Diffusion Tensor Imaging Assessment of Brain White Matter Maturation During the First Postnatal Year

Transcription

1 DTI of White Matter in infants Neuroradiology Original Research James M. Provenzale 1 Luxia Liang 1,2 David DeLong 1 Leonard E. White 3 Provenzale JM, Liang L, DeLong D, White LE Keywords: anisotropy, diffusion tensor imaging, diffusion-weighted imaging, infants, maturation DOI: /AJR Received March 8, 2006; accepted after revision March 28, Department of Radiology, Duke University Medical Center, Box 3808, Durham, NC Address correspondence to J. M. Provenzale (prove001@mc.duke.edu). 2 Beaconbioscience, Doylestown, PA. 3 Department of Community and Family Medicine, Duke University Medical Center, Durham, NC. AJR 2007; 189: X/07/ American Roentgen Ray Society Diffusion Tensor Imaging Assessment of Brain White Matter Maturation During the First Postnatal Year OBJECTIVE. The purpose of this study was to use diffusion-weighted and diffusion tensor imaging to investigate the status of cerebral white matter (WM) at term gestation and the rate of WM maturation throughout the first year of life in healthy infants. MATERIALS AND METHODS. Fifty-three children (35 boys) ranging in age from 1.5 weeks premature to 51.5 weeks (mean age, 22.9 weeks) underwent conventional MRI, diffusion imaging in three directions (b = 1,000 s/mm 2 ), and diffusion tensor imaging with gradient encoding in six directions, all on a 1.5-T MRI system. Apparent diffusion coefficient (ADC) and fractional anisotropy (FA) were measured in three deep WM structures (posterior limb of internal capsule, genu, and splenium of corpus callosum) and two peripheral WM regions (associational WM underlying prefrontal and posterior parietal cortex) with a standard region of interest (44 ± 4 cm 2 ). ADC and FA were expressed as a percentage of corresponding values measured in a group of healthy young adults. Mean ADC and FA values for deep and peripheral WM were plotted against gestational age normalized to term. The data were fit best with a broken-line linear regression model with a breakpoint at 100 days. ADC and FA values at term were estimated according to the intercept of the initial linear period (before day 100) with day 0. The slope of the linear fits was used to determine the rate of WM maturation in both the early and the late (after day 100) periods. Multivariate analysis of variance tests were used to compare deep and peripheral WM structures at term and at representative early and late ages (days 30 and 200) and to compare rates of ADC and FA maturation in early and late periods within the first year. RESULTS. At term, peripheral WM was less mature than deep WM according to results of extrapolation of ADC and FA values in the first 100 days of life to day 0 (p < 0.01). Mean ADC and FA value (percentage of mean adult value) for peripheral WM were mm 2 /s (163%) and 0.16 (32%), respectively, and mm 2 /s (143%) and 0.36 (54%), respectively, for deep WM. On day 30 and day 200, estimated mean ADC and FA continued to show greater diffusion (higher ADC) and less anisotropy (lower FA value) in peripheral WM (p < 0.01). During the first year of postnatal life, both ADC and FA matured at higher rates before postnatal day 100 compared with a later time. Differences were observed in rates of maturation in the first 100 days when rates of decrease in ADC and increase in FA were compared between peripheral WM and deep WM; however, the maturational trends differed whether ADC or FA was examined. The early rate of ADC decrease (maturation) was twice as great for peripheral WM than for deep WM (p < 0.01) unexpectedly, but the opposite pattern was observed for FA. The early rate of FA increase (maturation) was approximately one half as great for peripheral WM as for deep WM (p = 0.01). Throughout the rest of the first year, no differences were observed in the rates of change in either index between peripheral WM and deep WM. CONCLUSION. At term, both ADC and FA differ significantly in peripheral WM and deep WM, deep WM structures being more mature. Both deep WM and peripheral WM mature more rapidly during approximately the first 3 months in comparison with the rest of the first year. Unexpected differences in early (first 100 days) rates of maturation assessed with diffusion-weighted (ADC) and diffusion tensor (FA) imaging suggest that these two techniques may be sensitive to different aspects of WM maturation in the early perinatal period. D iffusion imaging and diffusion tensor imaging (DTI) are MRI techniques used to assess disorders of white matter (WM) in pediatric and adult patients [1 3]. Results of a number of investigations have shown that the apparent diffusion coefficient (ADC), which can be obtained with three-direction diffu- 476 AJR:189, August 2007

2 DTI of White Matter in infants sion-weighted imaging (DWI), and fractional anisotropy (FA), which can be measured with DTI, are sensitive indicators of microstructural changes in the WM of children [1 3]. Numerous studies have shown that cerebral myelination in prenatal and postnatal life occurs in an ordered sequence [4 6]. Studies also have shown that generally ADC decreases and FA increases during childhood and that the changes are thought to parallel myelination [7]. Developmental decreases in ADC and increases in FA, however, may reflect factors other than myelination, such as maturational changes in the cytoskeletal structure of neuronal axonal cylinders and compartmentalization of WM extracellular spaces all changes that affect the diffusion of water in cerebral WM [8 10]. It is not clear how such diverse developmental changes in the structure and composition of WM are reflected in quantitative assessments of ADC and FA in children of different ages. Our overall purpose was to use DWI and DTI to assess the postnatal maturation of WM in human infants in an attempt to relate developmental changes to changes in structure and composition of WM. Previous DWI and DTI studies have typically focused on premature infants, neonates in the first week of life, and children in the first decade of life [1, 2, 11 13]. Few studies have focused on maturational changes throughout the first year of life [14], when substantial neuroanatomic and functional maturation occurs [15]. The maturational changes in WM during this period were the focus of this study. It is well established that in the brains of older children and adults, substantial differences exist between the diffusional properties of tissue in deep WM structures (long compact WM pathways of the forebrain) and those in peripheral WM structures (less compact, subcortical WM) [16]. We had two goals in studying maturational changes in WM of infants. The first goal was to determine whether such differences exist in term infants. Such differences between deep WM and peripheral WM, if they exist, are not well documented in the radiologic literature. We compared the water diffusion properties of long tracks of the deep forebrain with those of more peripheral WM regions that underlie associational cortex to examine disparate degrees of structural and functional maturity in this early phase of postnatal development [17, 18]. Our second goal was to study whether rate of change in diffusion characteristics generally thought to be representative of myelination (e.g., FA and ADC) differs between compact WM and peripheral WM during the first year of life. The results of one study [3] suggest that a difference may exist in the rate of increase in FA among peripheral WM structures in comparison with deep tracks in children 1 6 years old, peripheral WM having a slightly greater rate of increase. Our second goal was to determine whether this previously reported greater rate of maturation of peripheral WM could be confirmed and extended to the first postnatal year in a sample with greater temporal resolution. Measurements of rates of diffusional changes would be expected to increase our understanding of how and when different neural systems mature in infancy and early childhood. Furthermore, documentation of normative radiologic data against which abnormal brains can be measured would be valuable for assessment of infants and children with various CNS disorders, such as leukodystrophy. Materials and Methods Study Population The initial study population consisted of 57 infants 1 week to 11 months old who underwent clinical MRI that included DTI. Four infants were excluded because results of a neurologic examination within the follow-up period indicated developmental delay necessitating further assessment by the pediatric neurology service at our institution. The other 53 infants (35 boys, 18 girls) had normal MRI findings. Continuous review of the electronic medical records at our institution over a 5-year period showed these infants had never had neurologic disease. The institutional review board at our medical center granted a waiver of informed consent for inclusion of patients. Images were obtained over 14 months with a total of four 1.5-T MRI units manufactured by the same vendor (Signa, GE Healthcare) and operating on the same imaging platform. All infants underwent one MRI examination. The distribution of gestational ages at birth was as follows: 32 weeks (n = 1), 35 weeks (n =1), 36 weeks (n = 2), 37 weeks (n =3), 38 weeks (n =6), 39 weeks (n = 2), 40 weeks (n = 34), and 41 weeks (n = 4). We consulted a pediatric neurologist at our institution, who was not an investigator in this study and who had 30 years of experience with multiple-institution infant studies, about the appropriateness of including premature infants in the study. For the purposes of our study, infants born at 38 weeks of gestation or later were considered so close to the standard postconceptual age at birth of 40 weeks that we did not consider the premature infants substantially different from infants born at 40 weeks. We were advised to include premature infants (who did not have adverse birth events and who had no subsequent neurologic disability on review of medical records over a period of years). These premature infants typically had undergone imaging many weeks after birth, and this factor minimized the effect of prematurity. The age at which imaging was performed, adjusted for premature birth, was 43 weeks for the infant born at 32 weeks; 30 weeks for the infant born at 35 weeks; 16 weeks and 46 weeks for the two infants born at 36 weeks; and 9, 21, and 32 weeks for the infants born at 37 weeks. FA data were available for analysis in all 53 cases. ADC data derived from the three-direction diffusion imaging sequence were available for 51 infants. For two infants, however, ADCs were not available because the optical disks on which their data sets were stored became corrupted. All other pulse sequences had already been stored on our PACS network. Indications for clinical MRI were as follows: scalp nevus or other scalp or facial lesion (n = 14), increased head circumference (n = 6), possible tethered spinal cord (n = 4), nonaccidental trauma (n = 4), suspected visual abnormality (n = 4), irritability (n = 4), sleep apnea (n = 2), suspected encephalocele (n =2), hearing deficit (n = 2), mild hypertonia (n = 2), congenital cardiac or abdominal abnormality (n = 2), cranial nerve palsy (n = 1), tachypnea (n = 1), aspiration (n = 1), microcephaly (n = 1), abnormal movements (n = 1), cranial bruit (n = 1), and congenital nystagmus (n =1). The age of each subject in weeks at the time of imaging was available in the medical record. We normalized this age to gestational age at birth and expressed age at imaging as an adjusted postnatal age at the time of imaging. For infants born prematurely (before 40 weeks of gestation), adjusted postnatal age at imaging was calculated as follows: age in weeks at imaging (40 weeks gestational age at birth). Thus an infant who was born at 38 weeks of gestational age and underwent imaging 10 weeks after birth was assigned a normalized age of 8 weeks. For infants born after term (after 40 weeks of gestation), adjusted postnatal age at imaging was calculated as follows: age in weeks at imaging + (gestational age at birth 40 weeks). Thus an infant who was born at 41 weeks of gestational age and underwent imaging 10 weeks after birth was assigned a normalized age of 11 weeks. Adjusted postnatal ages at imaging (reported in 4- week epochs) were as follows: 2 to 0 weeks (n = 1), 1 4 weeks (n = 4), 5 8 weeks (n = 5), 9 12 weeks (n = 7), weeks (n = 6), weeks AJR:189, August

3 (n = 4), weeks (n = 1), weeks (n =5), weeks (n = 4), weeks (n = 4), weeks (n = 0), weeks (n = 7), weeks (n = 3), and weeks (n =2). Imaging Parameters For each MRI examination, unenhanced transverse T1-weighted, transverse intermediateweighted, and transverse T2-weighted images were obtained in addition to the DTI protocol. All imaging was performed with a 1.5-T clinical MRI unit (Signa, GE Healthcare) with a standard head coil. Diffusion imaging was performed in the transverse plane with a spin-echo echo-planar imaging sequence and the following parameters: TR/TE, 12,000/100; inversion time, 2,200 milliseconds; diffusion gradient encoding in three orthogonal directions; b = 1,000 s/mm 2 ; field of view, cm; matrix size, ; slice thickness, 5 mm; gap, 2.5 mm; number of signals acquired, 1. ADC was calculated with the following equation: lns(g)=ln S(0) 2[γ 2 G 2 δ 2 (Δ δ / 3)] ADC, where G is the amplitude of the pulsed diffusion gradient, γ is the gyromagnetic ratio, Δ is the interval between the diffusion gradients, δ is the duration of the diffusion gradients, S(G) is the signal strength with pulsed diffusion gradient on, and S(0) is the signal strength with the pulsed diffusion gradient off [16]. The DTI protocol consisted of the following single-shot spin-echo echo-planar sequence: 12,000/101; field of view, 22 cm 2 ; matrix size, ; 6-mm contiguous slices through entire brain; number of excitations, 2. Diffusion gradients were encoded in six directions with a b value of 1,000 s/mm 2 and an additional image with no diffusion gradient (b = 0 s/mm 2 ). Imaging was performed through the entire brain. The T2- weighted sequence parameters were 2,800/100; field of view, 22 cm 2 ; matrix size, 256 (frequency direction) 192 (phase direction); slice thickness, 5 mm; gap, 2.5 mm; number of excitations, 2. The T1-weighted sequence parameters were 500/14; field of view, 22 cm; matrix size, 256 (frequency direction) 194 (phase direction); slice thickness, 5 mm; gap, 2.5 mm; number of excitations, 2. Generation of ADC Maps and FA Maps ADC and FA maps were generated with commercial (Functool, GE Healthcare) and proprietary software. For computation of diffusion tensors, the raw diffusion tensor data were transferred to an independent workstation (Advantage Windows, GE Healthcare) and processed. The six independent elements of the diffusion tensor, D xx, D yy, D zz, D xy, D xz, and D yz, were statistically calculated for each voxel with a previously described method and based on the following equation: ln [A(b) / A(b = 0)] = Σ i Σ j b ij D ij, where b ij is the component of the ith row and jth column of the diffusion gradient matrix b, A(b) is the resulting echo intensity for a gradient sequence with directions and magnitudes of the diffusionsensitizing gradients described by the b matrix, A(b = 0) is the echo intensity when b is the zero matrix (no diffusion gradient), and D ij is the corresponding component of the diffusion tensor matrix D [19, 20]. After the elements of the diffusion tensor were obtained, its eigenvalues were determined by diagonalization of the tensor matrix. FA was chosen for the index of anisotropy because it is rotationally invariant, provides excellent gray matter to WM contrast, and has a high contrast-to-noise ratio [21]. FA is also the most widely used index of anisotropy in recent literature, which facilitates comparison with data from previous studies and other investigators. FA represents the anisotropic portion of total diffusion. The following equation is used for FA: FA = (3/2) 1/2 [(E 1 d) 2 +(E 2 d) 2 +(E 3 d) 2 ] 1/2 / [E 1 2 +E 2 2 +E 3 2 ] 1/2, where E i = the three eigenvalues and d = (E 1 +E 2 +E 3 ) / 3 [16]. Values for FA range from 0 to 1, where 0 represents isotropic diffusion and 1 represents extremely anisotropic diffusion [16]. FA does not have a unit because it is a ratio of diffusion coefficients. The calculations for FA were performed for each voxel and displayed as an anisotropy map. One neuroradiologist with 7 years of neuroradiology experience conducted a visual and qualitative inspection of FA maps before quantitative analysis was performed. Regions of Interest Using Functool, a single observer blinded to age drew standard (44 ± 4 cm 2 ) regions of interest (ROIs) on ADC and FA maps with reference to corresponding T2-weighted images, DW images, and DTI images displayed side by side (Fig. 1). The analyst identified the anatomic structures of interest on the T2-weighted images and the DW image (for ADC measurement) and the DT image (for FA measurement). In the analysis program we used, ROIs placed on the DW image were automatically simultaneously placed on the corresponding site on the relevant ADC map. In a similar manner, ROIs placed on one of the DTI images were automatically simultaneously placed on the corresponding site on the relevant FA map. Our aim was to reliably and consistently sample representative long tracks of the deep WM and more peripheral WM regions underlying the associational cortex. The deep WM structures sampled were the posterior limb of the internal capsule and the genu and splenium of the corpus callosum. ROIs were drawn on axial images obtained from the center of the diencephalon, where the internal capsule becomes most distinct (Fig. 1). The peripheral WM regions sampled were subcortical WM underlying the superior frontal gyrus and the superior parietal lobule. ROIs for the frontal and parietal WM were drawn on the image in a position one or two slices superior to the roof of the lateral ventricle. In each case, the observer placed the ROI in the site that showed the lowest ADC or the greatest FA within the specified region. For instance, to sample ADCs in the posterior limb of the internal capsule, the observer placed the ROI on the slice that presented the widest diameter of the internal capsule and moved the ROI within the posterior limb until the region showing the lowest ADC was found. Analysis of ADC and FA Measurements Mean ADC and FA values from the five ROIs for each subject were averaged categorically so that for each infant, we obtained mean ADC and FA values for deep and peripheral WM. Regression models were used for each set of diffusion indexes (ADC and FA) against age. Both data sets were best fit with linear regression models with fitted breakpoints at day 100. Using these models, we obtained estimated ADC and FA values at term for deep and peripheral WM using the intercept of the initial linear period (the fitted line before day 100) with day 0 (day of birth adjusted for gestational age). We similarly obtained estimated ADC and FA values at day 30 and day 200 from the early (before day 100) and late (after day 100) linear fits. These days were arbitrarily chosen to provide additional cross-sectional time points for statistical evaluation of differences between deep and peripheral WM in the early and late phases of maturation during the first year. Corresponding values for deep and peripheral WM structures and regions were evaluated with Student s t tests. For assessment of possible differences in the rate of maturation in peripheral WM and deep WM, the equality of the regression slopes was tested within the regression model with multivariate analysis of variance. To evaluate the relative maturation of infant ADCs and FA, the data were also expressed in terms of percentage of mean adult values. For ADC, the data were normalized to previously published adult values for the same structures and regions [22] (summary data provided by Gilmore JH, 478 AJR:189, August 2007

4 DTI of White Matter in infants personal communication). For FA, the data were normalized to adult values obtained from a cohort of 16 healthy adults (eight women) years old (mean, 23.4 years) who underwent imaging in a previous study [23]. To facilitate comparison with other studies in which diffusion data are grouped by postnatal month, we also present our ADC and FA measurements grouped into 12 intervals of 4 weeks beginning with day 0 (day of birth adjusted for gestational age). A B C D E F Fig. 1 Healthy 24-week-old girl born at 40 weeks of gestational age. A, T2-weighted MR image at level of internal capsule shows normal appearance of brain. B, Diffusion-weighted image at level of A shows region of interest (ROI) placement in posterior limbs of internal capsules. C, Apparent diffusion coefficient (ADC) map at level corresponding to B shows ROIs in posterior limbs of internal capsules. Mean ADC was mm 2 /s. D, Diffusion-weighted image slightly cephalic to C shows ROI placement in genu and splenium of corpus callosum. E, ADC map at level corresponding to D shows ROIs placed in genu and splenium of corpus callosum. Mean ADC in genu is mm 2 /s and in splenium is mm 2 /s. F, Diffusion tensor image shows ROI placement in posterior limb of internal capsules and genu and splenium of corpus callosum. (Fig. 1 continues on next page) Results ADC and FA at Term Table 1 shows mean ADC and FA values at term for the five WM regions studied. ADCs for the two peripheral WM regions were higher than ADCs for the deep WM structures. In categoric combination, the mean ± standard error of the mean (SEM) ADCs were 1.32 ± mm 2 /s for peripheral WM and 1.09 ± mm 2 /s for deep WM (p < 0.01). Given the general inverse relation between ADC and FA [24], it follows that the FA values for the two peripheral WM regions were lower than the values for the deep WM structures. In categoric combination, the mean ± SEM FA values were 0.16 ± 0.01 for peripheral WM and 0.36 ± 0.02 for deep WM (p < 0.01). Expressed as percentage of mean adult value, the higher ADC and lower FA values of pe- AJR:189, August

5 G H I J K L Fig. 1 (continued) Healthy 24-week-old girl born at 40 weeks of gestational age. G, Fractional anisotropy (FA) map at level corresponding to E shows ROIs placed in posterior limb of internal capsule and genu and splenium of corpus callosum. Mean FA value in posterior limb of internal capsule is 0.492, in genu is 0.514, and in splenium is H, T2-weighted MR image at level of subcortical white matter shows normal appearance of brain. I, Diffusion-weighted image at level corresponding to H shows ROI placement in frontal subcortical white matter and parietal subcortical white matter. J, ADC map at level corresponding to I shows ROI placement. Mean ADC in frontal subcortical white matter is mm 2 /s and in parietal subcortical white matter is mm 2 /s. K, Diffusion tensor image shows ROI placement in frontal subcortical white matter and parietal subcortical white matter. L, FA map at level corresponding to K shows ROI placement. Mean FA value in frontal subcortical white matter is and in parietal subcortical white matter is ripheral WM were further from corresponding mean adult values than were the ADC and FA values of deep WM. Thus mean ADC and FA in peripheral WM were 163% and 32%, respectively, of mean adult values, whereas the ADC and FA for deep WM were 143% and 54%, respectively, of the mean adult values. These differences in normalized diffusion indexes between peripheral WM and deep WM were highly significant (p <0.01). Although our chief interest was a categoric comparison of representative peripheral WM regions and deep WM structures, we did detect statistically significant differences in both ADC and FA at term within the deep WM group (p < 0.01) (Table 1). For the two peripheral WM regions sampled, there were no differences in absolute ADC or FA; however, after normalization to mean adult values, differences were detected (p < 0.01). Because we had no a priori hypothesis regarding differences within WM categories, no further analysis of these differences was pursued. 480 AJR:189, August 2007

6 DTI of White Matter in infants TABLE 1: Mean Fractional Anisotropy and Apparent Diffusion Coefficient Values (± Standard Error) for Deep and Peripheral White Matter Structures at Term Structure Deep white matter b Fractional Anisotropy Apparent Diffusion Coefficient ( 10 3 mm 2 /s) a Genu of corpus callosum 0.38 ± 0.03 (57 ± 4) 1.18 ± 2 (157 ± 3) Splenium of corpus callosum 0.34 ± 0.02 (45 ± 3) 1.11 ± 2 (138 ± 3) Posterior limb of internal capsule 0.37 ± 0.02 (58 ± 3) 0.97 ± 1 (135 ± 1) Peripheral white matter Frontal subcortical white matter 0.16 ± 0.01 (35 ± 3) 1.33 ± 2 (179 ± 2) Parietal subcortical white matter 0.15 ± 0.01 (29 ± 3) 1.31 ± 1 (148 ± 1) Note Values in parentheses are percentage of mean adult value and are derived from a linear model with a fitted breakpoint at 100 days (Figs. 2 and 3). The intercept of the initial linear period with day 0 (day of birth adjusted for gestational age) is an estimate of fractional anisotropy and apparent diffusion coefficient at term. a Mean adult apparent diffusion coefficients from [22], provided by Gilmore JH (personal communication). b Significant differences detected within group of deep white matter structures for both fractional anisotropy and apparent diffusion coefficient (p < 0.01). ADC Age at Imaging (wk) Fig. 2 Graph shows apparent diffusion coefficients (ADCs) for peripheral and deep white matter throughout first year of life. Each symbol represents mean deep ( ) and peripheral ( ) white matter values for one subject. Figure 1 shows regions of interest in representative subject. At right of each graph are adult mean ± standard error of the mean. Values were fit by broken-line linear regression functions with break points at 100 days. Increased rates of change in ADC in first 3 months are indicated by steeper slopes of initial linear fits compared with second fitted regression lines. Age normalized to term. ADC and FA Throughout the First Year Mean ADC and FA for peripheral WM and deep WM in each case were plotted against age (normalized to term) at imaging (Figs. 2 and 3). Regression models were fit to each distribution. The best fits were obtained with brokenline models with breakpoints at day 100 (all p < 0.01). Thus the rate of change in ADC and FA during the first year in both peripheral WM and deep WM was not constant. Both diffusion indexes matured at a faster rate in approximately the first 3 months (before day 100) compared with the rest of the year (all p < 0.01). In particular, the rates of change for peripheral WM were nearly fivefold and twofold greater for ADC and FA, respectively, before day 100 than in the rest of the year. For deep WM, the rate of change between early and later periods increased approximately twofold for ADC and threefold for FA. We next asked whether categoric differences existed between peripheral WM and deep WM in rates of change in ADC and FA in the early (before day 100) and later (after day 100) periods. In the early period, statistically significant differences were detected between peripheral WM and deep WM in both rate of ADC change and rate of FA change; however, discordant results were obtained with the two diffusion indexes. In the early period, ADC in peripheral WM decreased at approximately twice the rate of ADC in deep WM (rate of change for peripheral WM, mm 2 /s/wk; rate of change for deep WM, mm 2 /s/wk; p < 0.01). Unexpectedly, the opposite pattern was observed for FA. FA in peripheral WM in the early period increased at only approximately one half of the rate of FA in deep WM (rate of change for peripheral WM, 0.005/wk; rate of change for deep WM, 0.009/wk; p = 0.01). Throughout the late period, no differences were observed between peripheral WM and deep WM in rates of change in either ADC (rate of change for peripheral WM, mm 2 /s/wk; rate of change for deep WM, mm 2 /s/wk; p = 0.64) or FA (rate of change for peripheral WM, 0.002/wk; rate of change for deep WM, 0.003/wk; p =0.08). In addition to estimating rates of change in ADC and FA in peripheral WM and deep WM during the first year, we used the regression models to estimate differences in absolute ADC and FA at two cross-sections through the distributions: one date in the early period (day 30) and one in the later period (day 200). The significant differences in ADC and FA between peripheral WM and deep WM present at term persisted at both time points. Peripheral WM regions maintained a higher ADC and less FA than deep WM structures (p < 0.01) (Figs. 2 and 3). To facilitate comparison of our data with values found in other studies of infant subjects, we grouped our sample into 12 intervals of 4 weeks beginning on day 0 (day of birth adjusted for gestational age). These grouped values for ADC and FA throughout the first year are reported in Tables 2 and 3. We examined the effect of gender on the analysis of ADC and FA maturation in peripheral WM and deep WM. We compared the age distribution in the two gender classes by means of a two-group Student s t test and found no statistically significant difference in age distribution (p > 0.3). We then tested ADC and FA for peripheral WM and deep WM in a general linear model including gender effects for level and slopes. No significant gender effects were detected (p >0.2). AJR:189, August

7 FA Age at Imaging (wks, normalized to full-term) Fig. 3 Graph shows fractional anisotropy (FA) for peripheral and deep white matter throughout first year of life. Each symbol represents mean peripheral ( ) or deep ( ) white matter FA values for one subject. At right of each graph are adult mean ± standard error of the mean. For each data set, values were fit by broken-line linear regression functions with break points at 100 days. As with apparent diffusion coefficient (Fig. 2), rates of change in FA were greater in the first 3 months compared with the rest of the year, as indicated by steeper slopes of initial linear fits compared with second fitted regression lines. TABLE 2: Mean Apparent Diffusion Coefficients for Deep and Peripheral White Matter During First Year Deep White Matter Peripheral White Matter Interval n Genu of Corpus Callosum Splenium of Corpus Callosum Posterior Limb of Internal Capsule Frontal Parietal (153) 1.10 (136) 0.95 (133) 1.32 (177) 1.28 (144) (144) 1.05 (130) 0.91 (127) 1.17 (158) 1.16 (131) (143) 1.10 (136) 0.90 (125) 1.16 (156) 1.14 (129) (133) 0.99 (123) 0.85 (118) 1.06 (143) 1.03 (116) (129) 1.02 (126) 0.86 (121) 1.05 (141) 1.02 (114) (110) 1.06 (130) 0.82 (115) 1.03 (138) 0.95 (107) (131) 0.93 (115) 0.80 (112) 1.00 (135) 0.96 (109) (129) 0.86 (107) 0.80 (112) 1.00 (134) 0.95 (107) (129) 0.82 (101) 0.79 (110) 0.96 (128) 0.94 (106) (122) 0.84 (104) 0.79 (111) 0.93 (124) 0.92 (103) (133) 0.75 (93) 0.78 (109) 0.93 (125) 0.91 (103) (127) 0.74 (91) 0.77 (107) 0.92 (123) 0.88 (99) SD of estimate a 0.05 (6) 0.04 (5) 0.02 (3) 0.05 (6) 0.03 (3) Note Values are apparent diffusion coefficient 10 3 mm 2 /s. Values in parentheses are percentage of mean adult value. Mean adult apparent diffusion coefficients obtained from [22], provided by Gilmore JH (personal communication). Measurements for first year are grouped into 12 intervals of 4 weeks beginning with gestational age adjusted to term as day 0. a SD estimates derived from one-way analysis of variance with 4-week interval as the classification variable and data pooled over the 12 intervals (39 degrees of freedom). Discussion The purpose of this study was to use DWI and DTI to investigate the status of cerebral WM at term and the rate of WM maturation during the first year of life in healthy infants. Our results can be summarized as four principal findings. First, substantial differences exist in the diffusional properties of representative peripheral WM regions and deep WM structures in the infant brain at term. Second, ADC and FA mature at a faster rate in the first 3 months than in the rest of the year. Third, ADC and FA change at different rates in peripheral and deep WM during this early (first 3 months after birth) period of rapid maturation. Finally, in the other months of the first year, ADC and FA continue to mature at slower but very similar rates in peripheral WM and deep WM. Status of Cerebral WM at Term The first major finding in our study was that substantial differences exist between the diffusional properties of peripheral WM and those of deep WM in the infant brain, even at term gestation. The ADC in peripheral WM was significantly higher and FA significantly less than the corresponding values in deep WM (Table 1). Deep WM structures had already achieved approximately one half of the mean adult ADC and FA at this age, and peripheral WM remained barely discernible at the thresholds of detection (Table 1). ADC and FA are thought to reflect a number of WM features, including degree of myelination and volume of the extracellular compartment, amount of extracellular water, changes in composition of the extracellular matrix, density and organization of axons in WM structures, maturation of neurofilaments and other constituents of the axonal cytoskeleton, and maturation of voltage-gated conductances along axonal membranes [9, 25 28]. At term, the deep structures we studied (portions of the corpus callosum and internal capsule) are still immature with respect to each of these cytologic and histologic features of WM [5]. Nevertheless, each deep WM structure is well established at this age, having been pioneered in the second trimester by long axonal projections arranged into relatively compact parallel bundles of axonal fascicles [29 31]. In contrast, the peripheral WM of the prefrontal and posterior parietal cortex that we sampled was much less developed at term, being only poorly myelinated and sparsely populated by associational fibers that have yet to establish widespread corticocortical connections [10, 32]. These peripheral WM structures consequently are likely to retain larger extracellular volumes with more extracellular fluid. Therefore, our findings of a significantly greater ADC (greater water diffusion) and less FA (more isotropic diffusion) in peripheral WM than in deep WM are consistent with known microanatomic characterizations. 482 AJR:189, August 2007

8 DTI of White Matter in infants TABLE 3: Mean Fractional Anisotropy Values for Deep and Peripheral White Matter Structures During First Year Deep White Matter Peripheral White Matter Interval n Genu of Corpus Callosum Splenium of Corpus Callosum Posterior Limb of Internal Capsule Frontal Parietal (56) 0.38 (50) 0.40 (63) 0.15 (33) 0.15 (29) (66) 0.41 (54) 0.38 (60) 0.19 (41) 0.19 (37) (72) 0.40 (53) 0.46 (73) 0.21 (46) 0.21 (40) (81) 0.52 (68) 0.48 (75) 0.21 (45) 0.24 (45) (67) 0.52 (68) 0.50 (79) 0.20 (44) 0.25 (48) (71) 0.52 (68) 0.55 (87) 0.21 (47) 0.23 (44) (76) 0.54 (71) 0.53 (83) 0.22 (49) 0.26 (49) (72) 0.58 (76) 0.53 (84) 0.27 (58) 0.25 (48) (79) 0.56 (74) 0.51 (80) 0.25 (54) 0.26 (50) (86) 0.62 (82) 0.58 (92) 0.24 (52) 0.32 (61) (85) 0.61 (80) 0.54 (85) 0.29 (64) 0.32 (61) (94) 0.66 (86) 0.57 (90) 0.32 (69) 0.28 (54) SD of estimate a 0.06 (9.2) 0.06 (7.5) 0.03 (5.5) 0.03 (5.9) 0.03 (5.9) Note Values in parentheses are percentage of mean adult value. Measurements for first year are grouped into 12 intervals of 4 weeks beginning with gestational age adjusted to term as day 0. a SD estimates derived from one-way analysis of variance with 4-week interval as the classification variable and data pooled over the 12 intervals (41 degrees of freedom). One goal of our study was to characterize the status of peripheral WM and deep WM in terms of mean ADC and FA in the infant brain at term. Rather than obtaining a large sample of term neonatal infants, we used statistical methods to estimate mean diffusional indexes at this important developmental mark. To confirm the validity of this approach, it is important to compare our neonatal values with those obtained in previous studies of term infants. Because of the lack of consistent, well-defined ROI locations, it is a challenge to compare our mean ADC and FA findings in peripheral WM with the findings of previous studies. It is possible, however, to compare our data on deep WM structures with those from previous studies because of the greater consistency and precision with which deep structures can be resolved and assessed. For instance, in one study [13], the mean FA values in the first 2 months of life were 0.38 for the genu and 0.42 for the splenium, similar to the mean FA values of 0.37 for the genu and 0.38 for the splenium during the first month of life in our study. Likewise, in the other study, the mean FA in the first 2 months of life for the internal capsule (presumably the posterior limb, because the sample was acquired with the aim of including the corticospinal tract) was 0.45, only slightly greater than our value of 0.40 in the first month. In another study [22], which differed from ours in the use of a 3-T MRI unit and 12 signals averaged in the DTI sequence, neonatal (mean age, 16 days) and adult ADC and FA were reported only in bar graphs, but we acquired mean values from the authors (Gilmore JH, personal communication). In that study, the ADCs and FA values were mm 2 /s and 0.47 for the genu of the corpus callosum, mm 2 /s and 0.63 for the splenium of the corpus callosum, and mm 2 /s and 0.45 for the internal capsule. These ADCs are reasonably close to our values at term; however, the FA values are considerably higher than ours, especially for the corpus callosum (Table 1). The differences in field strength and DTI sequence might have contributed to the discrepant FA values between the studies. Further studies from several institutions are needed for definitive establishment of normative ADCs and FA values for peripheral WM regions and deep WM structures in term neurologically healthy infants. Our data are intended to provide a clinically meaningful contribution toward that end. Changes in Cerebral WM Early in the First Year of Life Additional findings of our study pertain to changes in diffusion indexes that occur during the first year of life as WM continues to differentiate and mature. The first of these findings was that ADC and FA change at a faster rate in the first 3 months after birth at term gestation than in the rest of the year. This observation is qualitatively similar to previous findings [2, 25, 28, 33] of an exponential decline in ADC in several peripheral WM and deep WM (and gray matter) structures, the period of greatest change occurring before the end of the first month after birth at term gestation. Our ADC (and FA) data were best fit with broken-line linear regression models with a breakpoint at 100 days, indicating that the period of rapid change in ADC and FA continues beyond the first month of life. When we compared the rates of change in ADC and FA for peripheral WM and deep WM during this early period of rapid maturation, we found statistically significant differences between WM categories. Results of a study by Forbes et al. [25] also suggested a difference between the rate of decrease in ADC in peripheral WM and the rate in deep WM in the first few months of life, although the rates of ADC change were not quantified. In that study, graphs indicated that the greatest rate of decline in ADC was in peripheral (anterior and posterior subcortical) WM, with less-rapid change occurring in deep WM (anterior and posterior limbs of internal capsule). Consistent with that suggestion, our findings showed that the ADC in peripheral WM decreased at approximately twice the rate it did in deep WM in the first 100 days after birth at term (steeper slope of the peripheral WM regression line in Fig. 2). Unexpectedly, we did not observe a comparable difference in the rates of change in FA between peripheral and deep WM in this same early period. Instead, the opposite pattern was observed: The rate of change in FA in peripheral WM was approximately one half of the rate of change in FA in deep WM (steeper slope of the deep WM regression line in Fig. 3). That both ADC and FA change rapidly during the first 3 months of life attests to the accelerated progression of WM development in both peripheral WM regions and deep WM structures in the perinatal period. However, the disparities we observed in the rates of change in ADC and FA in deep WM and peripheral WM attests to the heterogeneous nature and differential progression of WM histogenesis and maturation in different parts of the forebrain. A recent report [14] of maturational changes in mean diffusivity and FA in the infant brain during the first 4 months of life described similar isolated changes in mean diffusivity and FA (presumably, dispro- AJR:189, August

9 portionate changes in one diffusional index not matched by complementary changes in the other). Taken together these findings raise the possibility that ADC and FA are sensitive to different aspects of WM maturation in this early period of rapid change [9, 27]. One possibility is that ADC is most sensitive to earlier molecular and histologic changes associated with the so-called pre myelin sheath stage of cerebral myelination [34]. If so, the greater rate of ADC change in peripheral WM would reflect the relative delay in progression toward a larger number of mature myelin sheaths in peripheral WM than in deep WM in this phase of the perinatal period [5, 6]. This possibility does not exclude contributions from other factors that restrict the content of water and its diffusional patterns in WM, nor does it imply that early rapid changes in FA may not also reflect processes that occur before myelination [14, 26]. Results of histologic studies of peripheral WM in human infants suggest a variety of changes that can account for the rapid decrease in ADC (and rapid increase in FA) that we observed in the first 3 months, including an increase in number and density of axons and an increase in number and phosphorylation of axonal neurofilaments, in addition to proliferation and compaction of myelin sheaths and expression of myelin basic protein [10, 34]. In one such postmortem study [10], the results of which characterized the development of peripheral WM underlying parietal associational cortex in human infants, the period of the most rapid change occurred in two early time periods: and postconceptional weeks. The breakpoints at 100 days identified in our linear regression models (the inflection points when the rates of change in ADC and FA decreased) corresponded approximately to postconceptional week 54. Thus our early period of rapid change in diffusional indexes corresponds remarkably well to the progression of WM histogenesis and maturation, at least for peripheral WM in the parietal lobe. Further detailed postmortem investigations of developing WM in the human brain of the type undertaken by Haynes et al. [10] are necessary to understand the differences between the developmental progression of peripheral and deep WM and how such events relate to radiologic assessments of water diffusion in cerebral tissue. Changes in Cerebral WM Later in the First Year of Life After the first 3 months of life, we detected no differences between the rate of change in ADC and FA in peripheral WM and the rate in deep WM. Both types of WM had a slowed but steady progression in maturation toward adult values throughout the rest of the first year. These findings are generally consistent with those of previous reports of ADC and FA data on infants in the first year of life [2, 25, 28, 33, 35]. However, methodologic differences and variation in the size, shape, and precise placement of ROIs make it difficult to compare ADC and FA for infants of any age and their rates of change throughout early infancy. Nevertheless, some meaningful comparisons can be made. Other authors [3, 15] have found that ADC and FA in various WM structures mature in both preterm and term infants and have shown differences between the ADC and FA of earlier maturing tracts (e.g., components of the corpus callosum) and those of later maturing tracts (e.g., subcortical WM regions). In one early study, in which FA was measured with only two orthogonal diffusion gradients, the authors [1] concluded that changes in anisotropy occur only in the first 6 months of life. In subsequent studies, however, investigators [2, 3, 15] found that anisotropy increases in major WM regions well beyond that period. Our findings support the assertion that WM anisotropy continues to increase and ADC to decrease throughout the first year of life, although not at the same rate of change as in the first 3 months. This finding would be expected given that both marked developmental changes and alterations in signal intensity consistent with myelination continue throughout and beyond the first year of life [9]. For example, Forbes et al. [25] calculated ADC in the posterior limb of the internal capsule through the first year of life. The mean ADC at birth appeared to be approximately mm 2 /s, similar to the mean value we found for the first month of life ( mm 2 /s). In that study, the mean ADC in the posterior limb of internal capsule in the last 3 months of the first year of life appears to have been approximately mm 2 /s (20% decrease), similar to the mean value of mm 2 /s (18% decrease) over the same period in our study. Thus both the mean values and the rate of ADC decline throughout the first year in the two studies are similar. These changes compare relatively well with those found in a third study [2], in which the mean ADCs in the posterior limb of the internal capsule were approximately mm 2 /s at birth and approximately mm 2 /s (30% decrease) in the last 3 months of the first year of life. When we compared changes in ADC of posterior WM with results from previous studies, we found similar rates of decrease in ADC throughout the first year. Forbes et al. [25] calculated ADC in posterior subcortical WM on the same slice that showed the basal ganglia, which was at a more inferior level than the one we used to calculate parietal ADC above the roof of the lateral ventricles. Nonetheless, some limited comparisons can be made. The mean ADC of posterior WM calculated by Forbes et al. appears to have been approximately mm 2 /s in the first postnatal month compared with mm 2 /s in our study during that period. The mean ADC of posterior WM in the last 3 months of the first year in the study by Forbes et al. appears to have decreased to approximately mm 2 /s (27% decrease) compared with mm 2 /s (29% decrease) in our study during that period. Thus the rate of decrease in ADC over the first year of life for ROIs admittedly placed at different locations in the posterior WM appears quite similar in multiple studies. Implications for Relative Contributions of Axonal Growth and Myelination By the end of the first year of life, mean ADCs in all WM regions were within 27% of mean adult values. Although statistically significant differences between the ADC of deep WM and that of peripheral WM were detected at the day 200 cross section through our data set (and persisted throughout the rest of the first year), it would be difficult to attribute any biologic or clinical significance to such small differences between the ADCs of the WM categories at this later stage of the first year (Fig. 2). Rather, these findings likely indicate that the molecular and microstructural changes in WM that are responsible for the volume of extracellular water and the restriction of water diffusion are substantially mature after 1 year. This inference can be extended to diffusional anisotropy but only for deep WM. Each of the deep WM structures assessed had achieved FA no less than 86% of mean adult values by the end of the first year (Table 3), again indicating substantial maturation of diffusion anisotropy in deep WM during the first year. In contrast, the two peripheral WM regions had achieved only 69% and 54% of mean adult values for frontal and parietal WM, respectively, by the end of the first year (Table 3). This finding suggests the ongoing development of molecular and microstructural factors that promote anisotropic diffusion beyond the first year in this type of WM. It is 484 AJR:189, August 2007

10 DTI of White Matter in infants well known that the process of myelination in peripheral WM continues beyond the first year [4 6]. The ongoing consolidation and proliferation of myelin sheaths is one factor that should contribute to the remaining increase in FA that occurs after the first year and to ongoing changes in MR signal intensity thought to reflect progressive myelination [9, 36]. A variety of factors other than myelination may contribute to ongoing increases in anisotropy late in the first year and thereafter, especially in peripheral WM underlying associational cortical regions. Supporting evidence is most clear in studies of animals deficient in myelin. In one study [37], anisotropy within WM was clearly present in myelin-deficient rats, although to a lesser degree than in control rats. The authors concluded that myelination is not a prerequisite for the development of anisotropy but that the presence of myelination increases the degree of anisotropy. In another study, the investigators [38] examined anisotropy in two directions, parallel to axonal fibers (axial diffusivity) and perpendicular to their length (radial diffusivity), in mice with incomplete myelin formation reflecting dysmyelination (shiverer mice). That study showed that axial diffusivity was maintained at normal levels in shiverer mice compared with control mice, suggesting that axonal fibers (and not solely myelination) are the main contribution to axial diffusivity in that animal model. The results of these animal studies raise the possibility that a significant factor in the ongoing increase in FA in peripheral WM is the continued addition of new axonal projections and collateral fibers. Interestingly, growthassociated protein 43 (GAP43), a marker of axonal growth and elongation, continues to be expressed at high levels, relative to the levels in adult samples, in parietal WM throughout the first year, although peak expression is in the preterm period [10]. This finding suggests that axonal growth and elongation in parietal WM continue at a substantial pace at least during the second half of the first year of postnatal life, when the mean FA of peripheral WM is only approximately one half of the normal adult value. Limitations Our study was subject to a number of limitations, which need to be considered in assessing our findings. First, all ADCs and FA values were single recordings made by single observers. Although the observers had substantial experience in recording ADCs and FA values in the same structures in previous studies, we cannot provide any estimate of intraobserver or interobserver variability. Second, because our images were acquired for clinical purposes in the care of infants who had normal findings on conventional MRI studies and did not need subsequent imaging, we recorded data at only a single time point for each child. Therefore, our data set reflects trends in a sample of individuals, whereas serial recordings for individuals might have produced different rates of ADC decreases and FA increases. The third limitation was that coregistration of ADC and FA maps and spin-echo MR images was not performed, which might have resulted in more precise measurements of neuroanatomic regions and structures. However, one would not expect such inaccuracy to result in systematic bias of the data in one direction or another. Some authors [14, 28, 33] have contended that a lower b value (e.g., 700 or 900 s/mm 2 ) than the one used in our acquisitions (1,000 s/mm 2 ) may be more optimal for infants. However, a change in b value would be expected to affect ADC and FA relatively equally in all structures at all ages. Fourth, we obtained MR images with four units and did not measure variability within or between units. However, our results did not differ from those of other published studies on this account and, as with the other limitations, are not expected to have produced a systematic bias that would affect data interpretation. Fifth, our study was aimed at understanding categoric differences in the early status and maturation of peripheral WM and deep WM. We did not develop hypotheses aimed at particular WM regions and tracks. It is possible, therefore, that differences within categories of WM (including regions and structures that we did not measure) might have deviated from the average trends reported. Differences within the deep WM at term were detected statistically and will be the subject of a future investigation. The last limitation was that a head coil specifically designed for neonatal brain imaging, used in some research environments, was not used in this study. The data obtained in this study may differ from results obtained with alternative coils and with better-designed pulse sequences and scanning protocols. However, our purpose was to assess the diffusional properties of WM in the infant brain by use of instrumentation, acquisition protocols, and imaging times representative of most standard clinical imaging environments. Even when the limitations are considered, our normative data on the status of WM in the term infant brain and the rates of change in ADC and FA that occur in the first year should be broadly applicable for clinicians and clinical investigators who must evaluate infants with suspected disorders of cerebral WM. References 1. Nomura Y, Sakuma H, Takeda K, Tagami T, Okuda Y, Nakagawa T. Diffusional anisotropy of the human brain assessed with diffusion-weighted MR: relation with normal brain development and aging. Am J Neuroradiol 1994; 15: Mukherjee P, Miller JH, Shimony JS, et al. Normal brain maturation during childhood: developmental trends characterized with diffusion tensor MR imaging. Radiology 2001; 221: McGraw P, Liang L, Provenzale JM. Evaluation of normal age-related changes in anisotropy during infancy and childhood as shown by diffusion tensor imaging. AJR 2002; 179: Yakovlev PI, Lecours AR. The myelogenetic cycles of regional maturation of the brain. In: Minkowski A, ed. Regional development of the brain in early life. Oxford, UK: Blackwell Scientific, 1967: Brody BA, Kinney HC, Kloman AS, Gilles FH. Sequence of central nervous system myelination in human infancy. Part 1. An autopsy study of myelination. J Neuropathol Exp Neurol 1987; 46: Kinney HC, Brody BA, Kloman AS, Gilles FH. Sequence of central nervous system myelination in human infancy. Part 2. Patterns of myelination in autopsied infants. J Neuropathol Exp Neurol 1988; 47: Paus T, Collins DL, Evans AC, Leonard G, Pike B, Zijdenbos A. Maturation of white matter in the human brain: a review of magnetic resonance studies. Brain Res Bull 2001; 54: Brady ST, Witt AS, Kirkpatrick LL, et al. Formation of compact myelin is required for maturation of the axonal cytoskeleton. J Neurosci 1999; 19: Barkovich AJ. Concepts of myelin and myelination in neuroradiology. Am J Neuroradiol 2000; 21: Haynes RL, Borenstein NS, Desilva TM, et al. Axonal development in the cerebral white matter of the human fetus and infant. J Comp Neurol 2005; 484: Partridge SC, Mukherjee P, Henry RG, et al. Diffusion tensor imaging: serial quantitation of white matter tract maturity in premature newborns. Neuroimage 2004; 22: Mukherjee P, Miller JH, Shimony JS, et al. Diffusion-tensor MR imaging of gray and white matter development during normal human brain mat- AJR:189, August