 mm3, matrix size 96 � 60, NSA: 2, 15 diffusion gra-dient directions (29), TR/TE: 8800/41 msec, b ¼ 400s/mm2, G: 61 mT/m, D/d: 20/9.4 msec, halfscan:0.625, EPI-train length: 37, fat suppression: spectraladiabatic inversion recovery (SPAIR).

Postprocessing

Postprocessing of the data was done using a custom-built toolbox for Wolfram Mathematica 8 (WolframResearch, Champaign, IL) and consisted of four steps:1) application of a Rician noise-suppression algorithm(30) to the diffusion-weighted images; 2) registration ofthe diffusion-weighted images (Sn, n ¼ 1, 2, . . ., 15, b ¼400 s/mm2) to their corresponding nonweighted image(S0, b ¼ 0 s/mm2); 3) phase unwrapping (31) of the GEimages to derive a DB0-field map (32) and calculation ofthe displacement field (33); 4) Translation correction ofthe diffusion tensor based on this displacement field(34). The steps are explained in more detail below.

Rician Noise Suppression

The echo time of the SE-EPI sequence was 41 msec,which is of the same order as the T2 relaxation time ofmuscle at 3 T. As a result, the DTI images of all thedatasets had an average SNR of 24 for the non-weighted images (Fig. 2A) and 13 for the diffusion-weighted images (35,36) before application of thenoise suppression algorithm. In order to decrease thenegative effect of noise on the fiber tractography qual-ity, we applied the noise suppression technique intro-duced by Aja-Fernandez et al (30). The algorithm usesa Rician linear minimum mean square error (LMMSE)estimator, which is convolved with the DTI imagesusing a 5 � 5 pixels Gaussian kernel. The noise var-

iance s2, which is needed as input for the LMMSE es-timator, was estimated from the background noiselevel. The effect of the algorithm is shown in Fig. 2B.The difference between the original and noise-sup-pressed image is depicted in Fig. 2C.

Image Registration

To correct for motion and eddy currents induced by geo-metric distortions the diffusion-weighted images wereregistered to the nonweighted images using an affinetransformation. The 3D registration was a normalizedgradient fields-based algorithm and used Gauss-New-ton minimization (37,38). The registration was accom-panied with the appropriate b-matrix reorientation (39).

Magnetic Field Calculation and InhomogeneityCorrection

The dual-echo GE acquisition (Fig. 1D) produced twosets of phase images, one with echo time TE1 ¼ 4.6msec (u1, Fig. 2D) and one with TE2 ¼ 9.6 msec (u2,Fig. 2E). Dephasing in each voxel during the 5 msecbetween TE1 and TE2 is related to the magnetic fieldinhomogeneity DB0, according to:

gDB0 ¼ u2 � u1

TE2 � TE1: ½1�

Before Eq. [1] could be applied, the phase images asshown in Fig. 2D,E needed phase unwrapping, becausethey suffered from 2p phase jumps. For this purposewe implemented the phase-unwrapping algorithmintroduced by Herraez et al (31). Basically, the algo-rithm classifies pixels on their position in the imagewith respect to phase jumps. The larger homogeneousparts in the phase image are taken as the starting pointof the phase unwrapping procedure, followed byregions with more phase jumps. We extended the algo-rithm to exclude background pixels, which we obtainedfrom a binary mask of the T1-weighted images.

From the resulting magnetic field map DB0, the dis-placements in the frequency-encoding direction dpy andin phase-encoding directions dpxwere calculated using:

dpy ¼ gDB0N

BW; ½2�

and

dpx ¼ gDB0NM

BW¼ dpyM : ½3�

Here, g is the proton gyromagnetic ratio, BW thepixel bandwidth, and N and M the number of k-linesin frequency- and phase-encoding directions, respec-tively (33). DTI acquisitions were performed with M ¼60 phase-encoding k-lines, which means that dis-placements in the frequency-encoding direction dpy

are a factor 60 smaller than the displacements in thephase-encoding direction dpx. Therefore, displace-ments in the frequency-encoding direction were re-stricted to the subvoxel level and were neglected. Fig-ure 2F shows the DB0 map determined from the 2 GE

DTI of Human Forearm Muscle Architecture 3

Figure 2. Postprocessing of raw image data. A: Diffusion-tensor SE-EPI imaging (b ¼ 0 s/mm2) before noise suppression. B:Diffusion-tensor SE-EPI imaging (b ¼ 0 s/mm2) after noise suppression. C: Difference between A and B. D: GE phase imagewith TE ¼ 4.6 msec. E: GE phase image with TE ¼ 9.6 msec. Note the 2p phase jumps that need unwrapping. F: DB0-map,calculated from images A and B after phase unwrapping. G: T1-weighted image. The dashed red lines contour the muscles,the radius, and the ulna. H: Uncorrected SE-EPI image from the DTI acquisition, overlaid with the contours of G. A clear mis-registration with the contours is visible. The green lines depict the displacement calculated from the DB0-map. Mismatch islargest in regions of high displacement (inset). I: After correction the registration between the SE-EPI and the contours of theT1-weighted image is markedly improved. J: Dxx tensor component of raw data. K: Dxx tensor component after postprocess-ing. L: Color-coded FA map of raw data. M: Color-coded FA map after postprocessing. [Color figure can be viewed in theonline issue, which is available at wileyonlinelibrary.com.]

4 Froeling et al.

phase images shown in Fig. 2D,E. After correction forthe displacements in the phase-encoding direction,images were resampled on a Cartesian grid using lin-ear interpolation.

Fiber Tractography and Muscle Segmentation

Fiber tractography was performed using the DTIToolprogram (http://bmia.bmt.tue.nl/software/dtitool)developed in-house (40). Fiber tracts continued bidir-ectionally (0.2 voxel integration steps) from a seedingROI until empirically determined stopping criteriawere satisfied (fractional anisotropy [FA] <0.1 or >0.5,an angle change >5 degrees/integration step). Theseeding ROIs were drawn based on the T1-weightedimages. With fiber tractography in the brain one caneasily distinguish different structures based on a col-ored FA map, but in the forearm most of the fibertracts run in the same direction (see Fig. 2L,M). Thismakes it impossible to identify different musclesbased on the FA images and fiber tracts alone andthus the anatomical information of the T1-weightedimages proved necessary. For each of the 19 forearmmuscles between 1 and 5 seeding ROIs were drawn,amounting to on average 56 ROIs in total per arm.

Diffusion Parameters

After segmentation, for each muscle mean values offive DTI parameters were calculated, the three eigenval-ues (l1, l2, and l3), the mean diffusivity (MD), and thefractional anisotropy (FA). The appropriate voxels wereselected based on the fiber tracts. The mean value andstandard deviation of the parameters for the five sub-jects were calculated for six different muscles—Prona-tor teres (PT), Flexor carpi ulnaris (FCU), Flexor Digito-rum Profundus (FDP), Pronator quadratus (QUA),Extensor digitorum communis (EDC), Supinator(SUP)—and the whole muscle volume (WMV).

Cadaver Dissection and Photography

Dissection was performed on a formalin-fixed humancadaver right forearm of an elderly person. After re-moval of the skin and subcutis, 19 individual musclesof the forearm were carefully dissected. Each musclewas isolated by detaching the proximal bony insertionand transecting the distal tendon. Twenty photo-graphs of each isolated muscle were taken at differentangles around the proximal–distal axis of the forearm.The photographs of each muscle were merged, creat-ing a movie showing the muscle rotating around theproximal–distal axis. These animations were com-pared with the segmented fiber tracts.

RESULTS

Postprocessing

Figure 2F shows a representative displacement field ofa single slice, calculated from the DB0 map. The colorrepresents the calculated displacement map in pixelsaccording to the scale on the left and the overlay of

black lines visually illustrates the actual displace-ment. The displacement maps of all slices in thisdataset are available as an animation in the Support-ing Digital Content 1, which provides an overview ofthe geometrical distortions caused by field inhomoge-neities throughout the entire forearm.

The improvements introduced by the displacementcorrections are exemplified in Fig. 2G–I. Figure 2Gshows a T1-weighted image with three contours(dashed red lines) enclosing the muscle compartmentsof the forearm and the radius and ulna. In Fig. 2H thecorresponding SE-EPI image is overlaid with the samecontours. There is a clear misregistration between thecontours and the SE-EPI image, induced by geometri-cal distortions. The green lines in Fig. 2H depict thedisplacement calculated from the DB0 map. The mis-match between the contours and the SE-EPI image islargest in regions of high displacement (inset in Fig.2H). After correction (Fig. 2I) the agreement of thecontours of the T1-weighted image and the SE-EPIimage is markedly improved.

Figure 2J,K shows the Dxx tensor component before(Fig. 2J) and after (Fig. 2K) postprocessing. One caneasily see the improvement of the quality of the tensorwithout losing detail. The improvement can alsoclearly be seen in the color-coded FA map shown inFig. 2L (before processing) and M (after processing).

The postprocessing steps also greatly improved thefiber tracking of the muscles in the human forearm.Figure 3 displays fiber tractography of the wholehuman forearm before (Fig. 3A,C) and after (Fig. 3B,D)displacement. The fiber tracts are shown on top of or-thogonal cross-sections and surface renderings of theT1-weighted images of the forearm. Regions where T1-weighted data and fiber tracts were severely misregis-tered are indicated with red arrows. After postprocess-ing the misregistration virtually disappeared anddenser fiber tracking continued toward the proximalend of the forearm (indicated by the red asterisk). Thisis made clearer in Fig. 3E,F, where the Flexor digitorumprofundus muscle of the five subjects is shown before(Fig. 3E) and after (Fig. 3F) postprocessing. For everymuscle two axially placed seeding ROIs were used atone- and two-thirds of the length of the muscle.

Muscles of the Human Forearm

All 19 muscles of the human forearm could be seg-mented on the basis of the fiber tracking, as shown inFig. 4, which depicts the muscles of the volar groups(Fig. 4A,B) and dorsal groups (Fig. 4C,D) next to text-book illustrations (41) for one of the subjects. Onecan appreciate the close resemblance between thetextbook drawings and the DTI-based muscle recon-structions. For fiber tracking only a low number—between one and five—of seeding ROIs were needed tosegment each individual muscle. The fiber tracts ofdifferent muscles stayed well separated, even whenneighboring muscles locally had very similar fiber tra-jectories. To illustrate the rich detail that can beobserved in the reconstructions of individual muscles,two muscles, ie, the FCU (Fig. 4A4) and the SUP(Fig. 4D1), are shown in larger magnification in

DTI of Human Forearm Muscle Architecture 5

Figs. 5 and 6, respectively, together with photographsof the corresponding human cadaver muscles.

Flexor Carpi Ulnaris

The FCU is part of the superficial volar muscle group.The FCU has two sites of origin. The ulnar head origi-nates from an aponeurosis that arises from the proxi-mal two-thirds of the dorsal ulna, ie, the medial ulnarsurface and olecranon. The humeral head originatesfrom the common flexor aponeurosis that arises fromthe medial humeral epicondyle. The FCU ends in a dis-tal tendon that starts halfway along the forearm and

predominantly lies at the anterior side of the muscle. Inboth the fiber tracking reconstruction as well as thephotographs these distinct features of the FCU are wellrecognized. Figure 5A–C shows the FCU in the anatom-ical context of the forearm in several orientations. Fig-ure 5D displays three T1-weighted images at the posi-tions indicated by the red arrows in Fig. 5A. Thedashed white lines contour the FCU. Arrows in Fig. 5Dindicate the viewing directions for the FCU in Fig.5A,E,F,G. At the superficial surface of the FCU, shownin Fig. 5E, fibers curve around the body of the muscleand reveal a small part of the superficial surface of thedistal tendon, which is denoted by the dashed black

Figure 3. Fiber tractography of the whole human forearm. Muscle fibers are shown on top of orthogonal cross-sections and sur-face renderings of the T1-weighted images. A,C: Tractography before displacement and diffusion-tensor shear corrections. B,D:Tractography after corrections. Regions where T1-weighted data and fiber tracts were severely misregistered are indicated withred arrows. After postprocessing the misregistration virtually disappeared and denser fiber tracking continued toward the proxi-mal end of the forearm as indicated by the red asterisk. E,F: Segmented Flexor digitorum profundus of all five subjects before (E)and after (F) postprocessing. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

6 Froeling et al.

contour on the right. In Fig. 5F,G, the deep surface ofthe distal tendon is fully visible, as indicated by thedashed black contours on the right. The pennate struc-ture of the muscle, where muscle fibers that originatefrom the proximal tendon plate attach to both sides ofthe distal tendon, is clearly visible. A movie of the fibertractography reconstruction of the FCU together withthe cadaver dissection photographs is available fromSupporting Digital Content 2.

Supinator

Figure 6A–C shows the SUP in the anatomical con-text of the forearm. In Fig. 6D three T1-weighted sli-

ces are shown at the positions indicated by the redarrows in Fig. 6A. The dashed white line contoursthe SUP. The SUP consists of two planes of musclefibers and is curved around the upper third of theradius. The muscle originates from tendinous tissue,which attaches to the lateral epicondyle of the hu-merus and the upper one-third of the ulna, ie, theposterior ulnar margin and the base of the olecra-non. The muscle is curved around the upper third ofthe radius and attaches to it by tendinous tissuebetween the radial tuberosity and the insertion ofthe PT. The white arrows in Fig. 6D point in theviewing directions of the fiber tracking reconstruc-tions and cadaver dissections of the SUP in Fig.

Figure 4. Fiber tractography of all human forearm muscles next to textbook anatomical illustrations (41). A: Superficial vo-lar group. B: Deep volar group. C: Superficial dorsal group. D: Deep dorsal group. The black dashed lines indicated the bor-der between two neighboring fiber clusters based on their seeding ROIs.

DTI of Human Forearm Muscle Architecture 7

6A,E–H. In Fig. 6E,F the curved, sheet-like musclefiber organization of the SUP is visible. The originand insertion of the muscle, indicated by the dashedblack lines in the images of both the DTI fiber recon-struction and the cadaver dissection, are in goodcorrespondence. Due to partial volume effects asmall fraction of superficial fibers seems to follow theEDC, which neighbors the SUP, as indicated withthe asterisk in Fig. 6G. A movie of the fiber tractog-raphy reconstruction of the SUP together with thecadaver dissection photographs is available fromSupporting Digital Content 3.

Diffusion Parameters

The mean values of the five diffusion tensor param-eters before and after postprocessing are shown inTable 1. No large differences of the mean values ofthe parameters were observed between the different

muscles and values are in agreement with valuespreviously reported (29). After postprocessing, themean values of the first and second eigenvalue, theMD and FA on average drop by 2%, 1%, 1%, and9%, respectively. The third eigenvalue increases 5%.The standard deviations values of the first and sec-ond eigenvalue, the MD and FA decrease 32%,34%, 26%, and 15%, respectively. The standarddeviation of the third eigenvalue slightly increaseswith 5%.

DISCUSSION

In this study we present detailed DTI muscle fiberreconstructions of the human forearm. All musclescould be segmented and fiber tractography showedconvincing visual correspondence with dissections ofa human cadaver forearm. Tractography of human

Figure 5. Fiber tractography ofthe FCU. A–C: FCU in the anatom-ical context of the forearm. D:Three T1-weighted images at thepositions of the red arrows in A.The dashed white line contoursthe FCU. E–G: FCU tractographyand cadaver dissections from sev-eral viewing points indicated bythe white arrows in D. The dashedblack lines on the right and on theleft indicate the positions of thedistal tendon and proximal apo-neurosis, respectively. [Color fig-ure can be viewed in the onlineissue, which is available atwileyonlinelibrary.com.]

8 Froeling et al.

forearm muscles has been shown before (29,42). How-ever, to the best of our knowledge these are the firstdetailed DTI tractography reconstructions showing allof the human forearm muscles using a single acquisi-tion. The time-efficient MRI examination makes thisprotocol clinically usable.

Most of the DTI muscle tractography studies havefocused on reconstructions of the larger muscles ofthe lower extremities, such as the calf (10,21,43) and

thigh (11). DTI of the human forearm is particularlychallenging. In order to reveal the 19 smaller andlarger muscles of the forearm a high spatial resolutionis required. Even minor magnetic field inhomogeneity-induced distortions may be of similar dimensions asthe thickness of the smaller muscles. Part of the vol-ume of the human forearm is occupied by intermus-cular and subcutaneous fat, bone, blood vessels,nerves, and tendinous tissue. The close proximity of

Figure 6. Fiber tractography of the SUP. A–C: SUP in the anatomical context of the forearm. D: Three T1-weighted images atthe positions of the red arrows in A. The dashed white line contours the SUP. E–H: SUP tractography and cadaver dissectionsfrom several viewing points indicated by the white arrows in D. The dashed black lines indicate the origin and insertion. Theasterisk indicates a small fraction of superficial fibers that seems to follow the EDC, which neighbors the SUP. [Color figurecan be viewed in the online issue, which is available at wileyonlinelibrary.com.]

DTI of Human Forearm Muscle Architecture 9

these tissues to muscle results in partial volumeeffects with negative effects on the quality of the trac-tography. Moreover, muscle has a low T2 at 3 T,resulting in low SNR of the diffusion-weighted SE-EPIacquisitions and introducing inaccuracies in the diffu-sion tensor estimation (28,29,44). Resolution andSNR may be improved at the expense of a longer scantime. However, the goal of the present study was todevelop a patient-friendly clinically applicable proto-col, which required a short total examination time. Wetherefore decided to cope with the above difficultiesand, where possible, correct for them using appropri-ate postprocessing of the image and tensor data.

Spatial deformations of the diffusion tensor as aconsequence of field inhomogeneities were adequatelycorrected using a DB0-field map, which was obtainedfrom a short gradient-echo acquisition. Nevertheless,this method does not fully recover the original signalintensities from the squeezed or stretched voxels,which could lead to incorrect tensor estimations(32,45,46). An improvement to the present field-inho-mogeneity correction protocol might be found in apixel-wise measurement of the point-spread-function,which could be used to recover original signal inten-sities even in heavily deformed regions (47). However,the latter approach increases the total examinationtime, as it requires a longer B0-mapping and point-spread-function acquisition.

The mean values and standard deviations of the dif-fusion parameters changed after postprocessing.These changes were in line with the expectationsbased on simulations done by Damon (44). With thecurrent acquisition parameters and SNR one wouldexpect a slight overestimation of the first and secondeigenvalue and the MD. After noise suppression,registration, and correction of the data these meanvalues indeed dropped slightly. Also, the third eigen-value would be underestimated, which is also clearlyindicated by the 5% increase of the parameter afterpostprocessing. As a result of the slight overestima-tion of l1 and l2 and underestimation of l3 the FAwas considerably overestimated. After postprocessingthis parameter decreased almost 10%. Due to the var-

iance in SNR between the different subjects the over-and underestimation was different for each subject,thus increasing the intergroup variability. The stand-ard deviation of the parameters decreased with 15%to 30% by correcting for all the artifacts, except forl3, where the standard deviation slightly increased.The reduced standard deviation of the parameters willincrease the repeatability of the measurement andthus allow for smaller group sizes or detection ofsmaller differences in parameters between groups.

Validation of the forearmmuscle fiber tracking recon-structions consisted of comparisons to human cadaverdissections. A shortcoming of this approach is that DTIacquisitions were performed on a healthy young sub-ject, while cadaver dissections were of an elderly sub-ject. Also, the validation is subjective and not quantifi-able. Nevertheless, the correspondence betweencadaver muscle specimens and DTI muscle tractogra-phy was very convincing. Alternatively, one could fur-ther compare the muscle tractography quantitativelyby performing DTI measurements on cadaver speci-mens, an approach that previously has been used to es-tablish alignment of the principal eigenvector with localmuscle fiber direction in the ex vivo rat tibialis anteriormuscle (3). However, cadaver tissue does not have thesame diffusion properties as living tissue, which com-plicates a straightforward extrapolation of the fibertractography accuracy toward the in vivo measure-ments. Another possibility for validation could be ultra-sound. This technique is widely used to quantitativelyestimate the muscle architectural characteristics in2D. However, determination of 3D fascicle orientationhas only recently been proven possible with limited re-solution (48). Currently, there exists no other noninva-sive technique to compare with the muscle tractogra-phy, emphasizing the uniqueness of in vivo muscle DTI.

The postprocessing tool, developed in this work,was fully automated and required no user input. Incontrast, segmentation of individual muscles wasbased on manually drawn ROIs. Even though we onlyneeded a low number of ROIs per muscle, this was atime-consuming iterative and potentially subjectiveprocess. ROI drawing and segmentation was

Table 1

Mean Value 6 Standard Deviation of Five Diffusion Tensor Parameters of the Six Selected Muscles and the Whole Muscle Volume

Before postprocessing

PT FCU FDP QUA EDC SUP WMV

k1 [10-3 mm2/s] 1.94 6 0.13 1.91 6 0.16 1.94 6 0.13 1.92 6 0.27 1.95 6 0.07 2.14 6 0.24 1.96 6 0.14

k2 [10-3 mm2/s] 1.43 6 0.11 1.40 6 0.13 1.43 6 0.09 1.39 6 0.18 1.39 6 0.09 1.55 6 0.19 1.43 6 0.09

k3 [10-3 mm2/s] 1.11 6 0.10 1.07 6 0.05 1.12 6 0.11 1.02 6 0.10 1.06 6 0.07 1.16 6 0.09 1.08 6 0.06

MD [10-3 mm2/s] 1.50 6 0.11 1.46 6 0.11 1.50 6 0.09 1.46 6 0.20 1.47 6 0.07 1.62 6 0.16 1.49 6 0.10

FA [-] 0.28 6 0.03 0.29 6 0.03 0.28 6 0.05 0.30 6 0.02 0.30 6 0.01 0.30 6 0.03 0.29 6 0.02

After postprocessing

PT FCU FDP QUA EDC SUP WMV

k1 [10-3 mm2/s] 1.88 6 0.06 1.91 6 0.11 1.93 6 0.11 1.89 6 0.19 1.96 6 0.07 1.98 6 0.09 1.92 6 0.06

k2 [10-3 mm2/s] 1.38 6 0.06 1.39 6 0.09 1.41 6 0.08 1.42 6 0.16 1.40 6 0.05 1.47 6 0.07 1.39 6 0.06

k3 [10-3 mm2/s] 1.10 6 0.04 1.14 6 0.09 1.18 6 0.07 1.13 6 0.14 1.15 6 0.09 1.16 6 0.08 1.12 6 0.05

MD [10-3 mm2/s] 1.46 6 0.06 1.47 6 0.10 1.50 6 0.07 1.48 6 0.17 1.50 6 0.06 1.54 6 0.08 1.47 6 0.06

FA [-] 0.27 6 0.02 0.26 6 0.02 0.25 6 0.03 0.26 6 0.01 0.28 6 0.02 0.27 6 0.02 0.27 6 0.01

10 Froeling et al.

supported by anatomical information from the T1-weighted images, which could be used directly for seg-mentation purposes because of the good registrationof DTI and T1-weighted data. Nevertheless, for appli-cation of these techniques to large study groups, fur-ther improvements are needed to automate the seg-mentation process. One could envision combininganatomical information derived from T1-weightedimages, like positions of bones, tendons, and bloodvessels, with DTI muscle tractography to improve theautomatic segmentation algorithms. Another solutioncould be the seed surface technique as proposed byLansdown et al (20). However, due to the complexmuscle anatomy in the human forearm this approachis very difficult to implement and potentially moretime-consuming. For example, the common flexor ten-don is the origin, in part, of all the five superficial vo-lar muscles and the Flexor digitorum superficialismuscle arises from three different locations and endsin four tendons. The current protocol can be adaptedfor other regions, which makes the method a usefultool in clinical education and research of anatomyand movement science. Finally, subject-specific mus-cle architectural parameters, such as fiber lengths,physiological cross-sectional area, and pennationangles, could be extracted.

In conclusion, we have presented a time-efficientpatient-friendly acquisition protocol and a postpro-cessing tool for DTI tractography and segmentation ofhuman forearm muscles. The protocol can be appliedto study human muscle architecture in vivo, whichmay serve as input for models of musculoskeletalmechanics, as well as aid in diagnosis and assess-ment of muscle injury and pathology.

ACKNOWLEDGMENT

We thank Leon Huiberts and Johan Hekking for assis-tance with the dissection and photography of thecadaver forearm.

REFERENCES

1. Burkholder TJ, Fingado B, Baron S, Lieber RL. Relationshipbetween muscle fiber types and sizes and muscle architecturalproperties in the mouse hindlimb. J Morphol 1994;221:177–190.

2. Lieber RL, Friden J. Functional and clinical significance of skele-tal muscle architecture. Muscle Nerve 2000;23:1647–1666.

3. Van Donkelaar CC, Kretzers LJ, Bovendeerd PH, et al. Diffusiontensor imaging in biomechanical studies of skeletal muscle func-tion. J Anat 1999;194(Pt 1):79–88.

4. Mori S, Crain BJ, Chacko VP, van Zijl PC. Three-dimensionaltracking of axonal projections in the brain by magnetic resonanceimaging. Ann Neurol 1999;45:265–269.

5. Mori S, van Zijl PC. Fiber tracking: principles and strategies — atechnical review. NMR Biomed 2002;15:468–480.

6. Damon BM, Ding Z, Anderson AW, Freyer AS, Gore JC. Validationof diffusion tensor MRI-based muscle fiber tracking. Magn ResonMed 2002;48:97–104.

7. Fornage BD. The case for ultrasound of muscles and tendons.Semin Musculoskelet Radiol 2000;4:375–391.

8. Kawakami Y, Abe T, Fukunaga T. Muscle-fiber pennation anglesare greater in hypertrophied than in normal muscles. J ApplPhysiol 1993;74:2740–2744.

9. Rutherford OM, Jones DA. Measurement of fibre pennation usingultrasound in the human quadriceps in vivo. Eur J Appl PhysiolOccup Physiol 1992;65:433–437.

10. Kan JH, Heemskerk AM, Ding Z, et al. DTI-based muscle fibertracking of the quadriceps mechanism in lateral patellar disloca-tion. J Magn Reson Imaging 2009;29:663–670.

11. Budzik JF, Le Thuc V, Demondion X, Morel M, Chechin D, CottenA. In vivo MR tractography of thigh muscles using diffusion imag-ing: initial results. Eur Radiol 2007;17:3079–3085.

12. Heemskerk AM, Damon BM. Diffusion tensor MRI assessment ofskeletal muscle architecture. Curr Med Imaging Rev 2007;3:152–160.

13. Kreulen M, Smeulders MJ. Assessment of flexor carpi ulnarisfunction for tendon transfer surgery. J Biomech 2008;41:2130–2135.

14. Holl N, Echaniz-Laguna A, Bierry G, et al. Diffusion-weightedMRI of denervated muscle: a clinical and experimental study.Skeletal Radiol 2008;37:1111–1117.

15. Saotome T, Sekino M, Eto F, Ueno S. Evaluation of diffusionalanisotropy and microscopic structure in skeletal muscles usingmagnetic resonance. Magn Reson Imaging 2006;24:19–25.

16. Yanagisawa O, Kurihara T, Kobayashi N, Fukubayashi T. Strenu-ous resistance exercise effects on magnetic resonance diffusionparameters and muscle-tendon function in human skeletal mus-cle. J Magn Reson Imaging 2011;34:887–894.

17. Zaraiskaya T, Kumbhare D, Noseworthy MD. Diffusion tensorimaging in evaluation of human skeletal muscle injury. J MagnReson Imaging 2006;24:402–408.

18. Blemker SS, Pinsky PM, Delp SL. A 3D model of muscle revealsthe causes of nonuniform strains in the biceps brachii. J Bio-mech 2005;38:657–665.

19. Hatakenaka M, Matsuo Y, Setoguchi T, et al. Alteration of protondiffusivity associated with passive muscle extension and contrac-tion. J Magn Reson Imaging 2008;27:932–937.

20. Lansdown DA, Ding Z, Wadington M, Hornberger JL, Damon BM.Quantitative diffusion tensor MRI-based fiber tracking of humanskeletal muscle. J Appl Physiol 2007;103:673–681.

21. Deux JF, Malzy P, Paragios N, et al. Assessment of calf musclecontraction by diffusion tensor imaging. Eur Radiol 2008;18:2303–2310.

22. Heemskerk AM, Sinha TK, Wilson KJ, Damon BM. Change inwater diffusion properties with altered muscle architecture. In:Proc 16th Annual Meeting ISMRM, Toronto; 2008. p 1787.

23. Heemskerk AM, Sinha TK, Wilson KJ, Ding Z, Damon BM. Quan-titative assessment of DTI-based muscle fiber tracking and opti-mal tracking parameters. Magn Reson Med 2009;61:467–472.

24. Sinha S, Sinha U. Reproducibility analysis of diffusion tensorindices and fiber architecture of human calf muscles in vivo at1.5 Tesla in neutral and plantarflexed ankle positions at rest.J Magn Reson Imaging 2011;34:107–119.

25. Sinha U, Sinha S, Hodgson JA, Edgerton RV. Human soleus mus-cle architecture at different ankle joint angles from magnetic reso-nance diffusion tensor imaging. J Appl Physiol 2011;110:807–819.

26. Stanisz GJ, Odrobina EE, Pun J, et al. T1, T2 relaxation andmagnetization transfer in tissue at 3T. Magn Reson Med 2005;54:507–512.

27. Gold GE, Han E, Stainsby J, Wright G, Brittain J, Beaulieu C.Musculoskeletal MRI at 3.0 T: relaxation times and image con-trast. Am J Roentgenol 2004;183:343–351.

28. Heemskerk AM, Sinha TK, Wilson KJ, Ding Z, Damon BM.Repeatability of DTI-based skeletal muscle fiber tracking. NMRBiomed 2010;23:294–303.

29. Froeling M, Oudeman J, van den Berg S, et al. Reproducibility ofdiffusion tensor imaging in human forearm muscles at 3.0 T in aclinical setting. Magn Reson Med 2010;64:1182–1190.

30. Aja-Fernandez S, Niethammer M, Kubicki M, Shenton ME, WestinCF. Restoration of DWI data using a Rician LMMSE estimator.IEEE Trans Med Imaging 2008;27:1389–1403.

31. Herraez MA, Burton DR, Lalor MJ, Gdeisat MA. Fast two-dimen-sional phase-unwrapping algorithm based on sorting by reliabilityfollowing a noncontinuous path. Appl Opt 2002;41:7437–7444.

32. Chen NK, Wyrwicz AM. Correction for EPI distortions usingmulti-echo gradient-echo imaging. Magn Reson Med 1999;41:1206–1213.

33. Koch KM, Rothman DL, de Graaf RA. Optimization of static mag-netic field homogeneity in the human and animal brain in vivo.Prog Nucl Magn Reson Spectrosc 2009;54:69–96.

DTI of Human Forearm Muscle Architecture 11

34. Alexander DC, Pierpaoli C, Basser PJ, Gee JC. Spatial transfor-mations of diffusion tensor magnetic resonance images. IEEETrans Med Imaging 2001;20:1131–1139.

35. Henkelman RM. Measurement of signal intensities in the pres-ence of noise in MR images. Med Phys 1985;12:232–233.

36. Kaufman L, Kramer DM, Crooks LE, Ortendahl DA. Measuringsignal-to-noise ratios in MR imaging. Radiology 1989;173:265–267.

37. Haber E, Modersitzki J. Intensity gradient based registration andfusion of multi-modal images. Methods Inf Med 2007;46:292–299.

38. Haber E, Modersitzki J. Intensity gradient based registration andfusion of multi-modal images. Med Image Comput Comput AssistInterv 2006;9(Pt 2):726–733.

39. Leemans A, Jones DK. The B-matrix must be rotated when cor-recting for subject motion in DTI data. Magn Reson Med 2009;61:1336–1349.

40. Vilanova A, Berenschot G, van de Pul C. DTI visualization withstreamsurfaces and evenly-spaced volume seeding. Proc VisSym.Konstanz, Germany. 2004. pp.173–182.

41. Gray H. Gray’s anatomy of the human body. Philadelphia: Lea &Febiger; 1918.

42. Levin DI, Gilles B, Madler B, Pai DK. Extracting skeletal musclefiber fields from noisy diffusion tensor data. Med Image Anal2011;15:340–353.

43. Sinha S, Sinha U, Edgerton VR. In vivo diffusion tensor imaging ofthe human calf muscle. J Magn Reson Imaging 2006;24:182–190.

44. Damon BM. Effects of image noise in muscle diffusion tensor(DT)-MRI assessed using numerical simulations. Magn ResonMed 2008;60:934–944.

45. Jezzard P, Balaban RS. Correction for geometric distortion inecho planar images from B0 field variations. Magn Reson Med1995;34:65–73.

46. Jezzard P, Barnett AS, Pierpaoli C. Characterization of and cor-rection for eddy current artifacts in echo planar diffusion imag-ing. Magn Reson Med 1998;39:801–812.

47. Zeng H, Constable RT. Image distortion correction in EPI: com-parison of field mapping with point spread function mapping.Magn Reson Med 2002;48:137–146.

48. Rana M, Wakeling JM. In-vivo determination of 3D muscle archi-tecture of human muscle using free hand ultrasound. J Biomech2011;44:2129–2135.

12 Froeling et al.