Skip to main content
NCBI home page
Human Brain Mapping logoLink to Human Brain Mapping
. 2017 Apr 21;38(7):3603–3614. doi: 10.1002/hbm.23614

The structural connectome of children with traumatic brain injury

Marsh Königs 1,2,✉, L W Ernest van Heurn 3, Roel Bakx 3, R Jeroen Vermeulen 4,5, J Carel Goslings 6, Bwee Tien Poll‐The 7, Marleen van der Wees 8, Coriene E Catsman‐Berrevoets 9, Jaap Oosterlaan 1,2,10, Petra J W Pouwels 11,12
PMCID: PMC6866988  PMID: 28429381

Abstract

This study aimed to investigate the impact of mild to severe pediatric TBI on the structural connectome. Children aged 8–14 years with trauma control (TC) injury (n = 27) were compared to children with mild TBI and risk factors for complicated TBI (mildRF+, n = 20) or moderate/severe TBI (n = 16) at 2.8 years post‐injury. Probabilistic tractography on diffusion tensor imaging data was used in combination with graph theory to study structural connectivity. Functional outcome was measured using neurocognitive tests and parent and teacher questionnaires for behavioral functioning. The results revealed no evidence for an impact of mildRF+ TBI on the structural connectome. In contrast, the moderate/severe TBI group showed longer characteristic path length (P = 0.022, d = 0.82) than the TC group. Furthermore, longer characteristic path length was related to poorer intelligence and poorer working memory in children with TBI. In conclusion, children have abnormal organization of the structural connectome after moderate/severe TBI, which may be implicated in neurocognitive dysfunction associated with pediatric TBI. These findings should be interpreted in the context of our exploratory analyses, which indicate that the definition and weighting of connectivity (e.g., streamline density, fractional anisotropy) influence the properties of the reconstructed connectome and its sensitivity to the impact and outcome of pediatric TBI. Hum Brain Mapp 38:3603–3614, 2017. © 2017 Wiley Periodicals, Inc.

Keywords: pediatrics, traumatic brain injury, structural connectivity, functional outcome

INTRODUCTION

Traumatic brain injury (TBI) is the leading cause of death and acquired disability in children and adolescents [World Health Organization, 2006]. The essence of the neuropathology associated with TBI is thought to be represented by widespread axonal injury [Sharp et al., 2014], threatening the integrity of brain networks that facilitate efficient relay and integration of information throughout the brain [Park and Friston, 2013]. Ultimately, children with TBI are at risk of neurocognitive impairments [Babikian and Asarnow, 2009], behavior problems [Li and Liu, 2013] and poor academic attainment [Vu et al., 2011].

Diffusion tensor imaging (DTI) studies have revealed widespread abnormalities in white matter integrity (i.e., fractional anisotropy [FA]) after moderate and severe TBI in children, suggesting the presence of cerebral edema in the acute phase after injury and the presence of axonal degeneration and/or demyelination in the chronic phase [for a meta‐analysis, see Roberts et al., 2014]. Even after mild TBI, acute and subacute white matter abnormalities have been shown in children [Babcock et al., 2015; Van Beek et al., 2015; Yuan et al., 2015]. Taken together, these findings indicate that white matter is implicated in the neuropathology of TBI across the full range of injury severity.

In recent years, white matter tractography and graph theory have shown to be promising methods to study the impact of neurological disorders on the organization of structural connectivity [for a review, see Fornito et al., 2015]. Among a handful of studies that have used graph theory to assess structural connectivity after TBI, the majority has investigated adults with moderate/severe TBI. These studies have used a varying set of network parameters to show that TBI distorts the small‐world topology of the connectome, that is, the equilibrium between structural segregation (i.e., local clustering of connections) and structural integration [i.e., interconnectivity between clusters, Rubinov and Sporns, 2010] toward a lower degree of integration [Caeyenberghs et al., 2012a, 2012b, 2014; Kim et al., 2014]. In adults, complex network measures have shown promising associations with postconcussional symptoms [Dall'Acqua et al., 2016], postural control [Caeyenberghs et al., 2012a], neurocognitive functioning [Caeyenberghs et al., 2014; Kim et al., 2014] and ratings of disruptive behavior [Kim et al., 2014], highlighting the prognostic potential of network analysis for the outcome of patients with TBI.

The scarcely available studies on structural connectivity after pediatric TBI appear to be in line with literature on adults with TBI. The findings indicate an increase in small world organization of the structural connectome (i.e., increased segregation and reduced integration) in the acute phase of mild TBI [Yuan et al., 2015] and in the chronic phase of complicated mild to severe TBI [Yuan et al., 2017]. In addition to the evident lack of studies into the structural connectome of pediatric TBI patients, the sparse literature into adult TBI shows great methodological variability that further hampers an integrative interpretation of results across studies. For example, studies differ in TBI severity of the patient sample (i.e., mild, moderate, severe or moderate/severe), the recovery phase of interest (i.e., acute, subacute or chronic phase) and the method chosen to reconstruct white matter tracts (e.g., deterministic vs. probabilistic tractography) [see Qi et al., 2015 for a review]. Another crucial difference between studies is the metric chosen to define the structural network of interest, such as the most commonly used streamline density (SLD, a measure of connectivity probability) [Behrens et al., 2003] or FA (a measure of connectivity integrity) [Fagerholm et al., 2015; Qi et al., 2015]. Differential definitions of the structural network are known to influence network parameters, although it remains unknown how this influences connectome reconstruction [Qi et al., 2015]. Taken together, the current literature indicates that TBI alters structural brain connectivity; while it remains unclear how more severe forms of TBI affect structural connectivity in the developing pediatric brain, and how varying network definitions capture differential aspects of the neuropathology in the structural connectome.

This study aims to elucidate the impact of mild to severe pediatric TBI on the structural connectome in relation to neurocognitive and behavioral outcome in the chronic phase of recovery. Based on the available literature on adult TBI, we expect that pediatric TBI will affect the degree of structural integration in the connectome. Furthermore, we will explore the influence of two commonly used network definitions (SLD and FA) on the reconstruction of the connectome. The results of this study will contribute to our understanding of the impact of TBI on neural connectivity and its implications for functional outcome in children.

METHODS

Participants

This study compared a group of 36 children with TBI to a group of 27 children with trauma control (TC) injury not involving the head, in order to control for pre‐injury risk factors of traumatic injury [Max et al., 1998]. Data were collected as part of a follow‐up on a consecutive cohort that was retrospectively recruited from three university‐affiliated level I trauma centers and three rehabilitation centers in the Netherlands [Königs et al., 2015]. Inclusion criteria were: (1) age 8–14 years at time of follow‐up; (2) proficient in the Dutch language; (3) children in the TBI group were required to have a history of hospital admission with a clinical diagnosis of either: (a) mild TBI (GCS = 15–13, loss of consciousness [LOC] duration ≤ 30 min, post‐traumatic amnesia [PTA] duration ≤ 1 h) with at least one of the following risk factors for complicated TBI (mildRF+ TBI) according to the European Federation of Neurological Societies guidelines on mild TBI: impaired consciousness (GCS = 13–14), focal neurological deficits, persistent vomiting (≥ 3 episodes), post‐injury epileptic seizure, progressive headache and abnormal head CT‐scan [Vos and Battistin, 2002]; or (b) moderate/severe TBI (GCS = 12–3, LOC duration > 30 min, PTA duration > 1 h [Teasdale and Jennett, 1976]); and (4) children in the TC group were required to have a history of hospital admission for traumatic injuries below the clavicles [American College of Surgeons, 2004]. Exclusion criteria were: (1) previous TBI; (2) visual disorder interfering with neurocognitive testing; or (3) current neurological condition with known effects on neurocognitive functioning, other than TBI, as documented in medical records and/or reported in a parent‐questionnaire on premorbid functioning.

Background Information

Data on gender, age, socio‐economic status (SES) and diagnosed psychiatric or learning disorders were collected using a parental questionnaire. SES was defined as the average level of parental education ranging from 1 (no education) to 8 (postdoctoral education) [Statistics Netherlands, 2006].

Injury Severity

Diagnosed injuries, the lowest score on the GCS on the day of admission, presence of the described risk factors for complicated mild TBI [Vos and Battistin, 2002], and length of hospital stay were extracted from medical files, as was information on any executed surgical procedures. Based on this information, children were assigned to the TC group, mildRF+ TBI group or moderate/severe TBI group.

MRI Acquisition and Pre‐Processing

MRI was performed at an average 2.8 years post‐injury (SD = 1.1) on a 3 Tesla whole‐body unit (Discovery MR750, GE Healthcare, Milwaukee, Wisconsin) using an 8‐channel head‐coil. Three‐dimensional T1‐weighted images were acquired using a fast spoiled gradient‐echo sequence (176 slices, acquisition matrix 256 × 256, voxel‐size 1 × 1 × 1 mm, TR/TE/TI = 8.2/3.2/450 ms, flip angle 15˚). Furthermore, two‐dimensional echo‐planar diffusion‐tensor images were acquired in 5 volumes without diffusion weighting and 30 volumes with non‐collinear diffusion gradients (b‐value = 750 s/mm2) in 47 slices of 2.5‐mm thickness (axially angulated parallel to the line connecting the pituitary to the fastigium of the fourth ventricle) covering the whole brain (TR/TE = 5,000/74 ms). The acquired in‐plane resolution was 2.5 × 2.5 mm, reconstructed to 1 × 1 mm using interpolation. Parallel imaging was applied with an acceleration factor of 2. All processing of MR images was performed using the Functional MRI of the Brain (FMRIB) Software Library (FSL) version 5.0.8 [Jenkinson et al., 2012]. Pre‐processing of diffusion‐tensor images, including correction for motion and eddy currents is described in the Supporting Information. Subsequent data processing is described below and a visual representation of the data processing pipeline is provided in Figure 1. An explanatory table of study constructs and variables described in this method section is provided in the Supporting Information (Table S1).

Figure 1.

Figure 1

Open in a new tab

Pipeline of data processing for the construction of connectivity matrices. Note. A visual representation of the processing pipeline for the construction of SLD‐FA matrices for the main analysis and SLD‐SLD and FA‐FA for exploratory analyses (see Supporting Information). Also see “Connectivity matrix construction” in the Methods section for an elaborate description of the procedure. AAL = Automated Anatomical Labeling (atlas) [Gong et al., 2009]; FIRST [Patenaude et al., 2011] refers to a model‐based segmentation tool from the Functional MRI of the Brain Software Library (FSL); SLD = streamline density; ROIs = regions of interest; FA = fractional anisotropy. [Color figure can be viewed at http://wileyonlinelibrary.com]

Probabilistic Tractography

We performed probabilistic tractography to infer the structural connectivity between a priori defined regions‐of‐interest (ROI). The Automated Anatomic Labeling (AAL) atlas [Gong et al., 2009] was used to define neocortical ROIs (n = 78) in standard space, while FIRST [Patenaude et al., 2011] was used to define subcortical ROIs (n = 14: thalamus, caudate nucleus, putamen, globus pallidum, hippocampus, amygdala and accumbens nuclei in both hemispheres) in T1‐space. All ROIs (for a complete list, see Table S2 in the Supporting Information) were combined in T1 space and subsequently rimmed to confine tractography from and to the border between grey matter and white matter. For the cortical ROIs, we calculated a rim of two voxels thickness by subtracting an eroded white matter mask from a dilated white matter mask (obtained using FAST) [Zhang et al., 2001]. We subsequently multiplied the resulting rim with the AAL atlas in subject space to obtain rim masks of the cortical ROIs. For subcortical ROIs, rim masks were obtained by the original subcortical ROI masks that were subtracted by two‐voxel eroded versions of the original masks. The resulting cortical and subcortical rim masks (n = 92) were each used as both seeds and targets for probabilistic tractography. BEDPOSTX [Behrens et al., 2003] was used to model multiple fiber orientations per voxel in the diffusion weighted maps, and finally probabilistic tracking was performed using PROBTRACKX2 in DTI space [Behrens et al., 2003]. A total of 500 streamlines were generated from each voxel in each seed ROI (using a curvature threshold of 0.2 and step length of 0.5 mm) and streamlines were terminated if they reached one of the target ROIs (n = 91). Streamlines with interhemispheric crossings outside the corpus callosum and fornix were discarded.

Network Definition

Connectivity matrix

As indicated, we explored the influence of network definition on the reconstruction of the connectome. To this end, we used SLD (describing the probability of a tract connecting two ROIs) [Behrens et al., 2003] and FA (describing the integrity of a tract connecting two ROIs) [Fagerholm et al., 2015] to create three types of connectivity matrices for each subject. Visual representations of these networks are displayed in Figure S1 (see Supporting Information). We used SLD to define connectivity matrices that capture the network with the highest connectivity probability (i.e., the connectivity probability network). These binary SLD‐defined matrices were subsequently weighted for FA in order to account for differences in connectivity integrity within the connectivity probability network (SLD‐FA matrices). We additionally constructed SLD‐defined matrices that were weighted by SLD to account for differences in connectivity probability between brain regions (SLD‐SLD matrices). Likewise, FA‐defined matrices were constructed in order to capture the network with the highest connectivity integrity (i.e., the connectivity integrity network), which were weighted for FA to account for differences in connectivity integrity between brain regions in this network (FA‐FA matrices). The main analyses were based on FA‐SLD matrices as the exploratory analyses (see Supporting Information) revealed that these matrices were most sensitive to the functional outcome of pediatric TBI. A visual representation of the SLD‐FA network is displayed in Figure 2 (as produced with qgraph) [Epskamp et al., 2012].

Figure 2.

Figure 2

Open in a new tab

Visual representation of the reconstructed connectome. Note. Also see “Network definition” in the Methods section for an explanation of the SLD‐FA network. The displayed network is based on the trauma control group average (thresholded at the 20% strongest connections). Line weight and distance between nodes refer to link weight (i.e., weight of the connection between nodes in the network), whereas nodal size refers to the importance of a node in the network, defined by the number of links attached to that node (i.e., degree). Node lateralization is included in the node name (R for right and L for left; see Table S2 in the Supporting Information for an explanation of abbreviated node names). [Color figure can be viewed at http://wileyonlinelibrary.com]

Connectivity matrix construction

SLD of a given tract was defined by the number of streamlines that connect the involved seed‐target pair (ROI X to ROI Y and vice versa), where higher SLD corresponds to higher connectivity probability. SLD was obtained as part of the output from PROBTRACKX2 (see “Probabilistic tractography”). FA of a given tract was defined by mean FA in a set of streamlines connecting the involved seed‐target pair (ROI X to ROI Y and vice versa), where higher FA corresponds to higher connectivity integrity. FA of each tract was derived according to a method described by Squarcina et al. [2012], which has been found to produce accurate and robust results in patients with TBI [Fagerholm et al., 2015]. A white matter tract atlas was constructed based on a subgroup of five subjects from the TC group that was representative for the total TC group (age: M = 10.6, SD = 1.4, gender: 60% males, SES: M = 6.0, SD = 0.9). For these five subjects, spatial SLD distributions connecting each seed‐target pair were registered to MNI‐152 space and subsequently averaged across the subjects. The resulting group‐based spatial SLD distributions were then confined to the 5% voxels with the highest SLD per tract connecting each seed‐target pair and binarized to produce a conservative atlas of white matter tract masks. The mask of each tract was superimposed on the skeletonized FA map of each subject (FA > 0.2, constructed using FSL's Diffusion Toolbox) to calculate mean FA across streamlines connecting each seed‐target pair within the white matter skeleton, which was subsequently used as FA weight.

Global Network Parameters

Graph theory was used to study the properties of the constructed networks. Graph theory is an influential method in neural connectivity analysis that describes the organization of a network according to the distribution of links between nodes. An in‐depth review of graph theory is provided elsewhere [Rubinov and Sporns, 2010]. Graph analysis can be used to describe connectivity at the level of the network using global network parameters. The Brain Connectivity Toolbox [Rubinov and Sporns, 2010] was used to calculate global network parameters.

We used characteristic path length, transitivity, assortativity, and modularity to describe the organization of structural connectivity of the global network. In short, characteristic path length is a measure of integration defined by the average shortest distance between nodes, where distance is defined by the average number of links that connect a node to all other nodes. Transitivity is a measure of local clustering, defined by the ratio of triangles (i.e., three nodes connected in a closed triangle) to triplets (i.e., three nodes connected in an open triangle), where higher transitivity reflects stronger clustering of nodes. Modularity is a measure of clustering at the level of modules, which are subdivisions of the network in groups of nodes with many links to nodes within the group and few links to nodes outside the group. Higher modularity refers to a stronger subdivision of the network into modules. Assortativity is a measure of hierarchy, defined by the correlation between the degree (i.e., number of links attached to a node) of connecting nodes. Consequently, higher (more positive) assortativity coefficients describe a stronger tendency of nodes to connect to other nodes with a similar degree.

Functional Outcome

We used aspects of neurocognitive and behavioral functioning with proven sensitivity for TBI as measures of functional outcome. In a previous study, we already showed that the current mildRF+ TBI and moderate/severe TBI groups had lower intelligence, poorer working memory performance, poorer encoding of information in working memory and more internalizing behavior (e.g., symptoms of anxiety and depression) and externalizing behavior problems (e.g., aggression and delinquent behavior) than the TC group [Königs et al., 2017]. Consequently, we selected these outcome measures to study the relation between network parameters and functional outcome.

Intelligence was measured by a Wechsler Intelligence Scale (WISC)‐III short‐form estimation of age‐standardized full‐scale IQ (FSIQ), involving the Vocabulary and Block Design subtests. This short form has high validity and reliability in estimating full FSIQ [Sattler, 2001]. Working memory was measured using the age‐standardized score on the Digit Span subtest of the WISC‐III [Wechsler, 1991]. The Rey Auditory Verbal Learning Test (RAVLT) [van den Burg and Kingma, 1999] was used to measure encoding in verbal memory using the age‐standardized z‐scores on the direct recall condition. Behavioral functioning was measured using parent and teacher ratings of internalizing problems and externalizing problems, obtained using the Child Behavior Checklist and the Teacher Rating Form, respectively [Verhulst and van der Ende, 2013]. Age‐ and gender‐standardized T‐scores of parents and teacher ratings were averaged to yield composite scores of internalizing and externalizing problems. For clarity reasons, all functional outcome scores were transformed to z‐scores where lower values intuitively correspond to poorer neurocognitive performance/less behavior problems.

Procedure

The current study represents a follow‐up of an earlier investigated sample of children [Königs et al., 2015]. Of all 123 children that were eligible for the current follow‐up (TBI: n = 67; TC: n = 56), 11 were not reached (TBI: n = 8; TC: n = 3) and 36 declined participation (TBI: n = 17; TC: n = 19). Main reasons not to participate were: not interested (TBI: 41%; TC: 32%), objection to MRI (TBI: 0%; TC: 21%), and lack of time (TBI: 24%; TC: 21%). A total of 12 children were excluded from participation due to dental braces incompatible with MRI (TBI: n = 2; TC: n = 6), claustrophobia (TBI: n = 2; TC: n = 1), major brain damage after neurosurgical resection (TBI: n = 1; TC: n = 0), or no show (TBI: n = 1; TC: n = 0). The remaining children in the TBI and TC groups (TBI: n = 36; TC: n = 27) did not differ from their respective cohorts in terms of age, gender and SES (TBI: Ps ≥ 0.28; TC: Ps ≥ 0.07), or GCS score (TBI: P = 0.68). The current follow‐up took place at an average 2.8 years post‐injury, ranging between 0.8 and 6.2 years in the TBI sample.

Written informed consent was provided by parents and children aged >11 years. Trained examiners administered the neurocognitive tests in a fixed order, while parents filled out questionnaires in a waiting room. Subsequently, children were made familiar with the MRI procedure using a simulation scanner before actual MRI scanning was performed in the VU University Medical Center. Neurocognitive testing and MRI scanning were performed on the same day for all participants. The medical ethical committee of the VU University Medical Centre approved this study (NL37226.029.11) and the procedures performed in this study were in accordance with the 1964 Helsinki declaration and its later amendments.

Statistical Analysis

All dependent variables were screened for outliers using box‐plots. To explore group comparability, all groups (TC, mildRF+ TBI and moderate/severe TBI) were compared on demographics, injury severity variables and the prevalence of clinical diagnoses of psychiatric or learning disorders, using ANOVA or chi‐square tests, where appropriate.

To investigate the impact of TBI on network properties, global network parameters (characteristic path length, transitivity, modularity, and assortativity) were derived from the SLD‐FA matrices. As matrix threshold can influence network properties [Qi et al., 2015], global network parameters were calculated for a range of proportional thresholds (range: 5–30% strongest connections with steps of 1%). The area under the curve (AUC) of global network parameters as a function of matrix threshold was used in all analyses assessing network properties. This strategy prevents the production of spurious results that are specific to a certain, arbitrarily chosen, matrix threshold [Zhang et al., 2011]. AUC of each global network parameter was compared between groups using ANOVA with a linear contrast to assess the main effect of group (TC, mildRF+ TBI and moderate/severe TBI) on global network properties.

We replicated earlier reported [Königs et al., 2017] effects of TBI on measures of neurocognitive functioning (FSIQ, digit span, and RALVT encoding) and behavioral functioning (internalizing and externalizing problems) using ANOVA. Subsequently, we investigated the relations between network parameters that showed a main effect of group and measures of neurocognitive and behavioral functioning with a main effect of group. To this end, Pearson correlations were calculated in the sample of children with TBI and children in the TC group separately.

All ANOVAs with significant main effects of group were followed by pair‐wise comparisons between groups using post‐hoc LSD testing. For all statistical analyses, α was set at 0.05 (two‐sided) and effect sizes were expressed in Cohen's d [Cohen, 1988].

RESULTS

Patient Characteristics

Table 1 displays data on demographics, injury‐severity variables and prevalence of psychiatric and learning disorders for all groups. There were no group differences on demographic variables (Ps ≥ 0.17), except for lower SES in the mildRF+ TBI and moderate/severe TBI groups as compared to the TC group (Ps ≤ 0.035). With regard to injury‐related variables, as expected the moderate/severe TBI group had lower GCS scores than the mildRF+ TBI group (P < 0.001), while higher prevalence of neurosurgery and longer hospital stay were observed as compared to both the mildRF+ TBI group and TC group (Ps ≤ 0.009). As a result of our recruitment strategy, the TC group had higher prevalence of extracranial fractures and fracture surgery than both the mildRF+ TBI group (Ps < 0.001) and the moderate/severe TBI group (Ps < 0.001). Last, there were no significant group differences in the prevalence of psychiatric or learning disorders (Ps > 0.06).

Table 1.

Demographic and injury‐related characteristics in TBI and TC groups

Groups
TC Mild RF+ TBI Moderate/Severe TBI Contrasts
n 27 20 16
Demographics
Males, n (%) 12 (44) 13 (65) 10 (58) NS
Age at testing in y 10.2 (1.5) 10.5 (1.8) 9.9 (1.4) NS
SES 6.3 (1.1) 5.5 (1.3) 5.0 (1.1) TC > M, MS
Injury‐related information
Age at injury in y 7.5 (2.2) 7.7 (2.3) 6.9 (1.9) NS
Lowest GCS — 14.5 (0.7) 8.2 (2.8) M > MS
Hospital Stay in d 2.5 (2.0) 3.5 (2.4) 8.1 (7.3) TC, M < MS
Time since injury in y 2.7 (1.0) 2.8 (1.1) 3.0 (1.5) NS
Range 1.0–4.5 0.8–53 1.0–6.2
Extracranial fracture, n (%) 23 (85) 4 (20) 2 (13) TC > M, MS
>1 Extracranial fractures, n (%) 3 (11) 1 (5) 0 (0) NS
Fracture surgery, n (%) 22 (82) 2 (10) 0 (0) TC > M, MS
Neurosurgery, n (%) 0 (0) 0 (0) 4 (25) TC, M < MS
Diagnosed conditions
Psychiatric disorder, n (%) 1 (4) 2 (10) 1 (6) NS
Learning disorder, n (%) 2 (7) 4 (20) 0 (0) NS
Open in a new tab

Note. Data reflect mean (standard deviation), unless otherwise indicated. TC = traumatic control; TBI = traumatic brain injury; RF = risk factor; SES = socio‐economic status; NS = not significant; y = years; d = days; GCS = Glasgow Coma Scale; ADHD = Attention Deficit Hyperactivity Disorder; M = mildRF+ TBI group; MS = moderate/severe TBI group.

Global Network Properties

Figure 3 displays global network parameters as a function of matrix threshold. AUC of each global network parameter was compared between groups in order to assess the impact of TBI on network properties across matrix thresholds (Table 2). This analysis revealed a significant main effect of group on characteristic path length, while analyses on transitivity, modularity or assortativity did not reveal significant effects of group. Post‐hoc group comparisons revealed that the moderate/severe TBI group had longer characteristic path length as compared to the TC group (P = 0.022, d = 0.82), while no other significant group differences were observed (Ps ≥ 0.09). Taken together, these findings indicate that moderate/severe TBI reduces structural integration in the child's connectome.

Figure 3.

Figure 3

Open in a new tab

Global network parameters as a function of matrix threshold. Note. Error bars indicate standard error. Global network parameters were calculated in the SLD‐FA network. TC = trauma control; TBI = traumatic brain injury; RF = risk factor. [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 2.

Group comparisons on AUC of global network parameters

Groups ANOVA
TC Mild RF+ TBI Moderate/Severe TBI F(1,61) P Contrasts
n 27 20 16
Global network parameter
Characteristic path length 1.246 (0.058) 1.256 (0.070) 1.292 (0.556) 6.3 0.022 MS > TC
Modularity 0.102 (0.003) 0.102 (0.002) 0.102 (0.025) 0.3 0.63 —
Transitivity 0.071 (0.002) 0.071 (0.002) 0.070 (0.003) 0.4 0.09 —
Assortativity 0.014 (0.005) 0.015 (0.005) 0.015 (0.004) 0.0 0.58 —
Open in a new tab

Note. Data reflect mean (standard deviation), unless otherwise indicated. Global network parameters were calculated in the SLD‐FA network. AUC = area under the curve describing the relation between the global network parameter and matrix threshold; TC = traumatic control; RF = risk factor; TBI = traumatic brain injury; M = mildRF+ TBI group; MS = moderate/severe TBI group.

Functional Outcome

In a previous study [Königs et al., 2017], we found that the current samples of children with mildRF+ TBI and children with moderate/severe TBI had decreased neurocognitive functioning and increased behavior problems. Here, we replicate these findings (see Table III), by showing that mildRF+ TBI and moderate/severe TBI group had lower FSIQ (P = 0.003, d = −0.93 and P = 0.018, d = −0.84), poorer working memory performance (P = 0.005, d = −0.92 and P = 0.041, d = −0.67) and poorer encoding in long‐term verbal memory (P = 0.041, d = −0.63 and P = 0.014, d = −0.86) than the TC group. With regard to behavioral functioning, the mildRF+ TBI and moderate/severe TBI groups had higher ratings of internalizing problems than the TC group (P = 0.024, d = 0.85 and P = 0.005, d = 0.88) and the mildRF+ TBI group had higher ratings of externalizing problems as compared to the TC group (P = 0.001, d = 1.13).

Table 3.

Group comparisons on aspects of functional outcome

Groups ANOVA
TC Mild RF+ TBI Moderate/Severe TBI F(1,61) P Contrasts
n 27 20 16
Neurocognitive functioning
FSIQ 0.47 (0.82) −0.37 (1.04) −0.24 (0.93) 5.6 0.006 TC > M, MS
Digit Span 0.44 (0.87) −0.36 (0.91) −0.17 (1.01) 4.0 0.01 TC > M, MS
RAVLT Encoding 0.40 (0.90) −0.20 (1.05) −0.37 (0.95) 3.9 0.03 TC > M, MS
Behavior Problems
Internalizing Problems −0.44 (0.77) 0.20 (0.77) 0.43 (1.33) 5.1 0.009 TC < M, MS
Externalizing Problems −0.37 (0.88) 0.58 (0.84) −0.14 (1.10) 6.3 0.003 TC, MS < M
Open in a new tab

Note. Z scores are reported. Data reflect mean (standard deviation), unless otherwise indicated. TC = traumatic control; TBI = traumatic brain injury; RF = risk factor; FSIQ = full‐scale IQ; M = mildRF+ TBI group; MS = moderate/severe TBI group; RAVLT = Rey Auditory Verbal Learning Test.

Relations Between Structural Connectivity and Functional Outcome

We assessed the relations among network parameters with obtained effects of TBI (characteristic path length) and aspects of neurocognitive and behavioral outcome with sensitivity to TBI (FSIQ, working memory, RAVLT encoding and internalizing and externalizing behavior problems). In the sample of children with TBI, longer characteristic path length was associated with lower FSIQ (r = −0.34, P = 0.042) and poorer working memory performance (r = −0.45, P = 0.006). These relations were not observed in the TC group (–0.04 ≤ rs ≤ 0.00, Ps ≥ 0.83) and other relations between characteristic path length and aspects of neurocognitive and behavioral functioning in the TBI sample were also not significant (–0.10 ≤ rs ≤ 0.15, Ps ≥ 0.40). Taken together, these findings suggest that the influence of moderate/severe TBI on characteristic path length is related to important aspects of neurocognitive outcome (FSIQ and working memory), but not to behavioral functioning after moderate/severe TBI.

Exploratory Analysis of Network Definitions

A more elaborate description of the results from our exploratory analyses is provided in the Supporting Information. In summary, we found that network definition strongly influences the properties of the reconstructed network. More specifically, SLD‐SLD matrices capture a network with high sensitivity to intrahemispheric connectivity (95–98% of all connections). In contrast, SLD‐FA and FA‐FA matrices capture networks with progressively higher sensitivity to interhemispeheric connectivity (64–30% and 86–33% of all connections, respectively). This finding indicates that network definition and weighting greatly influences sensitivity of the reconstructed network for interhemispheric connectivity.

Group comparisons on the reconstructed networks furthermore revealed that TBI had a differential impact on global network parameters depending on network definition. More specifically, the effect of moderate/severe TBI group manifested differently in the SLD‐SLD network (increased transitivity, P = 0.029, d = 0.74) than in the SLD‐FA and FA‐FA networks (increased characteristic path length, Ps ≤ 0.022, ds ≥ 0.82). Only increased characteristic path length was related to aspects of functional outcome in terms of intelligence (SLD‐FA network: r = −0.34, P = 0.042) and working memory (SLD‐FA network: r = −0.45, P = 0.006; and FA‐FA network: r = −0.40, P = 0.015). These exploratory analyses further indicate that network definition also influences the sensitivity of the reconstructed network to the impact and functional outcome of pediatric TBI.

DISCUSSION

This study used probabilistic tractography in combination with graph theory to study the structural connectome in children with mild to severe TBI during the chronic phase of recovery. The results of this study show that children with TBI are at risk of abnormal organization of the structural connectome. More specifically, children with moderate/severe TBI have increased characteristic path length, indicative of reduced structural integration in the connectome. Furthermore, longer characteristic path length was associated with poorer neurocognitive outcome of pediatric TBI, while no evidence was found for relations between aberrant structural connectivity and behavioral outcome. The results of this study contribute to our understanding of the impact of moderate/severe TBI on the structural connectome and its potential implication in neurocognitive dysfunction after pediatric TBI.

Analyses aimed at elucidating the impact of pediatric TBI on the structural connectome revealed no evidence for an impact of mildRF+ TBI on structural connectivity. This finding is in line with our earlier study that used DTI in combination with tract‐based spatial statistics to study white matter integrity, revealing no evidence for reduced white matter integrity in children with mildRF+ TBI [Königs et al., 2017]. That study did show decreased white matter volume in children with mildRF+ TBI, but the current study suggests that this may not translate to abnormal organization of the connectome. The negative findings in children with mildRF+ TBI further suggest that the reported effects of mild pediatric TBI on structural connectivity in the acute phase of recovery [Yuan et al., 2015] may be essentially transient in nature, for example caused by temporary cerebral edema or compensated for by neural plasticity. This hypothesis awaits empirical testing in future research.

In contrast to mildRF+ TBI, children with moderate/severe TBI were found to have abnormal organization of the structural connectome. When compared to the TC group, children with moderate/severe TBI had higher characteristic path length indicative of decreased structural integration. The current findings converge with previous studies reporting increased path length in the connectivity probability network after moderate/severe TBI in adults [Caeyenberghs et al., 2012a, 2012b, 2014; Kim et al., 2014] and a pilot study in children with complicated to severe TBI [Yuan et al., 2017]. We further extend these findings by showing that longer characteristic path length is associated with poorer neurocognitive outcome of pediatric TBI in terms of intelligence and working memory. This indicates that globally decreased structural integration in the connectome may be implicated in neurocognitive dysfunction following moderate/severe TBI in children. Surprisingly, we found no associations between global network properties and behavior problems as observed by parents and teachers. The absence of such associations may suggest that behavioral functioning after pediatric TBI is not related to global connectivity, but is rather dependent on the organization of more specific subnetworks such as the frontostriatal circuit [Max et al., 2012].

The current findings should be interpreted in the context of the results from our exploratory analyses on network definition. These results indicate that network definition has great influence on characteristics of the reconstructed network, as well as its sensitivity to the neuropathology and outcome of TBI. SLD (a measure of connectivity probability) and FA (a measure of connectivity integrity) are common measures of connectivity [Qi et al., 2015], which were found to capture differential aspects of the structural connectome. Definition by SLD captures a network (i.e., SLD‐SLD network, or the connectivity probability network) consisting of predominantly intrahemispheric connections. In contrast, network definition by FA captures a network (i.e., FA‐FA network, or the connectivity integrity network) with higher sensitivity for interhemispheric connections. Furthermore, the impact of moderate/severe TBI on structural connectivity differed for the connectivity probability network (i.e., increased local clustering) and the connectivity integrity network (i.e., decreased structural integration). Although these differential findings may be compatible with the general view that TBI affects the balance between segregation and integration [Pandit et al., 2013; Sharp et al., 2014; Yuan et al., 2015], they also indicate that network definition influences the observed effect of TBI on the connectome (i.e., increased clustering or decreased integration). The main findings of the current study were based on a network that was produced by weighting the connectivity probability network for connectivity integrity. This SLD‐FA network was most consistently related to measures of functional outcome in children with TBI. We speculate that the use of multiple measures of connectivity may result in a richer network reconstruction, representing more biological meaningful dimensions of the structural connectome.

This study had some weaknesses. First, we investigated a relatively small sample of patients (n = 63), especially with regard to the moderate/severe TBI group (n = 16). Therefore, negative findings in this study may be the result of limited statistical power. Second, a limited set of instruments was used to measure outcome in terms of neurocognitive and behavioral functioning. Third, the moderate/severe TBI group had lower SES than the TC group, which potentially could have confounded the observed impact of moderate/severe TBI on structural connectivity. However, exploratory analysis revealed that SES was unlikely to have confounded the observed effects of moderate/severe TBI, as SES was not significantly associated with the relevant network measures in the TC group (0.00 ≤ rs ≤ 0.23 Ps ≥ 0.25). Strong points of this study include: (1) the use of a TC group to account for premorbid risk factors that may increase the risk of TBI (e.g., psychiatric disorder); (2) the inclusion of an exploratory analysis on the influence of network definition on reconstruction of the connectome; (3) and the use of neurocognitive and behavioral measures according to the guidelines for Common Data Elements in research on pediatric TBI.

In conclusion, this is the first study investigating the structural connectome of children with mild to severe TBI in the chronic phase of recovery. The results show that children with moderate/severe TBI have aberrant organization of the connectome that relates to neurocognitive outcome. In the assessed structural network, children with moderate/severe TBI had reduced structural integration, which was related to poorer intelligence and working memory. These findings suggest that abnormal connectivity is implicated in neurocognitive dysfunction after pediatric TBI. Findings from exploratory analyses further suggest that network definition greatly influences the properties of the reconstructed connectome in terms of sensitivity to the impact and outcome of pediatric TBI, which should be taken into account when integrating results across studies.

Conflicts of interest

The authors have nothing to disclose.

Supporting information

Supporting Information

Click here for additional data file. (323.7KB, docx)

ACKNOWLEDGMENTS

We are grateful to Dr. J.A. van der Sluijs from the department of Pediatric Orthopedics (VU University Medical Centre Amsterdam) and Dr. H.A. Heij from the Pediatric Surgical Center Amsterdam (VU University Medical Center and Academic Medical Center) for their assistance in the recruitment of participants for this study

REFERENCES

  1. American College of Surgeons (2004): Advanced Trauma Life Support Program for Doctors. Ed. Committee on Trauma 7th ed. Chicago: IL.
  2. Babcock L, Yuan W, Leach J, Nash T, Wade S (2015): White matter alterations in youth with acute mild traumatic brain injury. J Pediatr Rehabil Med 8:285–296. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Babikian T, Asarnow R (2009): Neurocognitive outcomes and recovery after pediatric TBI: Meta‐analytic review of the literature. Neuropsychology 23:283–296. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Behrens TEJ, Woolrich MW, Jenkinson M, Johansen‐Berg H, Nunes RG, Clare S, Matthews PM, Brady JM, Smith SM (2003): Characterization and propagation of uncertainty in diffusion‐weighted MR imaging. Magn Reson Med 50:1077–1088. [DOI] [PubMed] [Google Scholar]
  5. Caeyenberghs K, Leemans A, Decker Heitger DC, Drijkoningen M, Linden D, Vander C, Sunaert S, Swinnen SP (2012a): Brain connectivity and postural control in young traumatic brain injury patients: A diffusion MRI based network analysis. NeuroImage Clin 1:106–115. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Caeyenberghs K, Leemans A, Heitger MH, Leunissen I, Dhollander T, Sunaert S, Dupont P, Swinnen SP (2012b): Graph analysis of functional brain networks for cognitive control of action in traumatic brain injury. Brain 135:1293–1307. [DOI] [PubMed] [Google Scholar]
  7. Caeyenberghs K, Leemans A, Leunissen I, Gooijers J, Michiels K, Sunaert S, Swinnen SP (2014): Altered structural networks and executive deficits in traumatic brain injury patients. Brain Struct Funct 219:193–209. [DOI] [PubMed] [Google Scholar]
  8. Cohen J (1988): Statistical power analysis for the behavioral science. Hillsdale, NY: Erlbaum. [Google Scholar]
  9. Dall'Acqua P, Johannes S, Mica L, Simmen H‐P, Glaab R, Fandino J, Schwendinger M, Meier C, Ulbrich EJ, Müller A, Jäncke L, Hänggi J (2016): Connectomic and surface‐based morphometric correlates of acute mild traumatic brain injury. Front Hum Neurosci 10:127. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Epskamp S, Cramer AOJ, Waldorp LJ, Schmittmann VD, Borsboom D (2012): qgraph: Network visualizations of relationships in psychometric data. J Stat Softw 48:1–18. [Google Scholar]
  11. Fagerholm ED, Hellyer PJ, Scott G, Leech R, Sharp DJ (2015): Disconnection of network hubs and cognitive impairment after traumatic brain injury. Brain 138:1696–1709. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Fornito A, Zalesky A, Breakspear M (2015): The connectomics of brain disorders. Nat Rev Neurosci 16:159–172. [DOI] [PubMed] [Google Scholar]
  13. Gong G, He Y, Concha L, Lebel C, Gross DW, Evans AC, Beaulieu C (2009): Mapping anatomical connectivity patterns of human cerebral cortex using in vivo diffusion tensor imaging tractography. Cereb Cortex 19:524–536. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Jenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW, Smith SM (2012): FSL. Neuroimage 62:782–790. [DOI] [PubMed] [Google Scholar]
  15. Kim J, Parker D, Whyte J, Hart T, Pluta J, Ingalhalikar M, Coslett HB, Verma R (2014): Disrupted structural connectome is associated with both psychometric and real‐world neuropsychological impairment in diffuse traumatic brain injury. J Int Neuropsychol Soc 20:887–896. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Königs M, Heij HA, van der Sluijs JA, Vermeulen RJ, Goslings JC, Luitse JSK, Poll‐Thé BT, Beelen A, van der Wees M, Kemps RJJK, Catsman‐Berrevoets CE, Oosterlaan J (2015): Pediatric traumatic brain injury and attention deficit. Pediatrics 136:534–541. [DOI] [PubMed] [Google Scholar]
  17. Königs M, Pouwels PJW, van Heurn LWE, Vermeulen RJ, Goslings JC, Poll‐The BT, van der Wees M, Catsman‐Berrevoets CE, Oosterlaan J (2017): Relevance of neuroimaging for neurocognitive and behavioral impairment after pediatric traumatic brain injury. Brain Imaging Behav 1–15. 26780240 [Google Scholar]
  18. Li L, Liu J (2013): The effect of pediatric traumatic brain injury on behavioral outcomes: A systematic review. Dev Med Child Neurol 55:37–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Max J, Koele S, Smith W Jr (1998): Psychiatric disorders in children and adolescents after severe traumatic brain injury: A controlled study. Child Adolesc. Psychiatry 37:932–840. [DOI] [PubMed] [Google Scholar]
  20. Max J, Wilde E, Bigler E (2012): Neuroimaging correlates of novel psychiatric disorders after pediatric traumatic brain injury. Child Adolesc Psychiatry 51:1208–1217. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Pandit AS, Expert P, Lambiotte R, Bonnelle V, Leech R, Turkheimer FE, Sharp DJ (2013): Traumatic brain injury impairs small‐world topology. Neurology 80:1826–1833. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Park H, Friston K (2013): Structural and functional brain networks: From connections to cognition. Science 342:579–587. [DOI] [PubMed] [Google Scholar]
  23. Patenaude B, Smith SM, Kennedy DN, Jenkinson M (2011): A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 56:907–922. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Qi S, Meesters S, Nicolay K, Romeny BM, Ossenblok P (2015): The influence of construction methodology on structural brain network measures: A review. J Neurosci Methods 253:170–182. [DOI] [PubMed] [Google Scholar]
  25. Roberts R, Mathias J, Rose S (2014): Diffusion Tensor Imaging (DTI) findings following pediatric non‐penetrating TBI: A meta‐analysis. Dev Neuropsychol 39:600–637. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Rubinov M, Sporns O (2010): Complex network measures of brain connectivity: Uses and interpretations. Neuroimage 52:1059–1069. [DOI] [PubMed] [Google Scholar]
  27. Sattler JM (2001): Assessment of Children: Cognitive Applications. San Diego: Jerome M. Sattler. p 774.
  28. Sharp DJ, Scott G, Leech R (2014): Network dysfunction after traumatic brain injury. Nat Rev Neurol 10:156–166. [DOI] [PubMed] [Google Scholar]
  29. Squarcina L, Bertoldo A, Ham TE, Heckemann R, Sharp DJ (2012): A robust method for investigating thalamic white matter tracts after traumatic brain injury. Neuroimage 63:779–788. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Statistics Netherlands (2006): Education Categorization Standard [Standaard onderwijsindeling 2006]. Den Haag: NL. Available at: http://www.cbs.nl. http://www.cbs.nl/nl-NL/menu/methoden/classificaties/overzicht/soi/2006/default.htm
  31. Teasdale G, Jennett B (1976): Assessment and prognosis of coma after head injury. Acta Neurochir (Wien) 34:45–55. [DOI] [PubMed] [Google Scholar]
  32. Van Beek L, Vanderauwera J, Ghesquiere P, Lagae L, De Smedt B (2015): Longitudinal changes in mathematical abilities and white matter following paediatric mild traumatic brain injury. Brain Inj 9052:1–10. [DOI] [PubMed] [Google Scholar]
  33. van den Burg W, Kingma a (1999): Performance of 225 Dutch school children on Rey's Auditory Verbal Learning Test (AVLT): Parallel test‐retest reliabilities with an interval of 3 months and normative data. Arch Clin Neuropsychol 14:545–559. [DOI] [PubMed] [Google Scholar]
  34. Verhulst F, van der Ende J (2013): Handleiding ASEBA Vragenlijsten voor leeftijden 6 t/m 18 jaar: CBCL/6‐18, YSR & TRF. ASEBA. Rotterdam: NL.
  35. Vos P, Battistin L (2002): EFNS guideline on mild traumatic brain injury: Report of an EFNS task force. Eur J Neurol 9:207–219. [DOI] [PubMed] [Google Scholar]
  36. Vu JA, Babikian T, Asarnow RF (2011): Academic and Language outcomes in Children after Traumatic Brain Injury A Meta Analysis. Except Child 77:263–281. [Google Scholar]
  37. Wechsler D (1991): WISC‐III: Wechsler Intelligence Scale for Children. San Antonio, TX: Psychological Corporation. [Google Scholar]
  38. World Health Organization (2006): Neurological Disorders: Public Health Challenges. Geneva, CH: World Health Organization. [Google Scholar]
  39. Yuan W, Wade SL, Babcock L (2015): Structural connectivity abnormality in children with acute mild traumatic brain injury using graph theoretical analysis. Hum Brain Mapp 36:779–792. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Yuan W, Treble‐Barna A, Sohlberg MM, Harn B, Wade SL (2017): Changes in structural connectivity following a cognitive intervention in children with traumatic brain injury: A pilot study. Neurorehabil Neural Repair 31:190–201. [DOI] [PubMed]
  41. Zhang Y, Brady M, Smith S (2001): Segmentation of brain MR images through a hidden Markov random field model and the expectation‐maximization algorithm. IEEE Trans Med Imaging 20:45–57. [DOI] [PubMed] [Google Scholar]
  42. Zhang Z, Liao W, Chen H, Mantini D, Ding JR, Xu Q, Wang Z, Yuan C, Chen G, Jiao Q, Lu G (2011): Altered functional‐structural coupling of large‐scale brain networks in idiopathic generalized epilepsy. Brain 134:2912–2928. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supporting Information

Click here for additional data file. (323.7KB, docx)

Articles from Human Brain Mapping are provided here courtesy of Wiley

RESOURCES