Abstract

Cerebral white-matter injury is common in preterm-born infants and is associated with neurocognitive impairments. Identifying the pattern of connectivity changes in the brain following premature birth may provide a more comprehensive understanding of the neurobiology underlying these impairments. Here, we characterize whole-brain, macrostructural connectivity following preterm delivery and explore the influence of age and prematurity using a data-driven, nonsubjective analysis of diffusion magnetic resonance imaging data. T1- and T2-weighted and -diffusion MRI were obtained between 11 and 31 months postconceptional age in 49 infants, born between 25 and 35 weeks postconception. An optimized processing pipeline, combining anatomical, and tissue segmentations with probabilistic diffusion tractography, was used to map mean tract anisotropy. White-matter tracts where connection strength was related to age of delivery or imaging were identified using sparse-penalized regression and stability selection. Older children had stronger connections in tracts predominantly involving frontal lobe structures. Increasing prematurity at birth was related to widespread reductions in connection strength in tracts involving all cortical lobes and several subcortical structures. This nonsubjective approach to mapping whole-brain connectivity detected hypothesized changes in the strength of intracerebral connections during development and widespread reductions in connectivity strength associated with premature birth.

Introduction

Premature birth is a major issue for neurologic health. Despite improvements in neonatal intensive care that have led to a decrease in mortality, morbidity remains substantial and neurocognitive deficits involving attention, memory, language, and behavior are increasingly prevalent, persistent, and severe with more extreme prematurity (Bhutta et al. 2002; Johnson 2007; Moster et al. 2008; Saigal and Doyle 2008; Kerr-Wilson et al. 2011). The pathophysiology underlying these deficits is unclear, although cerebral white-matter injury is common in the preterm population and correlates, albeit loosely, with cognitive impairments (Woodward et al. 2006).

Cognitive abilities are likely to be subserved by distributed cortical–subcortical networks (Mesulam 2000; Choi et al. 2008), with efficient cognitive processing dependent on communication between brain regions and therefore the integrity of the connecting white-matter tracts (Khwaja and Volpe 2007; Tamnes et al. 2010). Previous studies with preterm-born subjects have demonstrated cortical and subcortical white-matter changes at term (Anjari et al. 2007; Rose et al. 2008; Ball et al. 2010; Hasegawa et al. 2011), during adolescence and in adulthood (Yung et al. 2007; Constable et al. 2008; Eikenes et al. 2011; Mullen et al. 2011). While changes in white-matter structures such as the corpus callosum have been widely emphasized (Hasegawa et al. 2011; Jo et al. 2012), changes in the rest of the white matter appear more variable, perhaps due to methodological differences between studies. Identifying the pattern and strength of connectivity following premature birth may therefore provide a more comprehensive understanding of the development of cognitive impairments.

We have developed a data-driven, automated method of mapping and analyzing whole-brain structural connectivity on millimeter scale using diffusion magnetic resonance imaging. This combines a modified probabilistic tracking algorithm that provides a global, unrestricted survey of connections (Robinson et al. 2010), with recent statistical techniques including sparse-penalized regression (Tibshirani 1996) and stability selection (Meinshausen and Bühlmann 2010) to detect biologically relevant patterns within the resulting connectivity matrix.

We explore the influence of 2 factors likely to affect observed connectivity in infants born preterm: the age at which the scan was obtained and the postconceptional age at delivery. The age of examination transiently reflects current brain development, while the age at birth may record more lasting effects of premature extrauterine life. We studied preterm children between their first and third year after birth, predicting that the rapid development in anterior brain structures during this period (Yakovlev and Lecours 1967; Brody et al. 1987; Barkovich 2005) would be reflected in a positive association between age at scan and connectivity strength in frontal regions. We then estimated the influence of increasing prematurity to determine whether younger postconceptional age at birth is related to reduced connectivity strength, and which brain structures might be affected.

Materials and Methods

Ethics

The Hammersmith and Queen Charlotte's and Chelsea Local Research Ethics Committee approved this study and written parental consent was obtained prior to scanning (Approval 07/H0707/101).

Subjects

Data from 49 preterm-born children were analyzed. Infants had a median postconceptional age at birth of 28.3 weeks (range 24.6–34.7 weeks) and median birth weight of 0.99 kg (0.56–3.71 kg). Nineteen subjects were scanned with a median corrected age of 13 months (11–16 months), and 37 were scanned with a median corrected age of 25 months (23.5–31.5 months). Seven infants had paired imaging data from both scanning sessions. Subjects were included if high-quality T1, T2, and 32-direction diffusion tensor imaging (DTI) data were all present. Individuals were excluded if focal destructive cerebral lesions (cystic periventricular leukomalacia, hemorrhagic parenchymal infarction, or posthemorrhagic ventriculomegaly) were present on T1-weighted or T2-weighted MRI scans. Twenty-six subjects (53.1%) required mechanical ventilation. Forty-eight had neonatal imaging data of which: 5 (10.4%) subjects had germinal matrix hemorrhage; 4 (8.3%) had punctate white-matter lesions mainly located in the posterior periventricular white-matter; and 11 (22.9%) had enlarged ventricles (mild: 10, moderate: 1; left: 2, right: 2, bilateral: 7). Demographic data are outlined in Table 1.

Table 1

Infant demographic data

Characteristics Total (n = 49) 
Gestational age, median (range) (weeks) 28.29 (24.56–34.71) 
Birth weight, median (range) (kg) 0.986 (0.56–3.71) 
Male, no. (%) 23 (46.9) 
Singleton/twins, no. (%) 28/21 (57/43) 
Premature rupture of membrane (PROM), no. (%) 11 (22.4) 
Late-onset sepsis positive blood culture, no. (%) 9 (18.4) 
Respiratory comorbidity, no. (%) 11 (22.4) 
Patent ductus arteriosus, no. (%) 6 (12.2) 
Intra-uterine growth restriction (IUGR), no. (%) 11 (22.4) 
Characteristics Total (n = 49) 
Gestational age, median (range) (weeks) 28.29 (24.56–34.71) 
Birth weight, median (range) (kg) 0.986 (0.56–3.71) 
Male, no. (%) 23 (46.9) 
Singleton/twins, no. (%) 28/21 (57/43) 
Premature rupture of membrane (PROM), no. (%) 11 (22.4) 
Late-onset sepsis positive blood culture, no. (%) 9 (18.4) 
Respiratory comorbidity, no. (%) 11 (22.4) 
Patent ductus arteriosus, no. (%) 6 (12.2) 
Intra-uterine growth restriction (IUGR), no. (%) 11 (22.4) 

Respiratory comorbidity is defined as the number of infants with chronic lung disease—as requiring oxygen at 36 weeks postmenstrual age. IUGR is defined here as being <10th centile weight. PROM is defined here as >18 h.

MR Imaging

T1-weighted and T2-weighted data were acquired according to a previously defined protocol (Counsell et al. 2008). Single-shot echo-planar diffusion MRI was acquired on a Phillips 3-T system in 32 noncollinear directions using the following parameters: TR 9000 ms, TE 49 ms, slice thickness 2 mm, field of view 224 mm, matrix 128 × 128, voxel size 1.75 × 1.75 × 2 mm3, b-value = 750 s/mm2, SENSE factor = 2, and scanning time of 7 min. A pediatrician trained in MRI procedure was always in attendance. Infants were sedated with oral chloral hydrate (100 mg/kg up to a maximum of 1 g) prior to scanning, and pulse oximetry, temperature, and electrocardiography data were monitored throughout. Ear protection was used for each infant, comprising earplugs molded from a silicone-based putty (President Putty, Coltene Whaledent, Mahwah, NJ, USA) placed in the external ear and neonatal earmuffs (MiniMuffs, Natus Medical, Inc., San Carlos, CA, USA).

Data Analysis and Processing Pipeline

The delineation of structural connections followed a previously described automated protocol optimized for this infant population (Robinson et al. 2010). The pipeline is graphically summarized in Figure 1 and described briefly below with further details in the Supplementary Materials and Methods section.

Figure 1.

Higher level postprocessing figure showing pipeline from raw data through to whole-brain structural connectivity maps and feature selection.

Figure 1.

Higher level postprocessing figure showing pipeline from raw data through to whole-brain structural connectivity maps and feature selection.

Data Acquisition

T1-weighted, T2-weighted images, and 32-direction DTI were obtained, and underwent brain extraction (Smith 2002) and bias correction (Smith et al. 2004). DTI data were inspected before further processing and eddy-current corrected (Fig. 1, step A).

Image Segmentation and Registration

Probabilistic tissue segmentations were obtained from the T1-weighted images using SPM 8 (www.fil.ion.ucl.ac.uk/spm) (Fig. 1, step B). Anatomical segmentation was performed to divide each subject's T1-weighted image into 83 anatomical regions of interest (ROIs; Gousias et al. 2008), listed in Supplementary Table 1 (Fig. 1, step C). Tissue and anatomical segmentations were merged to generate cortical ROIs at the white–gray-matter boundary with subcortical regions intact (see Fig. 1, step D). Cortical ROIs were eroded at the boundaries between regions to create an ROI “ribbon”: one voxel thick, through which structural connections in the brain must pass in order to terminate in the gray matter.

ROIs were transformed from T1 to diffusion (b0) space via an intermediate T1T2 registration (Fig. 1, step E). Image alignment was achieved using voxel-based registration in 3 steps: rigid, affine, and nonrigid registration using IRTK (Rueckert et al. 1999) (www.doc.ic.ac.uk/~dr/software/).

Whole-Brain Tractography

An adapted probabilistic tracking algorithm was implemented to delineate tracts between all ROI pairs, as described in Robinson et al. (Robinson et al. 2010) (Fig. 1, step F). Conventional probabilistic tractography estimates the likelihood that a tract exists between 2 ROIs (Behrens et al. 2003). In contrast, the approach we used integrates a measure of anisotropy in estimating diffusive exchange along the tract, providing a robust approximation of connection strength. This has been previously demonstrated in studies of neonates (Ball et al. 2012) and adults (Robinson et al. 2010).

A connectivity matrix (83 × 83) was generated for each subject each element of which corresponds to the mean anisotropy of a particular tract (Fig. 1, step G).

Statistical Analysis and Feature Stability Selection

Connections that were traced consistently across the group were retained, producing a new, robust n × p matrix—where n is the number of subjects, and p is the number of retained connections, and each row in the matrix represents the set of retained connection strengths for an individual. The matrix was used in 2 regression analyses: In the first, corrected age at scan was the response variable and postconceptional age at birth was treated as a confounder variable. In the second, postconceptional age at birth was the response and age at scan the confounder. We were interested in building a sparse, linear regression model with relatively few non-zero regression coefficients (Tibshirani 1996). The degree of sparsity in the solution can be controlled via a regularization parameter and, to determine the appropriate level of regularization, we applied the recently proposed stability selection procedure which is based on data subsampling with repeated regression, and examined a range of regularization parameter values (see Supplementary Methods section) (Meinshausen and Bühlmann 2010). Stability selection provides a list of selection probabilities for the connections, representing the importance of each in predicting the response. In order to provide information regarding the direction of association between covariates and the response variable, an average coefficient was obtained from all the subsamples computed for each connection.

Visualization and Selection

As a preliminary, an unweighted, sparse consensus macroconnectome of the coherent connections across the group was visualized using Cytoscape (Cline et al. 2007) (Fig. 2), and we performed a descriptive network analysis (see Supplementary Methods section).

Figure 2.

Consensus macroconnectome of preterm-born children. Regions of interest displayed as circles where the size of the circle corresponds to the degree and shading corresponds to the local betweenness centrality. Sup, superior; Inf, inferior; Pos, posterior; Lat, lateral; Med, medial; Ant, anterior.

Figure 2.

Consensus macroconnectome of preterm-born children. Regions of interest displayed as circles where the size of the circle corresponds to the degree and shading corresponds to the local betweenness centrality. Sup, superior; Inf, inferior; Pos, posterior; Lat, lateral; Med, medial; Ant, anterior.

To depict the influence of development or prematurity, we employed the Circos visualization tool (Krzywinski et al. 2009) to display the topological variation between whole-brain structural networks weighted by connection selection probability with either development (age at scan) or prematurity (postconceptional age at birth) as the outcome.

We created a neuroanatomical representation of the effects of development and prematurity on white matter, binarizing every connection in each of the individuals' subject space and transforming them to a common T2-weighted subject space. Each connection was then iteratively merged across the group, thresholded, and binarized to define the common spatial extent of the connections. These were subsequently weighted by selection probability with respect to either response variable. Merging this set of weighted, common-space connections, we generated a probability map in anatomical space in which voxels with stronger connections running through them have higher intensity (Supplementary Methods section).

Finally, we defined the minimal set of connections that were most stable with respect to each response variable. To this end, we visualized the 10 strongest connections above a 0.5 selection probability threshold (i.e., those connections which were selected in the regression model >50% of the time). Ten connections represent a small portion of the total studied. However, in order to gain a practical view of the unique, anatomical pattern of white-matter pathways associated with development or prematurity, this number was felt to be adequate. Using a sequence of registration steps to a common T2-weighted space, population density maps were generated for each of these connections, (Fig. 1, step H).

Results

Figure 2 gives a nonsubjective topological representation of the sparse consensus macroconnectome in this cohort of preterm-born infants, showing high-node degree and betweenness centrality in superior frontal and parietal cortex, thalamus and putamen, and insular cortex. A descriptive, graph theoretical analysis demonstrated typical properties of anatomical brain networks including small-world architecture (Watts and Strogatz 1998) and was shown to be stable with further sparsification (Supplementary Results). Anisotropy data from connections in the consensus macroconnectome were used for all further analyses.

Figure 3 shows the macroconnectome with connections unweighted (Fig. 3A) or weighted by their selection probability in predicting development (Fig. 3B) or prematurity (Fig. 3C). Supplementary Figure 1 provides additional information regarding the types of connection and their differences in the developmental and premature-associated connectome. Supplementary Tables 2 and 3 present the complete list of connections for each response variable and their respective selection probabilities and coefficients. Figure 4 merges the set of connections to create a white-matter selection probability map for each response variable in anatomical space. Figures 5 and 6 depict, for development and prematurity, respectively, a set of population density maps for the most selected connections. Respective seed and target regions for each population density map are illustrated in Supplementary Figures 2 and 3, and their statistical information is presented in Tables 2 and 3.

Table 2

Selected connections associated with age at scan

Connection Region 1 Region 2 Selection probability Average regression coefficient Mean anisotropy (range) 
Superior frontal gyrus, left Superior frontal gyrus, right 0.908 1.443 0.2837 (0.2135–0.3346) 
Posterior temporal lobe, right Parietal lobe, left 0.851 1.113 0.2096 (0.1462–0.2761) 
Putamen, left Lateral orbital gyrus, left 0.749 1.152 0.1805 (0.1331–0.2273) 
Frontal lobe, left, becomes middle frontal gyrus Superior frontal gyrus, left 0.748 1.036 0.1557 (0.134–0.2155) 
Superior frontal gyrus, right Postcentral gyrus, right 0.684 0.932 0.1761 (0.1483–0.2152) 
Cingulate gyrus, anterior part, right Superior frontal gyrus, left 0.660 0.884 0.2757 (0.199–0.344) 
Inferior frontal gyrus, right Superior frontal gyrus, right 0.546 0.814 0.1759 (0.1391–0.215) 
Connection Region 1 Region 2 Selection probability Average regression coefficient Mean anisotropy (range) 
Superior frontal gyrus, left Superior frontal gyrus, right 0.908 1.443 0.2837 (0.2135–0.3346) 
Posterior temporal lobe, right Parietal lobe, left 0.851 1.113 0.2096 (0.1462–0.2761) 
Putamen, left Lateral orbital gyrus, left 0.749 1.152 0.1805 (0.1331–0.2273) 
Frontal lobe, left, becomes middle frontal gyrus Superior frontal gyrus, left 0.748 1.036 0.1557 (0.134–0.2155) 
Superior frontal gyrus, right Postcentral gyrus, right 0.684 0.932 0.1761 (0.1483–0.2152) 
Cingulate gyrus, anterior part, right Superior frontal gyrus, left 0.660 0.884 0.2757 (0.199–0.344) 
Inferior frontal gyrus, right Superior frontal gyrus, right 0.546 0.814 0.1759 (0.1391–0.215) 

Selection probability is calculating by estimating the frequency of a connection being selected across a range of regularization parameters and iterations of the regression. Mean anisotropy represents the average value of integrated anisotropy for a particular tract across the group (see Materials and Methods section for further information).

Table 3

Selected connections associated with postconceptional age at birth

Connection Region 1 Region 2 Selection probability Average regression coefficient Mean anisotropy (range) 
Caudate nucleus, right Globus pallidus, right 0.95 0.072 0.1755 (0.1135–0.2749) 
Cingulate gyrus, anterior part right Subgenual frontal cortex right 0.912 0.046 0.1468 (0.0898–0.2281) 
Occipital lobe, left Superior parietal gyrus, left 0.856 0.095 0.2034 (0.1436–0.2597) 
Putamen, left Thalamus, left 0.854 0.024 0.2202 (0.1659–0.3176) 
Insula, left Frontal lobe, left, becomes middle frontal gyrus 0.832 0.041 0.1716 (0.1209–0.2113) 
Parietal lobe, right Superior parietal gyrus, left 0.817 0.060 0.3188 (0.2567–0.4076) 
Medial and inferior temporal gyri, right Posterior temporal lobe, right 0.779 0.043 0.1978 (0.1234–0.2733) 
Superior frontal gyrus, left Superior frontal gyrus, right 0.773 0.067 0.2837 (0.2135–0.3346) 
Frontal lobe, left, becomes middle frontal gyrus Inferior frontal gyrus, left 0.772 0.036 0.1589 (0.1269–0.2131) 
Frontal lobe, right, becomes middle frontal gyrus Inferior frontal gyrus, right 0.77 0.013 0.1630 (0.1373–0.2027) 
Connection Region 1 Region 2 Selection probability Average regression coefficient Mean anisotropy (range) 
Caudate nucleus, right Globus pallidus, right 0.95 0.072 0.1755 (0.1135–0.2749) 
Cingulate gyrus, anterior part right Subgenual frontal cortex right 0.912 0.046 0.1468 (0.0898–0.2281) 
Occipital lobe, left Superior parietal gyrus, left 0.856 0.095 0.2034 (0.1436–0.2597) 
Putamen, left Thalamus, left 0.854 0.024 0.2202 (0.1659–0.3176) 
Insula, left Frontal lobe, left, becomes middle frontal gyrus 0.832 0.041 0.1716 (0.1209–0.2113) 
Parietal lobe, right Superior parietal gyrus, left 0.817 0.060 0.3188 (0.2567–0.4076) 
Medial and inferior temporal gyri, right Posterior temporal lobe, right 0.779 0.043 0.1978 (0.1234–0.2733) 
Superior frontal gyrus, left Superior frontal gyrus, right 0.773 0.067 0.2837 (0.2135–0.3346) 
Frontal lobe, left, becomes middle frontal gyrus Inferior frontal gyrus, left 0.772 0.036 0.1589 (0.1269–0.2131) 
Frontal lobe, right, becomes middle frontal gyrus Inferior frontal gyrus, right 0.77 0.013 0.1630 (0.1373–0.2027) 
Figure 3.

Connectogram rendering of group structural connectivity with connections unweighted (A) or weighted by their probability of selection with respect to corrected age at scan (B) or postconceptional age at birth (C). Increased thickness and coloration of lines (legend) indicates higher selection probability. Regions of interest (ROIs) are divided in blocks according to lobe (red: frontal, cyan: insula, yellow: limbic, purple: temporal, green: parietal, blue: occipital, gray: subcortical structures) and shown in relative anatomical location (anterior and posterior correspond to the top and bottom of each connectogram). A full list of abbreviations for individual ROIs are provided in the Supplementary Information.

Figure 3.

Connectogram rendering of group structural connectivity with connections unweighted (A) or weighted by their probability of selection with respect to corrected age at scan (B) or postconceptional age at birth (C). Increased thickness and coloration of lines (legend) indicates higher selection probability. Regions of interest (ROIs) are divided in blocks according to lobe (red: frontal, cyan: insula, yellow: limbic, purple: temporal, green: parietal, blue: occipital, gray: subcortical structures) and shown in relative anatomical location (anterior and posterior correspond to the top and bottom of each connectogram). A full list of abbreviations for individual ROIs are provided in the Supplementary Information.

Figure 4.

Merged connectivity map showing white-matter regions through which run tracts associated with age at scan (A) and postconceptional age at birth (B). The color bar indicates level of voxelwise selection probability: White-matter regions with the highest intensity have the highest probability of selection following regression. In (A), these include anterior aspects of the centrum semiovale; corona radiata and corpus callosum and involve the medial and lateral frontal lobes on both sides. In (B), these regions also include posterior and superior aspects of the corona radiata and sagittal striatum and particularly involve the parietal and occipital lobes bilaterally.

Figure 4.

Merged connectivity map showing white-matter regions through which run tracts associated with age at scan (A) and postconceptional age at birth (B). The color bar indicates level of voxelwise selection probability: White-matter regions with the highest intensity have the highest probability of selection following regression. In (A), these include anterior aspects of the centrum semiovale; corona radiata and corpus callosum and involve the medial and lateral frontal lobes on both sides. In (B), these regions also include posterior and superior aspects of the corona radiata and sagittal striatum and particularly involve the parietal and occipital lobes bilaterally.

Figure 5.

Population density maps for a subset of age-associated connections with the highest selection probability. Connections are displayed in axial, coronal, and sagittal sections (left to right) with putative white-matter tracts involved in each. The color bar indicates the proportion of the group present in each voxel of a connection lighter areas reveal parts of the connection with the most consistent anatomy across the group. In the description, bold white-matter tracts represent known direct anatomical fibers between seed and target regions (see Supplementary Fig. 2). Tracts in regular font represent fibers, which are indirect and/or may be encompassed as part of each probabilistic connection.

Figure 5.

Population density maps for a subset of age-associated connections with the highest selection probability. Connections are displayed in axial, coronal, and sagittal sections (left to right) with putative white-matter tracts involved in each. The color bar indicates the proportion of the group present in each voxel of a connection lighter areas reveal parts of the connection with the most consistent anatomy across the group. In the description, bold white-matter tracts represent known direct anatomical fibers between seed and target regions (see Supplementary Fig. 2). Tracts in regular font represent fibers, which are indirect and/or may be encompassed as part of each probabilistic connection.

Figure 6.

Population density maps for a subset of prematurity-associated connections with the highest selection probability. Connections are displayed in axial, coronal, and sagittal sections (left to right) with putative white-matter tracts involved in each. The color bar indicates the proportion of the group present in each voxel of a connection: Seed and target regions are shown in Supplementary Figure 3. Fiber populations in this subset include commissural (F and H), projection (A and D), and associative (E, G, I, and J) white-matter connections.

Figure 6.

Population density maps for a subset of prematurity-associated connections with the highest selection probability. Connections are displayed in axial, coronal, and sagittal sections (left to right) with putative white-matter tracts involved in each. The color bar indicates the proportion of the group present in each voxel of a connection: Seed and target regions are shown in Supplementary Figure 3. Fiber populations in this subset include commissural (F and H), projection (A and D), and associative (E, G, I, and J) white-matter connections.

Postnatal Age is Associated with Increasing Frontal Connectivity

Figure 3B weights topographical connections by selection probability using corrected age at scan as the response variable. Although 131 of a possible 249 connections were selected as predicting age at scan, the majority of these connections (>86%) had a selection probability <0.2 and close to zero (Fig. 3B, Supplementary Table 2).

Connections with the highest level of association connected frontotemporal areas (Fig. 3B) and encompassed white-matter structures, including the anterior centrum semiovale, corona radiata, and the genu of the corpus callosum (Fig. 4A). These connections mainly coupled regions in the frontal lobe: one connection paired homologous regions between hemispheres (connection A, Fig. 5 and Supplementary Fig. 3); while one linked temporal and parietal cortices (connection B). The strength of connection was positively associated with development in all cases (Table 2). Connections involving commissural tracts (A, F) had on average, a greater mean anisotropy than connections predominated by associative tracts (B, D, E, G).

More Severe Prematurity is Associated with Decreasing Connectivity in Cortex and Subcortex

The influence of prematurity on white-matter connections was widespread and involved bilateral cortical and subcortical structures (Fig. 3C, 4). All of the connections in the sparse, consensus macroconnectome were selected as predicting postconceptional age at birth and, in contrast to the connections associated with development, approximately half of these had a selection probability >0.2 (Fig 3C, Supplementary Table 3).

The ten connections with the highest selection probability were retained for further visualization and analysis (Fig. 6 and Supplementary Fig. 3). Identified connections involved: frontal, parietal, temporal, and occipital cortices; subcortical structures (connections A and D, Fig. 4); and homologous (H) or symmetrical connections between hemispheres (I, J). Selected white-matter connections included bilateral commissural, projection, and associative tracts (Fig. 6). In all connections, higher connection strength correlated with greater postconceptional age at birth (Table 3). Connections involving commissural tracts (B, F, H) had on average, a greater mean anisotropy than connections involving projection fiber bundles (A, D), which in turn had larger anisotropy values compared with connections predominated by associative tracts (E, G, I, J).

In order to account for the 7 subjects with serial imaging, further analyses were performed after removing these individuals' earlier or later scan data. The results for all repeat analyses were comparable with few changes in the connections selected as maximally associated with development or prematurity (see Supplementary Results).

Discussion

This article presents a data-driven, nonsubjective analysis of whole-brain structural connectivity in the developing brain using diffusion MRI and provides a global, unrestricted survey of the macroscopic neural architecture without a priori hypotheses. It identified connections related to development, showing as anticipated that connections in frontal regions increased with age. It also detected widespread reductions in connectivity in all cortical lobes and several subcortical structures bilaterally associated with preterm birth.

This approach offers certain advantages. Use of the diffusive exchange between voxels integrated across the whole tract more closely approximates neural connectivity than commonly used scalar measures such as FA, which better reflects local microstructural integrity. This provides sensitivity to connective changes and has been previously used in infants (Ball et al. 2012) and adults (Robinson et al. 2010). The method identifies the cortical and subcortical targets for each connection, and it is not restricted to the examination of subjectively defined white-matter tracts or the center of major white-matter fiber bundles. Penalized regression and stability selection provided an objective approach to feature selection in the connectivity matrix, and useful results are obtained using a relatively small sample of subjects. The current implementation of an 83-region connectivity matrix is relatively coarse and detail is lost due to the spatial averaging of relatively large seed parcellations; however, intersubject correlation at this scale has been shown to be higher (Cammoun et al. 2012). The approach is scalable and will allow the iterative discovery of the macroconnectome in the developing brain with increasing precision.

Early postnatal WM development is typically characterized by processes of axonal growth, myelination, and synaptogenesis (Yakovlev and Lecours 1967; Haynes et al. 2005; Nelson and Luciana 2007). From mid-gestation, axons undergo rapid elongation and maturation, and by late infancy are relatively complete (Haynes et al. 2005). Neuroaxonal development is extended in the associative frontal cortex and appears to be related to the persistence and reorganization of the subplate layer (Kostović et al. 2012). Dendritic arborization and synaptogenesis are also protracted in the frontal cortex (Huttenlocher and Dabholkar 1997; Webb et al. 2001; Petanjek et al. 2011) and coincide with changes in cerebral metabolic activity (Chugani et al. 1987). Histological evidence suggests that myelination typically occurs in a caudal to rostral, proximal to distal, and central to peripheral pattern, with sensory pathways myelinating before motor pathways and frontotemporal regions being the last to mature (Yakovlev and Lecours 1967; Kinney et al. 1988). The findings of conventional MRI studies correspond with this sequence and indicate that myelination reaches an adult pattern of appearance between the second and third postnatal year (Barkovich 2000, 2005). Our whole-brain, diffusion tractography results are comparable to both histological and imaging findings, with connections that were stronger with increasing age predominantly identified in frontal structures. DTI data have suggested analogous changes in local microstructural integrity in relevant white-matter regions despite methodological differences. These regions include the anterior corona radiata or frontal white matter (Schneider et al. 2004; Hermoye et al. 2006; Gao et al. 2009; Löbel et al. 2009), the genu of the corpus callosum (Gao et al. 2009; Löbel et al. 2009), the internal capsule (McGraw et al. 2002; Schneider et al. 2004; Hermoye et al. 2006; Gao et al. 2009), and the superior longitudinal fasciculus (Zhang et al. 2007). A recent study of subjectively selected white-matter pathways, including the genu of the corpus callosum, anterior limb of the internal capsule, and arcuate fasciculi, showed significant increases in FA from birth up to 2 years in a large cohort of normal infants (Geng et al. 2012). The increasing integrity of frontotemporal white-matter connections in this period of development is likely to parallel neurocognitive improvements, including specifically, the growing use of language (Herschkowitz 2000).

In contrast, the effect of prematurity on structural connectivity involved all major brain regions. Prematurity-associated connections were widespread, bilateral, and affected all cortical lobes and several subcortical structures, including projection, commissural and association fibers. Previous region-of-interest studies of premature infants imaged at term-equivalent age, showed reduced FA in the sagittal striatum, frontal lobe white matter, genu and splenium of the corpus callosum, and centrum semiovale (Anjari et al. 2007; Rose et al. 2008; Hasegawa et al. 2011; Jo et al. 2012) with further anisotropy reductions in the posterior limb of the internal capsule, external capsule, and posterior corpus callosum in very premature groups (Anjari et al. 2007; Rose et al. 2008). Tractography-based approaches were consistent with the above findings (Dudink et al. 2008; Hasegawa et al. 2011) and detailed analysis also demonstrated disruption to thalamocortical pathways following premature birth (Ball et al. 2012). Disturbances in WM microstructure appeared to persist from childhood (Yung et al. 2007; Jo et al. 2012) to adolescence and adulthood, with wider reductions in FA across the whole brain (Yung et al. 2007) and in a range of additional pathways including the corona radiata, uncinate fasciculus, superior and inferior longitudinal fasciculi, superior and inferior fronto-occipital fasciculi, corticospinal tract, and cingulum bundle (Vangberg et al. 2006; Constable et al. 2008; Eikenes et al. 2011; Mullen et al. 2011). These areas of reduced microstructural integrity largely approximated the regions through which prematurity-associated connections ran. The specific reductions in connectivity strength attributable to increasing prematurity demonstrated in this work are thus consistent with previous measures of local tissue integrity. The sensitivity of our approach and imaging during early childhood together with the wide range of age at birth in our dataset may account for certain differences in our results. As we have excluded infants with macrostructural pathology, these results are likely to be due to the effects of premature extrauterine life and more specifically, the diffuse white-matter abnormalities common in this population (Volpe 2003). The pathophysiology of this injury probably involves destructive influences, such as hypoxia–ischemia, hyperoxia, and infection and/or inflammation, concentrating on several developmentally susceptible populations of cells, and is part of a spectrum of developmental abnormality that includes reduced gray- and white-matter volumes, reduced cell density, and hypomyelination (Volpe 2009). These abnormalities persist and would account for the reduced connectivity detected in this study.

A limitation in our approach is the substantial amount of data discarded during the process of mapping group whole-brain connectivity. In order to make robust groupwise comparisons, we used only consistent connections that were detected in all subjects. Given the significant heterogeneity in the brain structure of the groups studied, expected data loss would be great. High fidelity in registration and segmentation steps minimized this problem, but of a possible 3403 connections, only ∼250 were retained across the group. Our results thus represent a robust but coarse summary of structural connectivity at the cost of probable loss of subgroup effects. In spite of this, the preterm macroconnectome maintains a typical brain network architecture including small-world topology, and differences in connectivity related either to development or prematurity are maintained over increasing sparsity (Fig. 2, Supplementary Fig. 1, Supplementary Results). In this study, we model the influence of prematurity on an infant macroconnectome. This model is limited in that it neither accounts for gray-matter injury nor the effect of macroscopic white-matter injury (since these subjects were excluded). Both types of injury may independently lead to impaired structural and functional development (Ment et al. 2009; Volpe 2009; de Kieviet et al. 2012). We also acknowledge that we did not account for the effects of intrauterine growth restriction (IUGR) and extrauterine factors such as respiratory comorbidity. Respiratory disease has shown a propensity to affect white-matter microstructure in immature neonates, independent of prematurity (Anjari et al. 2009; Ball et al. 2010), and IUGR has been associated with global and local white-matter changes in infants (Batalle et al. 2012). In future work, we hope to extend our approach to subjects with macroscopic pathology and account for the above confounders.

In conclusion, this study presents a scalable, nonsubjective, data-driven method, which detected contrasting effects of development and prematurity in the sparse infant macroconnectome. The approach can be used to evaluate connectivity in ever-increasing detail and undertake iterative discovery of the macroconnectome with increasing precision.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.

Funding

This work was supported by: the Medical Research Council Clinical Sciences Centre (Doctoral Studentship to AP); the NIHR Imperial College Comprehensive Biomedical Research Centre; and NIHR Comprehensive Biomedical Research Centre at Guy's and St Thomas' NHS Foundation Trust in partnership with King's College London and King's College Hospital NHS Foundation Trust.

Notes

The authors thank the families who took part in the study and our colleagues in the Neonatal Intensive Care Unit at Queens Charlotte's and Chelsea Hospital. Conflict of Interest: None declared.

References

Anjari
M
Counsell
SJ
Srinivasan
L
Allsop
JM
Hajnal
JV
Rutherford
MA
Edwards
AD
The association of lung disease with cerebral white matter abnormalities in preterm infants
Pediatrics
 , 
2009
, vol. 
124
 (pg. 
268
-
276
)
Anjari
M
Srinivasan
L
Allsop
JM
Hajnal
JV
Rutherford
MA
Edwards
AD
Counsell
SJ
Diffusion tensor imaging with tract-based spatial statistics reveals local white matter abnormalities in preterm infants
NeuroImage
 , 
2007
, vol. 
35
 (pg. 
1021
-
1027
)
Ball
G
Boardman
JP
Aljabar
P
Pandit
A
Arichi
T
Merchant
N
Rueckert
D
Edwards
AD
Counsell
SJ
The influence of preterm birth on the developing thalamocortical connectome
Cortex
 , 
2012
 
Ball
G
Counsell
SJ
Anjari
M
Merchant
N
Arichi
T
Doria
V
Rutherford
MA
Edwards
AD
Rueckert
D
Boardman
JP
An optimised tract-based spatial statistics protocol for neonates: applications to prematurity and chronic lung disease
NeuroImage
 , 
2010
, vol. 
53
 (pg. 
94
-
102
)
Barkovich
AJ
Concepts of myelin and myelination in neuroradiology
AJNR Am J Neuroradiol
 , 
2000
, vol. 
21
 (pg. 
1099
-
1109
)
Barkovich
AJ
Magnetic resonance techniques in the assessment of myelin and myelination
J Inherit Metab Dis
 , 
2005
, vol. 
28
 (pg. 
311
-
343
)
Batalle
D
Eixarch
E
Figueras
F
Muñoz-Moreno
E
Bargallo
N
Illa
M
Acosta-Rojas
R
Amat-Roldan
I
Gratacos
E
Altered small-world topology of structural brain networks in infants with intrauterine growth restriction and its association with later neurodevelopmental outcome
NeuroImage
 , 
2012
, vol. 
60
 (pg. 
1352
-
1366
)
Behrens
TEJ
Woolrich
MW
Jenkinson
M
Johansen-Berg
H
Nunes
RG
Clare
S
Matthews
PM
Brady
JM
Smith
SM
Characterization and propagation of uncertainty in diffusion-weighted MR imaging
Magn Reson Med
 , 
2003
, vol. 
50
 (pg. 
1077
-
1088
)
Bhutta
AT
Cleves
MA
Casey
PH
Cradock
MM
Anand
KJS
Cognitive and behavioral outcomes of school-aged children who were born preterm: a meta-analysis
JAMA
 , 
2002
, vol. 
288
 (pg. 
728
-
737
)
Brody
BA
Kinney
HC
Kloman
AS
Gilles
FH
Sequence of central nervous system myelination in human infancy. I. An autopsy study of myelination
J Neuropathol Exp Neurol
 , 
1987
, vol. 
46
 (pg. 
283
-
301
)
Cammoun
L
Gigandet
X
Meskaldji
D
Thiran
JP
Sporns
O
Do
KQ
Maeder
P
Meuli
R
Hagmann
P
Mapping the human connectome at multiple scales with diffusion spectrum MRI
J Neurosci Methods
 , 
2012
, vol. 
203
 (pg. 
386
-
397
)
Choi
YY
Shamosh
NA
Cho
SH
DeYoung
CG
Lee
MJ
Lee
JM
Kim
SI
Cho
ZH
Kim
K
Gray
JR
, et al.  . 
Multiple bases of human intelligence revealed by cortical thickness and neural activation
J Neurosci
 , 
2008
, vol. 
28
 (pg. 
10323
-
10329
)
Chugani
HT
Phelps
ME
Mazziotta
JC
Positron emission tomography study of human brain functional development
Ann Neurol
 , 
1987
, vol. 
22
 (pg. 
487
-
497
)
Cline
MS
Smoot
M
Cerami
E
Kuchinsky
A
Landys
N
Workman
C
Christmas
R
Avila-Campilo
I
Creech
M
Gross
B
, et al.  . 
Integration of biological networks and gene expression data using Cytoscape
Nat Protoc
 , 
2007
, vol. 
2
 (pg. 
2366
-
2382
)
Constable
RT
Ment
LR
Vohr
BR
Kesler
SR
Fulbright
RK
Lacadie
C
Delancy
S
Katz
KH
Schneider
KC
Schafer
RJ
, et al.  . 
Prematurely born children demonstrate white matter microstructural differences at 12 years of age, relative to term control subjects: an investigation of group and gender effects
Pediatrics
 , 
2008
, vol. 
121
 (pg. 
306
-
316
)
Counsell
SJ
Edwards
AD
Chew
ATM
Anjari
M
Dyet
LE
Srinivasan
L
Boardman
JP
Allsop
JM
Hajnal
JV
Rutherford
MA
, et al.  . 
Specific relations between neurodevelopmental abilities and white matter microstructure in children born preterm
Brain
 , 
2008
, vol. 
131
 (pg. 
3201
-
3208
)
de Kieviet
JF
Zoetebier
L
van Elburg
RM
Vermeulen
RJ
Oosterlaan
J
Brain development of very preterm and very low-birthweight children in childhood and adolescence: a meta-analysis
Develop Med Child Neurol
 , 
2012
, vol. 
54
 (pg. 
313
-
323
)
Dudink
J
Kerr
JL
Paterson
K
Counsell
SJ
Connecting the developing preterm brain
Early Hum Dev
 , 
2008
, vol. 
84
 (pg. 
777
-
782
)
Eikenes
L
Løhaugen
GC
Brubakk
A-M
Skranes
J
Håberg
AK
Young adults born preterm with very low birth weight demonstrate widespread white matter alterations on brain DTI
NeuroImage
 , 
2011
, vol. 
54
 (pg. 
1774
-
1785
)
Gao
W
Lin
W
Chen
Y
Gerig
G
Smith
JK
Jewells
V
Gilmore
JH
Temporal and spatial development of axonal maturation and myelination of white matter in the developing brain
AJNR Am J Neuroradiol
 , 
2009
, vol. 
30
 (pg. 
290
-
296
)
Geng
X
Gouttard
S
Sharma
A
Gu
H
Styner
M
Lin
W
Gerig
G
Gilmore
JH
Quantitative tract-based white matter development from birth to age 2 years
NeuroImage
 , 
2012
, vol. 
61
 (pg. 
542
-
557
)
Gousias
IS
Rueckert
D
Heckemann
RA
Dyet
LE
Boardman
JP
Edwards
AD
Hammers
A
Automatic segmentation of brain MRIs of 2-year-olds into 83 regions of interest
NeuroImage
 , 
2008
, vol. 
40
 (pg. 
672
-
684
)
Hasegawa
T
Yamada
K
Morimoto
M
Morioka
S
Tozawa
T
Isoda
K
Murakami
A
Chiyonobu
T
Tokuda
S
Nishimura
A
, et al.  . 
Development of corpus callosum in preterm infants is affected by the prematurity: in vivo assessment of diffusion tensor imaging at term-equivalent age
Pediatr Res
 , 
2011
, vol. 
69
 (pg. 
249
-
254
)
Haynes
RL
Borenstein
NS
Desilva
TM
Folkerth
RD
Liu
LG
Volpe
JJ
Kinney
HC
Axonal development in the cerebral white matter of the human fetus and infant
J Comp Neurol
 , 
2005
, vol. 
484
 (pg. 
156
-
167
)
Hermoye
L
Saint-Martin
C
Cosnard
G
Lee
S
Pediatric diffusion tensor imaging: Normal database and observation of the white matter maturation in early childhood
NeuroImage
 , 
2006
, vol. 
29
 (pg. 
493
-
504
)
Herschkowitz
N
Neurological bases of behavioral development in infancy
Brain Dev
 , 
2000
, vol. 
22
 (pg. 
411
-
416
)
Huttenlocher
PR
Dabholkar
AS
Regional differences in synaptogenesis in human cerebral cortex
J Comp Neurol
 , 
1997
, vol. 
387
 (pg. 
167
-
178
)
Jo
HM
Cho
HK
Jang
SH
Yeo
SS
Lee
E
Kim
HS
Son
SM
A comparison of microstructural maturational changes of the corpus callosum in preterm and full-term children: a diffusion tensor imaging study
Neuroradiology
 , 
2012
 
Johnson
S
Cognitive and behavioural outcomes following very preterm birth
Semin Fetal Neonatal Med
 , 
2007
, vol. 
12
 (pg. 
363
-
373
)
Kerr-Wilson
CO
Mackay
DF
Smith
GCS
Pell
JP
Meta-analysis of the association between preterm delivery and intelligence
J Public Health
 , 
2011
, vol. 
34
 (pg. 
209
-
216
)
Khwaja
O
Volpe
JJ
Pathogenesis of cerebral white matter injury of prematurity
Arch Dis Child Fetal Neonatal Ed
 , 
2007
, vol. 
93
 (pg. 
F153
-
F161
)
Kinney
HC
Brody
BA
Kloman
AS
Gilles
FH
Sequence of central nervous system myelination in human infancy. II. Patterns of myelination in autopsied infants
J Neuropathol Exp Neurol
 , 
1988
, vol. 
47
 pg. 
217
 
Kostović
I
Jovanov-Milosević
N
Radoš
M
Sedmak
G
Benjak
V
Kostović-Srzentić
M
Vasung
L
Culjat
M
Radoš
M
Hüppi
P
, et al.  . 
Perinatal and early postnatal reorganization of the subplate and related cellular compartments in the human cerebral wall as revealed by histological and MRI approaches
Brain Struct Funct
 , 
2012
 
Krzywinski
M
Schein
J
Birol
I
Connors
J
Gascoyne
R
Horsman
D
Jones
SJ
Marra
MA
Circos: an information aesthetic for comparative genomics
Genome Res
 , 
2009
, vol. 
19
 (pg. 
1639
-
1645
)
Löbel
U
Sedlacik
J
Güllmar
D
Kaiser
WA
Reichenbach
JR
Mentzel
H-J
Diffusion tensor imaging: the normal evolution of ADC, RA, FA, and eigenvalues studied in multiple anatomical regions of the brain
Neuroradiology
 , 
2009
, vol. 
51
 (pg. 
253
-
263
)
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 Am J Roentgenol
 , 
2002
, vol. 
179
 (pg. 
1515
-
1522
)
Meinshausen
N
Bühlmann
P
Stability selection
J R Stat Soc Series B Stat Methodol
 , 
2010
, vol. 
72
 (pg. 
417
-
473
)
Ment
LR
Hirtz
D
Hüppi
PS
Imaging biomarkers of outcome in the developing preterm brain
Lancet Neurol
 , 
2009
, vol. 
8
 (pg. 
1042
-
1055
)
Mesulam
M
Brain, mind, and the evolution of connectivity
Brain Cognit
 , 
2000
, vol. 
42
 (pg. 
4
-
6
)
Moster
D
Lie
RT
Markestad
T
Long-term medical and social consequences of preterm birth
N Engl J Med
 , 
2008
, vol. 
359
 (pg. 
262
-
273
)
Mullen
KM
Vohr
BR
Katz
KH
Schneider
KC
Lacadie
C
Hampson
M
Makuch
RW
Reiss
AL
Constable
RT
Ment
LR
Preterm birth results in alterations in neural connectivity at age 16 years
NeuroImage
 , 
2011
, vol. 
54
 (pg. 
2563
-
2570
)
Nelson
CA
Luciana
M
Handbook of Developmental Cognitive Neuroscience
 , 
2007
(pg. 
1
-
702
)
Petanjek
Z
Judas
M
Šimic
G
Rasin
MR
Uylings
HBM
Rakic
P
Kostović
I
Extraordinary neoteny of synaptic spines in the human prefrontal cortex
Proc Natl Acad Sci USA
 , 
2011
, vol. 
108
 (pg. 
13281
-
13286
)
Robinson
EC
Hammers
A
Ericsson
A
Edwards
AD
Rueckert
D
Identifying population differences in whole-brain structural networks: A machine learning approach
NeuroImage
 , 
2010
, vol. 
50
 (pg. 
910
-
919
)
Rose
SE
Hatzigeorgiou
X
Strudwick
MW
Durbridge
G
Davies
PSW
Colditz
PB
Altered white matter diffusion anisotropy in normal and preterm infants at term-equivalent age
Magn Reson Med
 , 
2008
, vol. 
60
 (pg. 
761
-
767
)
Rueckert
D
Sonoda
LI
Hayes
C
Hill
DL
Leach
MO
Hawkes
DJ
Nonrigid registration using free-form deformations: application to breast MR images
IEEE Trans Med Imaging
 , 
1999
, vol. 
18
 (pg. 
712
-
721
)
Saigal
S
Doyle
LW
An overview of mortality and sequelae of preterm birth from infancy to adulthood
Lancet
 , 
2008
, vol. 
371
 (pg. 
261
-
269
)
Schneider
JFL
Il'yasov
KA
Hennig
J
Martin
E
Fast quantitative diffusion-tensor imaging of cerebral white matter from the neonatal period to adolescence
Neuroradiology
 , 
2004
, vol. 
46
 (pg. 
258
-
266
)
Smith
SM
Fast robust automated brain extraction
Hum Brain Mapp
 , 
2002
, vol. 
17
 (pg. 
143
-
155
)
Smith
SM
Jenkinson
M
Woolrich
MW
Beckmann
CF
Behrens
TEJ
Johansen-Berg
H
Bannister
PR
De Luca
M
Drobnjak
I
Flitney
DE
, et al.  . 
Advances in functional and structural MR image analysis and implementation as FSL
NeuroImage
 , 
2004
, vol. 
23
 (pg. 
S208
-
S219
)
Tamnes
CK
Østby
Y
Walhovd
KB
Westlye
LT
Due-Tønnessen
P
Fjell
AM
Intellectual abilities and white matter microstructure in development: A diffusion tensor imaging study
Hum Brain Mapp
 , 
2010
, vol. 
31
 (pg. 
1609
-
1625
)
Tibshirani
R
Regression Shrinkage and Selection via the Lasso
J R Stat Soc Series B Stat Methodol
 , 
1996
, vol. 
58
 (pg. 
267
-
288
)
Vangberg
TR
Skranes
J
Dale
AM
Martinussen
M
Brubakk
A-M
Haraldseth
O
Changes in white matter diffusion anisotropy in adolescents born prematurely
NeuroImage
 , 
2006
, vol. 
32
 (pg. 
1538
-
1548
)
Volpe
JJ
Brain injury in premature infants: a complex amalgam of destructive and developmental disturbances
Lancet Neurol
 , 
2009
, vol. 
8
 (pg. 
110
-
124
)
Volpe
JJ
Cerebral white matter injury of the premature infant-more common than you think
Pediatrics
 , 
2003
, vol. 
112
 (pg. 
176
-
180
)
Watts
DJ
Strogatz
SH
Collective dynamics of “small-world” networks
Nature
 , 
1998
, vol. 
393
 (pg. 
440
-
442
)
Webb
SJ
Monk
CS
Nelson
CA
Mechanisms of postnatal neurobiological development: implications for human development
Dev Neuropsychol
 , 
2001
, vol. 
19
 (pg. 
147
-
171
)
Woodward
LJ
Anderson
PJ
Austin
NC
Howard
K
Inder
TE
Neonatal MRI to predict neurodevelopmental outcomes in preterm infants
N Engl J Med
 , 
2006
, vol. 
355
 (pg. 
685
-
694
)
Yakovlev
PI
Lecours
AR
Minkowski
A
The myelogenetic cycles of regional maturation of the brain
Regional development of the brain in early life
 , 
1967
Oxford, UK
Blackwell Scientific
(pg. 
3
-
70
)
Yung
A
Poon
G
Qiu
D
Chu
J
Lam
B
Leung
C
Goh
W
Khong
P
White matter volume and anisotropy in preterm children: a pilot study of neurocognitive correlates
Pediatr Res
 , 
2007
, vol. 
61
 (pg. 
732
-
736
)
Zhang
J
Evans
A
Hermoye
L
Lee
S-K
Wakana
S
Zhang
W
Donohue
P
Miller
MI
Huang
H
Wang
X
, et al.  . 
Evidence of slow maturation of the superior longitudinal fasciculus in early childhood by diffusion tensor imaging
NeuroImage
 , 
2007
, vol. 
38
 (pg. 
239
-
247
)