Participants
Participants were invited to take part in a two-session, behavioural and MRI study over a total of 3 hours at Western University and all protocols were approved by Western's Research Ethics Board. Parents provided informed written consent for their children’s participation in the study and children provided verbal assent.
Ten children with hydrocephalus were recruited from a paediatric neurosurgery clinic in London, Ontario. All patients were clinically stable at time of recruitment and received a ventriculoperitoneal (VP) shunt within their first year of life. Out of the ten patients, six successfully completed the neuroimaging protocol and ranged from 6–10 years of age (M = 8.43 years, SD = 1.61). All patients who completed the DTI protocol are denoted in Table 1. Patient 4A completed the entire protocol twice almost a year apart (356 days) as patient 4A2 and was included separately in the analysis.
Twenty-two neurotypically developing children were recruited as healthy controls through the Developmental Research Participant Pool and the OurBrainsCAN Cognitive Neuroscience Research Registry at Western University in London, Ontario. Ages ranged from 6–10 years old, and the mean age was not significantly different from that of the patient group (M = 8.35 years old, SD = 1.54, p = .455). Three controls were unable to complete the diffusion MRI protocol and were thus excluded from the neuroimaging analysis.
Parental socioeconomic status (SES) is an important variable to control for when studying neurodevelopment as low SES is associated with adverse neurocognitive outcomes in children with brain injury [25]. SES was measured for every participant using the Hollingshead Four Factor scale, which assesses maternal and paternal educational and occupational status [26]. A t-test of parental SES scores revealed no group differences between our patients and controls (Mpatients= 45.71, SDpatients= 11.67, Mcontrols= 47.52, SDcontrols= 13.15, p = .367).
Table 1
Clinically relevant demographics of all patients recruited for the study. Patients who completed the diffusion neuroimaging protocol denoted with an asterisk (*).
Patient | Age (years) | Sex | Prematurity (weeks) | Birth- weight | Etiology | Age of Onset | # Shunt Revisions |
1A | 5.42 | M | Yes (25) | XLBW | IVH | Birth | 0 |
2A* | 9.92 | M | No | Normal | IVH | 4 months | 0 |
3A | 5.33 | M | Yes (27) | VLBW | IVH | Birth | 0 |
4A* | 6.25 | M | No | Normal | DWM | 3 weeks | 0 |
5A* | 8.92 | M | Yes (26) | XLBW | IVH + MG | Birth | 0 |
6A | 11.33 | M | Yes (28) | VLBW | IVH | Birth | 3 |
7A* | 10.08 | F | No | Normal | SB | 1 week | 0 |
8A* | 6.83 | M | No | Normal | N/A | 8 months | 0 |
9A* | 9.75 | M | No | Normal | SB | 1 week | 2 |
10A | 5.75 | F | Yes (26) | LBW | IVH | Birth | 3 |
LBW = low birth weight, VLBW = very low birth weight, XLBW = extra low birth weight, IVH = Intraventricular Hemorrhage, DWM = Dandy-Walker’s Malformation, MG = meningitis, SB = Spina Bifida |
BRIEF2
Parental reports of everyday functioning within the past 6 months were obtained using the second edition of the Behavioural Rating Inventory of Executive Functions (BRIEF2) [27]. The BRIEF is an ecologically valid assessment of behaviour outside the laboratory setting and a gold standard used in a range of childhood disorders [7, 28]. Responses are scored into sub-domains of executive functioning, which are combined into a Behaviour Regulation Index (BRI), Emotional Regulation Index (ERI), Cognitive Regulation Index (CRI), and an overall Global Executive Composite (GEC). All raw scores are converted into age and sex standardized t-scores, with higher scores indicating greater executive dysfunction.
Mri Acquisition
All participants were first introduced to a mock scanner for 10–15 minutes to reduce potential anxiety or fear, during which sounds are reproduced, and the child can had some direct feedback and a discussion while lying in the scanner. MRI images were then acquired at the Centre for Functional and Metabolic Mapping at the University of Western Ontario using a 3-Tesla Siemens Magnetom Prisma Fit scanner and a Siemens Prisma 32-channel head coil. The entire imaging protocol lasted approximately 1 hr 15 min, and included diffusion imaging, high-resolution T1-weighted imaging, and other structural and functional imaging [29]. Cartoon clips were played during image acquisition to reduce motion artifacts.
The imaging protocol began with a localizer scout image and proceeded with two consecutive series of diffusion weighted imaging (DWI) obtained in opposite phase-encoding directions along the anterior-posterior axis with the following parameters: gradient directions = 30; b-value = 1000s/mm2; isometric voxel size = 2\(\times\)2\(\times\)2 mm3; and matrix size = 192\(\times\)192 mm. For each DWI series, a single volume referred to as the b0 was acquired without diffusion weighting (b-value = 0 s/mm2). Next, a high-resolution T1-weighted anatomical image was acquired using a three-dimensional magnetization-prepared rapid acquisition with gradient echo (MP-RAGE) sagittal pulse sequence, with the following parameters: repetition time (TR) = 2300ms; echo time (TE) = 2.93 ms; flip angle = 9\(^\circ\); matrix size = 256\(\times\)256 pixels; and where one whole brain image consisted of 160, 1 mm-thick slices. The field of view of the anatomical image was oriented along the anterior and posterior commissure with a matrix of 256\(\times\)256 pixels and a 1\(\times\)1\(\times\)1mm3 isotropic voxel size. Once the T1-weighted image was collected, the exact same DWI series reported above was replicated, resulting in a total of four DWI series.
Image Processing
All preprocessing steps were implemented using the standard FMRIB Software Library (FSL) v6.0.3 FDT toolbox [30, 31]. Structural and diffusion images were first converted from DICOM format to NIfTI volumes and organized into Brain Imaging Data Structure (BIDS) [32]. T1-weighted anatomical images were then skull-stripped iteratively using robust brain centre estimation and a binary brain mask was generated. Volumes with skull-brain interface outlines were visualized to ensure correct brain extraction, and failures were corrected at the individual subject level. Next, b0 volumes were extracted from every diffusion series and concatenated into a single volume that was used for susceptibility-induced distortion correction. All 4 DWI series were concatenated into a single image, along with the respective metadata. This approach was taken to increase the signal to noise ratio (SNR) for tensor estimation. Subject movement was corrected by linear registration to the first b0 volume and eddy current induced distortion correction was applied.
Diffusion tensor models were fit at each voxel within the pre-processed diffusion volumes to obtain maps of corresponding eigenvectors (ε1, ε2, and ε3) and eigenvalues (λ1, λ2, and λ3). Measures of FA, an index for diffusion asymmetry, and MD, the average diffusivity across all three eigenvalues, were obtained at the voxel level, along AD and RD. To prepare data for probabilistic tractography, crossing fibers in preprocessed images were accounted for by fitting a multiple tensor model at each voxel and generating more accurate probability distributions of voxel-wise fiber orientations [30, 31].
Forty-eight cortical regions of interest (ROI) defined by the Harvard-Oxford Cortical FSL Atlas, and 7 striatal ROIs defined by the Oxford-GSK-Imanova Striatal Connectivity FSL Atlas were used to parcellate each brain in native diffusion space (Fig. 1). The seven striatal ROIs, including caudal-motor, limbic, rostral-motor, executive, parietal, occipital, and temporal regions, were previously defined by structural connectivity with cortical regions [33].
Probabilistic tractography was performed in native diffusion space between all 7 striatal and 48 cortical regions and volumetric maps of corticostriatal tracts seeded in the executive striatal sub-region were obtained. The tractography algorithm was applied with default parameters to iteratively sample 5000 streamlines per seed mask voxel. All cortical and striatal ROIs were used as exhaustive seed/target pairs to construct a 55\(\times\)55 matrix containing the structural connectivity index of each pair. (Fig. 1b). To avoid the possibility of streamlines following the directional diffusion of CSF along shunts, tractography was run for all patients with manually segmented 3D shunt masks used to terminate streamlines. Successful mapping of the shunt was reviewed and confirmed by a pediatric neurosurgeon along coronal, axial and horizontal slices in T1-weighted space. Tract volumes were defined by voxels containing streamline densities above a 10th percentile threshold, and gray-matter represented in seed regions was subtracted out. These volumes were binarized into a mask image used extract average FA, MD, AD, and RD
To correct for intra and inter-subject variability in ROI size, the connectivity index between ROIs was computed as the streamline count divided by the average number of voxels between seed and target ROIs [34]. This bi-directional, normalized connectivity index was used for all statistical tests.
Executive corticostriatal network generation
To reduce false positives associated with probabilistic tractography and obtain a stronger signal to noise ratio individual subjects were pooled into a group average connectivity matrix to and a threshold was applied [35–39]. To define the corticostriatal structural networks, a streamline-informed approach was taken [40]. A threshold of 10% the maximum connectivity index computed from the group average was used to exclude tracts. This resulting executive corticostriatal network contained 21 tracts. The structural network was assessed in terms of connectivity, and secondary measures of FA, MD, AD and RD (Fig. 2b–d).
Statistical Analyses
Dissimilarity between a patient’s executive corticostriatal network and the average control network was assessed by computing a Pearson correlation coefficient and determining the probability of obtaining such coefficient using a cumulative distribution function. To construct a distribution of correlation coefficients within a neurotypical sample, a bootstrapped estimation was performed. For every control, connectivity, FA, MD, AD, and RD correlations were bootstrapped 10,000 times between their executive corticostriatal network and a randomly generated average control network that was sampled with replacement from the rest of the control pool. The resulting distribution contained 190,000 coefficients and was used to identify the probability of obtaining a co-efficient for each patient. A Bonferroni correction was used for multiple comparison by dividing the critical value by 4 (FA, MD, AD, RD), and p < 0.0125 was used as the statistical threshold for determining significance. Subsequently, FA dissimilarity between patient and control executive corticostriatal network was used to assess the linear relationship between white matter integrity and GEC scores, and to investigate whether distance from the neurotypical network related to the extent of executive dysfunction. An exploratory regression with GEC and FA correlations of the striatal-occipital network was performed to investigate specificity of the findings to the executive network.