Original Research

Fast 3D 1H MRSI of the Corticospinal Tract inPediatric BrainDong-Hyun Kim, PhD,1* Meng Gu, MS,2 Charles Cunningham, PhD,2 Albert Chen, MS,3

Fiona Baumer, BS,3 Orit A. Glenn, MD,3 Daniel B. Vigneron, PhD,3

Daniel Mark Spielman, PhD,3 and Anthony James Barkovich, MD3

Purpose: To develop a 1H magnetic resonance spectro-scopic imaging (MRSI) sequence that can be used to imageinfants/children at 3T and by combining it with diffusiontensor imaging (DTI) tractography, extract relevant meta-bolic information corresponding to the corticospinal tract(CST).

Materials and Methods: A fast 3D MRSI sequence was de-veloped for pediatric neuroimaging at 3T using spiral k-spacereadout and dual band RF pulses (32 � 32 � 8 cm field of view[FOV], 1 cc iso-resolution, TR/TE � 1500/130, 6:24 minutescan). Using DTI tractography to identify the motor tracts,spectra were extracted from the CSTs and quantified. Initialdata from infants/children with suspected motor delay (n � 5)and age-matched controls (n � 3) were collected and N-acety-laspartate (NAA) ratios were quantified.

Results: The average signal-to-noise ratio of the NAA peakfrom the studies was �22. Metabolite profiles were suc-cessfully acquired from the CST by using DTI tractography.Decreased NAA ratios in those with motor delay comparedto controls of �10% at the CST were observed.

Conclusion: A fast and robust 3D MRSI technique targetedfor pediatric neuroimaging has been developed. By combin-ing with DTI tractography, metabolic information from theCSTs can be retrieved and estimated. By combining DTIand 3D MRSI, spectral information from various tracts canbe obtained and processed.

Key Words: volumetric MRSI; fast MRSI; CST tractographyJ. Magn. Reson. Imaging 2009;29:1–6.© 2008 Wiley-Liss, Inc.

SINGLE VOXEL SPECTROSCOPY has been used to ac-cess various diseases in the pediatric brain. The use ofa localizing technique via PRESS (1) or STEAM (2) en-ables collecting information from a rectangular regionof interest (ROI). For many cases this information suf-fices in determining the desired relevant metabolic in-formation if the ROI can be well defined. However, thereis need to expand the ROI to two dimensions (2D) oreven three dimensions (3D) for certain applications. Awide scope of studies falls into this category where thespatial information can be an important factor. In ad-dition, another realistic reason for multidimensionalapplication can be that the region to acquire the spec-troscopic information from might not be well character-ized from the scout or other anatomical images. Forexample, when the targeted ROI is the corticospinaltract (CST), this region might not be readily visible inthe conventional scout T1- or T2-weighted images whenimaging infants.

Multivoxel spectroscopy, 2D or 3D, can be used toovercome these limitations while increasing the spatialcoverage. Conventional phase-encoded multivoxelspectroscopy is a robust and simple method but it ac-companies an increase in scan time that is proportionalto the number of voxels resolved. For 3D applicationsthis can amount to a significant scan time, limiting itsuse for clinical applications. For pediatric exams thelong scan time can be crucial due to the fact that theyare more vulnerable to move during the longer scans.For this reason there is a strong desire for a fasterscanning method for this population. Sedation is oftenused which decreases susceptibility to motion, but afaster scan would still be desirable to reduce the overallsedation time.

The purpose of this study was to develop a 1H mag-netic resonance spectroscopic imaging (MRSI) se-quence that can be used to image infants/children at3T and by combining it with diffusion tensor imaging(DTI) tractography, extract relevant metabolic informa-tion corresponding to the CST. By linking spectroscopicimaging with DTI to extract voxels corresponding to theCST, comprehensive metabolic assessment of the CSTcould be obtained. To demonstrate the potential useful-ness of this approach, data were acquired from infant/children with developmental motor delay and controls

1School of Electrical and Electronic Engineering, Yonsei University,Seoul, Korea.2Department of Radiology, Stanford University, Stanford, California, USA.3Department of Radiology, University of California, San Francisco, Cal-ifornia, USA.Contract grant sponsor: National Institutes of Health; Contract grantnumbers: NIH NS40117, RR09784; Contract grant sponsor: NationalCenter for Research Resources, US Public Health Services; Contractgrant number: 5 M01 RR-01271 to the the pediatric clinical researchcenter, University of California, San Francisco.*Address reprint requests to: D.K., School of Electrical and ElectronicEngineering, Yonsei University, Seoul, Korea.E-mail: donghyunkim@yonsei.ac.krReceived August 1, 2007; Accepted March 3, 2008.DOI 10.1002/jmri.21394Published online in Wiley InterScience (www.interscience.wiley.com).

JOURNAL OF MAGNETIC RESONANCE IMAGING 29:1–6 (2009)

© 2008 Wiley-Liss, Inc. 1

and N-acetylaspartate (NAA) level ratios from the CSTswere compared. Previous studies have indicated therapid metabolic alterations during development and itsdifferences from abnormal developmental conditions(3–5). Polynomial, logarithmic, or exponential fits havebeen used to describe the changes for normal develop-ment. Specifically, these studies have noted the in-crease in NAA levels as normal development occurs.

The main objective of this study was not to derive aclinical conclusion but rather to show the potentialclinical utility of this approach. In demonstrating thisfeature we follow an approach similar to that of a clin-ical study. It should be noted that the limited number ofsubjects for this study does not validate any clinicalsuggestions. In the next section we illustrate the imple-mentation of a fast 3D MRSI pulse sequence for use inpediatric patients. We apply the pulse sequence to pe-diatric patients with delayed motor development.

MATERIALS AND METHODS

Fast 3D 1H MRSI Pulse Sequence

A fast 3D 1H MRSI sequence using PRESS excitationwas developed for pediatric brain imaging at 3T (Fig. 1).Two identical dual band spectral spatial spin echopulses, designed to excite 1% of the water signal andsuppress lipids while passing the NAA, creatine (Cr),and choline (Cho) resonances was optimized for 3Tbrain imaging and used instead of the conventionalPRESS 180° pulses (6).

In-plane (kx and ky) spatial information was encodedusing spiral k-space readout trajectories while chemi-cal shift information was encoded by repeating the spi-ral trajectories. The spiral trajectory was designed tocover a 32 � 32 matrix over a 32 � 32 cm field of view(FOV) (kxmax � kymax � 0.5 cm�1). The relatively largeFOV was chosen so that ringing artifacts, mostly arisingfrom the residual lipid components, do not alias intothe ROI. The resulting spiral trajectory had a length of1.024 msec, thereby enabling a spectral bandwidth of976 Hz using a 4 �s sampling receiver. During a singlereadout, 512 spirals were continuously played out,which required a readout length of 524 msec. To satisfythe required imaging parameter of a 32 � 32 spatial

coverage, 16 spatial interleaving of the spirals wereused. A detailed depiction of spiral-based k-space read-outs for spectroscopic imaging has been previously de-scribed (7). After data acquisition, data were recon-structed using gridding. A 4 Hz Lorentzian apodizationfunction was used in the spectral domain to increasethe spectral signal-to-noise ratio (SNR). No windowingwas performed in the spatial frequency domain. Sinceno zero-filling was used the nominal spatial resolution,as measured by the full-width half-maximum value ofthe point spread function, should be 1 cc with iso-resolution. To verify this, simulation of the point spreadfunction was performed by assuming an impulsive ob-ject in the image domain and reconstructing with themethod described above to validate this.

Eight phase encodes (kz) were also used to acquire 3Dinformation with nominal resolution of 1 cm. Spatialsaturation pulses were used to exclude excitation of thesubcutaneous lipid regions. The PRESS selection boxwas chosen as large as possible within brain tissue.This was to reduce lipid signals from being excited whilecollecting necessary metabolic information. Althoughspectral spatial refocusing pulses and spatial satura-tion pulses were both used, additional suppression ofthe lipid signal was typically required. The PRESS se-lection box size was normally 10 � 10 � 6 cm locatedwithin the pediatric brain. Data were collected using aneight-channel receiver coil providing increased SNR.The imaging parameters were as follows: 32 � 32 � 8matrix over a 32 � 32 � 8 cm FOV, 1 cc iso-resolution,TR/TE � 1500/130, 980 Hz spectral bandwidth, 2 sig-nal averages, 6:24 minute total scan time. Data werecollected on a GE 3T (General Electric Healthcare Tech-nologies, Milwaukee, WI) scanner located at our insti-tution.

Diffusion and Standard Imaging Sequences

The diffusion tensor imaging data is acquired in 3.1minutes using a multirepetition, single-shot echo pla-nar sequence with 15 gradient directions, b � 0 and1000 s/mm2, TR � 11.5 seconds, TE � 61.8 msec, 1repetition, FOV � 28 � 14 cm, matrix � 256 � 128,slice thickness � 2.2 mm with no gap, 167 kHz readoutbandwidth, and no ramp sampling. A sense factor oftwo was used to reduce the geometric distortions. Theresulting in-plane resolution was 1.1 mm. Interleavedaxial slices were acquired for full brain coverage.

Additional standard imaging sequences included: 1)T1-weighted sagittal and axial spin echo images (4 mmthickness) using repetition time (TR) � 500 msec, echotime (TE) � 11 msec, 1 excitation, and 192 � 256acquisition matrix; and 2) axial 3D fast spin echo (FSE)images using TR � 4000 msec, TE � 85 msec, echotrain length (ETL) � 16, bandwidth (BW) � 15, 1.5 mmslice thickness, 6 locations/slab, and 192 � 256 acqui-sition matrix; 3) Coronal volumetric 3D gradient echoimages with radiofrequency spoiling (SPGR) images (1mm thickness) with TR � 36 msec, TR � 9 msec, flipangle � 35°, NEX � 1.

Figure 1. Pulse sequence diagram. A dual band spectral spa-tial spin echo pulse is used. Spiral readout gradients areplayed for fast data acquisition and full coverage. VSS (veryselective saturation) pulses are used to further suppress sig-nals coming from the subcutaneous lipids.

2 Kim et al.

Data Processing and Analysis

Spectroscopic data reconstruction was performed foreach coil by apodizing using a 4-Hz Gaussian filterfollowed by gridding and inverse fast Fourier transform(FFT). Using residual water signal from the dual bandsequence, zero and first-order phasing was applied toevery voxel spectrum. Spectra combination from theindividual coils was accomplished by weighting accord-ing to the phased water signal amplitude to optimize forSNR.

Data quantification was performed following the re-construction. Using in-house software for DTI fibertracking (8), the CSTs were identified. Voxels from thespectroscopic image corresponding to the CSTs wereextracted. Voxel shifting was performed prior to theextraction to center the spectra on the CST wheneverpossible to reduce partial volume errors. Spectral quan-tification of the extracted voxels was performed usingLCModel fitting (9). These quantitative values corre-sponding to the location of the CSTs were summed andaveraged. Cramer–Rao bounds of the LCModel fit be-yond 10% were excluded from the analysis.

For data analysis, NAA ratios (NAA/Cr and NAA/Cho)from the infants with motor delay were compared withthose from the controls. For controls a logarithmicfunction was fit to the spectroscopic data as a functionof age to emulate maturation.

Data Acquisition

Patients and controls were recruited from a group ofpatients already scheduled for MRI at our outpatientfacility. Prior to the scan, all parents of the subjectswere approached and, if agreeable to participation,were consented for the protocol, which was approved byour Institutional Review Board (IRB). The five subjectinfants and children were being studied for motor de-lay, whereas the three controls, of similar ages, hadnormal motor function and were being studied for es-otropia, language delay, or evaluation of a facial hem-angioma. A more complete description of the patientpopulation is listed in Table 1. Spectroscopic data wereacquired using the newly developed 3D MRSI pulsesequence after the routine imaging sequences had beencompleted.

RESULTS

By using DTI tractography to identify the CST we wereable to extract and analyze spectroscopic informationon most of the supratentorial CST from the 3D MRSIdataset and test whether NAA ratios are reduced in theCST compared to normal age-matched controls.

Figure 2 shows representative image slices from amotor delayed child and its corresponding spectra priorto any voxel shifting. Spectral regions surrounding theCSTs are shown. Good spectral quality data can be seenusing the proposed sequence. The average SNR of theNAA metabolite from the gray matter, measured by di-viding the signal amplitude by the standard deviation ofthe noise, was �22 � 2.5. This was slightly higher thanthe SNRs we normally achieved from adult studies,which were �20 � 2.9. The smaller size of the head ofinfants accounts for less noise seen by the receivercoils, thereby increasing the SNR. Estimated standarddeviation from the LCModel fit typically resulted in�5–7% (NAA: 6.04 � 2, Cr: 6.35 � 1.66, Cho: 6.06 �

Table 1Patient Population Information for 1H MRSI Study of Motor Delay Infants

Subject (n�8) Age at Scan Sex Indication for MRI Clinical MRI Result

Control 1 7 months F Facial muscle weakness and exotropia Normal MRIControl 2 10 months F Periorbital hemangioma Normal MRIControl 3 62 months M Language delay (normal motor function) Normal MRIPatient 1 8 months M Mild developmental delay Normal MRIPatient 2 10 months M Significant global delay White matter volume decreasePatient 3 15 months F Global delay, hypotonia, microcephaly Nonspecific white matter

abnormalityPatient 4 21 months M Global delay, spastic diplegia making progress,

right side preferenceNormal MRI

Patient 5 40 months F Global delay White matter volume decrease

Figure 2. Representative slices (b � 0 images from the DTIacquisitions) and spectra from a motor-delayed child. The leftCST is identified via DTI tractography and the regions, markedin white spots, are indicated by the arrows in the image fig-ures. Spectra are voxel shifted and extracted to collect themetabolic information mainly from the CST. The typical spec-tral coverage includes most of CST, normally ranging from theprecentral gyrus to the cerebral peduncle.

3D MRSI of the Corticospinal Tract 3

1.79) and below 15% for the metabolites of interest.Simulation of the point spread function resulted in a 1cc resolution (data not shown).

Figure 3 shows an example of a resulting DTI trac-tography and summed MRSI spectra extracted from theCSTs from an age-matched control (control #2) and amotor-delayed child (patient #2). Metabolite ratiosshow a relatively decreased NAA value for the motor-delayed infant compared to age-matched control. Whilethe tracts identified (shown in yellow) represent theCST, the spectral analysis was performed within thosetracts as indicated by the anatomic image representingthe bottom-most plane and blue patch representing theupper-most plane. These regional constraints were dueto the spectroscopic acquisition limitations rather thanfrom limitations from the DTI analysis. Extending be-

yond this limit superiorly resulted in bad spectral qual-ity due to heavy lipid contamination even with the dualband spectral spatial RF pulses. In the inferior direc-tion, line broadening due to B0 inhomogeneity from theskull base and sinuses limited data quantification.

The NAA/Cho and NAA/Cr ratios obtained from theCSTs of all the study subjects are given in Fig. 4. Thethree control data are fit to a logarithmic curve to as-sume nominal growth curves. It can be seen that all ofthe motor-delayed children have NAA ratios below thenormative data curve, although the amount of falloffvaries. One patient (21 months old) was diagnosed asdiplegic with a right-sided preference, and data pointsare plotted for both the combined CSTs (upper datapoint) and only the most affected side (lower data point).This is an added benefit of using volumetric imaging inthat bilateral comparisons can be readily accom-plished. Although the sample size is too small to assertthis, the overall findings from our pilot study indicatethat NAA ratios are reduced for those with motor delaycompared to controls with an effect size of �10% (9.5 �6.4% for NAA/Cho and 6.2 � 3.5% for NAA/Cr), similarto the findings reported by Filippi et al (10), Kulak et al(11), and Fayed et al (12).

DISCUSSION

In this article we present our study on the developmentof a fast 3D MRSI acquisition technique targeted forpediatric applications. The pulse sequence obtained 1cc spectroscopic data from a 3D volume region within a6.5-minute scan time and provided good SNR (�22) fordata analysis. By using DTI tractography to accuratelyidentify the CST, spectroscopic information corre-sponding to this region were extracted and analyzed. Asan illustrative example, we used this technique to com-pare NAA ratios in infants with and without motor de-

Figure 3. Combined DTI and MRSI pediatric studies of theCSTs. Center: Example DTI tractography identifying the CSTs(the blue cut-plane indicated the top of the PRESS box whilethe anatomic image represents the bottom plane included inthe analysis). Left and right: Summed spectra from the MRSIdataset computed by adding voxels corresponding to the por-tion of the CSTs within the PRESS excitation box (using voxelshifting as necessary to center the spectra near the CST), froma control and a child with motor delay. Metabolite ratios werecalculated using LCModel software.

Figure 4. NAA/Cho (left) and NAA/Cho (right) ratios computed from the CSTs of three controls and five patients with motordelay. A logarithmic function is fit to the data acquired from controls to assume normative maturation curves. Given thisassumption, the overall findings indicate that NAA ratios are reduced for those with motor delay compared with controls with aneffect size of �10%. A diplegic patient (21 months old) showed NAA ratios further decreased in the affected hemisphere comparedto the other hemisphere.

4 Kim et al.

lay. Overall, the study including a T1-weighted imaging,DTI, and MRSI sequence can be performed within a12-minute total scan time.

The use of a rapid spiral-based readout acquisitionusing PRESS modules allows volumetric coveragewithin a clinically acceptable scan time. The eight-channel receiver coils at 3T provide additional SNR in-creases. A unique feature of combining DTI and 3DMRSI is the ability to segment and extract specific path-ways within the brain. Here we used DTI to acquirespectra along the CST through the use of a fiber track-ing algorithm. Other pathways can likely be extractedfor different studies of the brain using this or similarprotocols. These characteristics are the strengths ofthis study.

One of the major limitations of this study is that evenat a spatial resolution of 1 cc, which is relatively goodfor in vivo MRS, there are considerable partial volumeeffects with respect to the CSTs. Hence, metabolic in-formation was obtained from both the tracts of interestand the immediately surrounding tissue. This is a fun-damental limitation and, to the extent that the addi-tional tissue is normal, we lose statistical power indetecting changes in NAA. In relation to our study pop-ulation, the partial volume effects are confounded bythe fact that several of our patients were globally de-layed and, therefore, had other concurrent neurologicaldeficits in addition to motor delay. By using voxel shift-ing of our spectroscopic data to center the spectra voxelwith the CST voxel, this effect was reduced by a certainamount although the finite resolution of our acquisitionstill precludes a perfect CST spectra (13). A relatedconcern comes from the spatial distortions inevitablewhen using EPI readouts for diffusion imaging. Thisfeature, noticeable in the EPI images of Fig. 2, conse-quently is another source of inaccurate metabolic as-sessment of the CST.

The use of an interior PRESS box, in combinationwith outer-volume spatial selection saturation pulses,provides excellent reduction of subcutaneous lipid sig-nals and high-quality spectroscopic data. However, thisapproach has the distinct disadvantage of prohibitingthe acquisition of metabolic information from the entireCST as well as most of the cerebral cortex, as seen inFig. 3. Further improvements to implement a truly volu-metric 1H MRSI sequence capable of interrogating theentire brain would be attractive (14).

Although the main focus of this study was not in itsclinical use for assessing developmental delay, a noteon this subject is worthwhile. Developmental delay oc-curs in �5%–10% of the childhood population (15,16),with motor dysfunction occurring in 10%–20% of thesecases (17,18). There is growing evidence that early iden-tification of children with developmental delay is criticalto treatment of, or intervention for, a disability andlessening its impact on the functioning of the child andfamily (19). MRI plays an important role in the compre-hensive evaluation of children with many types of de-velopmental delay. With respect to motor delay, be-cause the CST play a critical role in normal motorfunction, it is reasonable to assume that at least asubset of motor dysfunction is associated with abnor-malities in the CST. The ability to accurately assess in

vivo the integrity of the CST has been limited with DTIperhaps playing the largest role to date (20,21).

There are several, albeit limited, reports in the liter-ature concerning 1H MRS findings in developmentaldelay. While Martin et al (22) did not find any significantchanges in brain metabolite ratios of children withglobal developmental delay, several other investigatorsreported detected abnormalities using 1H-MRS. Filippiet al (10), in a cohort of children over the age of 2 withmild developmental delay and normal MRI scans, re-ported significantly decreased NAA/Cr ratios and ele-vated Cho/Cr ratios in frontal and parieto-occipitalsubcortical white matter. In a two-slice 1H-MRSI (MRSimaging) study comparing autism, developmental de-lay, and normal control by Friedman et al (23), an 8%reduction of global NAA was found between the devel-opmentally delayed and control children. In a morerecent article, Kulak et al (11) reported NAA/Cr ratios inthe basal ganglia are negatively correlated with learningdisabilities in patients with spastic cerebral palsy. Astudy more closely related to our targeted populationwas published by Fayed et al (12). In that study of 12children with isolated developmental delay (6 had iso-lated motor delay and 2 had motor delay in combinationwith language expressive delay) and 11 controls, 7%–15% decreases in NAA ratios were reported from singlevoxels located in the left centrum semiovale. In ourstudy we found that, compared to age-matched con-trols, the NAA ratios were reduced in motor-delayedinfants with an effect size of �10%. Additional datafrom both controls and motor delayed infants need to becollected to further confirm this difference.

In conclusion, we have developed a fast and robust3D MRSI technique targeted for pediatric neuroimag-ing. In 6:24 minutes, 3D volumetric data with 1 ccresolution can be readily acquired. Necessary spectro-scopic information was extracted by combining withDTI tractography. The ability to analyze spectra fromthe CST pathway (and other pathways) is unique to 3DMRSI.

REFERENCES1. Bottomley PA. Spatial localization in NMR spectroscopy in vivo. Ann

N Y Acad Sci 1987;508:333–348.2. Frahm J, Bruhn H, Gyngell ML, Merboldt KD, Hanicke W, Sauter R.

Localized high-resolution proton NMR spectroscopy using stimu-lated echoes: initial applications to human brain in vivo. MagnReson Med 1989;9:79–93.

3. Kreis R, Ernst T, Ross BD. Development of the human brain: in vivoquantification of metabolite and water content with proton mag-netic resonance spectroscopy. Magn Reson Med 1993;30:424–437.

4. Kimura H, Fujii Y, Itoh S, et al. Metabolic alterations in the neonateand infant brain during development: evaluation with proton MRspectroscopy. Radiology 1995;194:483–489.

5. Horska A, Kaufmann WE, Brant LJ, Naidu S, Harris JC, Barker PB.In vivo quantitative proton MRSI study of brain development fromchildhood to adolescence. J Magn Reson Imaging 2002;15:137–143.

6. Cunningham CH, Vigneron DB, Marjanska M, et al. Sequence de-sign for magnetic resonance spectroscopic imaging of prostate can-cer at 3 T. Magn Reson Med 2005;53:1033–1039.

7. Adalsteinsson E, Irarrazabal P, Topp S, Meyer C, Macovski A, Spiel-man DM. Volumetric spectroscopic imaging with spiral-based k-space trajectories. Magn Reson Med 1998;39:889–898.

8. Berman JI, Mukherjee P, Partridge SC, et al. Quantitative diffusiontensor MRI fiber tractography of sensorimotor white matter devel-opment in premature infants. Neuroimage 2005;27:862–871.

3D MRSI of the Corticospinal Tract 5

9. Provencher SW. Automatic quantitation of localized in vivo 1Hspectra with LCModel. NMR Biomed 2001;14:260–264.

10. Filippi CG, Ulug AM, Deck MD, Zimmerman RD, Heier LA. Devel-opmental delay in children: assessment with proton MR spectros-copy. AJNR Am J Neuroradiol 2002;23:882–888.

11. Kulak W, Sobaniec W, Smigielska-Kuzia J, Kubas B, Walecki J.Metabolite profile in the basal ganglia of children with cerebralpalsy: a proton magnetic resonance spectroscopy study. Dev MedChild Neurol 2006;48:285–289.

12. Fayed N, Morales H, Modrego PJ, Munoz-Mingarro J. White matterproton MR spectroscopy in children with isolated developmentaldelay: does it mean delayed myelination? Acad Radiol 2006;13:229–235.

13. Haacke E, Brown R, Thompson M. Magnetic resonance imaging.New York: Wiley-Liss, 1999.

14. Gu M, Spielman D. Robust lipid suppression using multiple fre-quency selective pulses for brain spectroscopic imaging at 3T. In:14th Annual Meeting ISMRM, Seattle; 2006:3213.

15. Drillien CM, Pickering RM, Drummond MB. Predictive value ofscreening for different areas of development. Dev Med Child Neurol1988;30:294–305.

16. Petersen MC, Kube DA, Palmer FB. Classification of developmentaldelays. Semin Pediatr Neurol 1998;5:2–14.

17. Shevell M, Ashwal S, Donley D, et al. Practice parameter: evalua-tion of the child with global developmental delay: report of theQuality Standards Subcommittee of the American Academy of Neu-rology and the Practice Committee of the Child Neurology Society.Neurology 2003;60:367–380.

18. Edwards SL, Sarwark JF. Infant and child motor development. ClinOrthop Relat Res 2005:33–39.

19. Developmental surveillance and screening of infants and youngchildren. Pediatrics 2001;108:192–196.

20. Filippi CG, Lin DD, Tsiouris AJ, et al. Diffusion-tensor MR imagingin children with developmental delay: preliminary findings. Radi-ology 2003;229:44–50.

21. Glenn OA, Henry RG, Berman JI, et al. DTI-based three-dimen-sional tractography detects differences in the pyramidal tracts ofinfants and children with congenital hemiparesis. J Magn ResonImaging 2003;18:641–648.

22. Martin E, Keller M, Ritter S, Largo RH, Thiel T, Loenneker T.Contribution of proton magnetic resonance spectroscopy to theevaluation of children with unexplained developmental delay. Pe-diatr Res 2005;58:754–760.

23. Friedman SD, Shaw DW, Artru AA, et al. Regional brain chemicalalterations in young children with autism spectrum disorder. Neu-rology 2003;60:100–107.

6 Kim et al.