Neonatal functional and structural connectivity are associated with cerebral palsy at age 2

Abstract

Objective:

The accuracy of structural MRI to predict later cerebral palsy (CP) in newborns with perinatal brain injury is variable. Diffusion tensor imaging (DTI) and task-based functional MRI (fMRI) show promise as predictive tools. We hypothesized that infants who later developed CP would have reduced structural and functional connectivity as compared to those without CP.

Study Design:

We performed DTI and fMRI using a passive motor task at 40–48 weeks postmenstrual age in 12 infants with perinatal brain injury. CP was diagnosed at age 2 using a standardized exam.

Results:

Five infants had CP at age 2 and 7 did not have CP. Tract-based spatial statistics (TBSS) showed widespread reduction of fractional anisotropy (FA) in almost all white matter tracts in the CP group. Using the median FA value in the corticospinal tracts as a cutoff, FA was 100% sensitive and 86% specific to predict CP, compared with a sensitivity of 60–80% and specificity of 71% for structural MRI. During fMRI, the CP group had reduced functional connectivity from the right supplemental motor area as compared to the non-CP group.

Conclusion:

DTI and fMRI obtained soon after birth are potential biomarkers to predict CP in newborns with perinatal brain injury.

Introduction

Brain injury sustained in the perinatal period is the leading cause of later cerebral palsy (CP) in both term and preterm infants. Recent research has shown that early intervention in infants at risk of CP leads to improved motor outcomes¹. However, it is difficult to predict CP with certainty based on conventional structural MRI except in cases with very severe injury to motor areas. This means that children are followed and families are told to “wait and see” how the child’s motor development progresses, leading to both a lack of targeted early treatments for children who could benefit, and unnecessary general early intervention services for children who will have a normal motor outcome.

Diffusion tensor imaging (DTI) is a well-established technique to evaluate the white matter microstructure of the brain. Fractional anisotropy (FA), a measure derived from DTI, reflects the degree to which the motion of water molecules is directionally restricted in white matter. Multiple studies have shown that FA of various white matter tracts in former preterm infants is associated with later motor outcomes ^2-4. More recently, FA has been associated with later outcomes in infants with perinatal stroke and hypoxic-ischemic encephalopathy (HIE) ^5-7. FA in the corticospinal tracts correlates with Bayley motor score at age 1–2 in infants with HIE⁵. Infants with HIE and unfavorable later outcomes have significantly lower FA values in the corticospinal tracts and corpus callosum on neonatal MRI⁶. Functional connectivity derived from task-based and resting-state functional MRI (fMRI) is less well-studied in the neonatal population^8-12. Only one published study has correlated neonatal functional connectivity with later outcomes in infants with brain injury, finding that functional connectivity MRI predicted motor impairment at 4 and 8 months of age¹³.

Our long-term goal is to determine whether DTI and fMRI can be used as an adjunct to conventional structural clinical MRI in the first few weeks after birth to add additional information for clinicians and families about potential damage to motor pathways. Our goal for this study was to evaluate the prognostic value of DTI and functional connectivity derived from task-based fMRI in infants with perinatal brain injury at <48 weeks corrected age for motor outcomes at 2 years of age. We hypothesized that infants with perinatal brain injury who later developed CP would have reduced structural and functional connectivity in the neonatal period compared with those with normal motor outcomes.

Methods

Participants

Eighteen infants with perinatal brain injury due to hypoxic-ischemic encephalopathy (HIE), stroke, periventricular leukomalacia (PVL), or intraventricular hemorrhage (IVH) were identified based on clinical head ultrasound or clinical MRI performed during the NICU stay. Inclusion criteria were infants of any gestational age with grade 3–4 intraventricular hemorrhage, periventricular leukomalacia, perinatal stroke, or hypoxic-ischemic injury seen on clinical neonatal head ultrasound or MRI. Exclusion criteria were inability to undergo a study MRI prior to 48 weeks postmenstrual age and any metal implants. The study MRI protocol included structural MRI, DTI, and fMRI using a motor stimulation paradigm before 48 weeks postmenstrual age. The protocol was approved by the Institutional Review Board and informed parental consent was obtained. All imaging was performed during unsedated natural sleep. Hearing protection was provided with earplugs, MiniMuffs, and headphones. We have previously reported the results of a visual fMRI task from a subset of infants in this study in a previous publication¹⁰.

MRI acquisition parameters

MRI data were all acquired on a 3T Philips Achieva MRI scanner (Phillips Medical Systems, Best, the Netherlands) in the Imaging Research Center at Cincinnati Children’s Hospital using a 32-channel phased array head coil. Infants were swaddled for scanning and placed in the head coil with hearing protection.

Structural MRI

High spatial resolution (1mm isotropic) 3D T1-weighted and 3D T2-weighted scans were obtained for anatomic reference. For the 3D T1 sequence, TR/TE = 8.1 msec/3.7 msec, flip angle = 8^o, matrix size 192 × 192, number of slices = 140, with a TFE of 192 and total time 4:36. For the 3D T2 sequence, TR/TE = 4800 msec/230 msec, flip angle = 90^o, number of slices = 140 and total time 3:12.

Diffusion imaging

Diffusion imaging was obtained using a single shot spin echo EPI sequence in the axial plane using a b value of 800 s/mm² and 61 directions uniformly distributed over a sphere using an electrostatic repulsive model, interleaved with 7 b=0 scans. Scan parameters were TR/TE = 6389msec/77msec, field of view (FOV) 224mm x 224mm, matrix size 112×109, slice thickness = 2 mm, voxel size 2.0 × 2.05 × 2.0, mm³, number of slices = 60 , SENSE factor = 3, receiver BW = 1752.6Hz/pixel. Total time was 10:42.

fMRI

fMRI was performed in the axial plane using a single shot gradient echo echo-planar imaging sequence with TR/TE = 2000msec/35msec, flip angle = 90^o, field of view 180×180mm, matrix size 64 × 64, slice thickness = 3mm, voxel size 2.8 × 2.8 × 3 mm³, number of slices = 36, SENSE Factor = 2, receiver BW = 3423.6Hz/pixel, total time 5:38.

fMRI Motor task paradigm

The motor paradigm was administered in a periodic block design consisting of 5.5 cycles of 30 second ON/OFF conditions. For the ON condition, a sensorimotor stimulus (opening and closing of both hands) was delivered via a suction bulb placed in each of the subject’s hands. The suction bulb was inflated and deflated by the research assistant via a second suction bulb that was attached with tubing. The OFF condition was no movement of the hand via suction bulb. The total scan time was 5 minutes 38 seconds, during which 165 BOLD (blood oxygen level dependent) image volumes were obtained.

Analysis

Structural MRIs were read by a pediatric neuroradiologist blinded to all subject information except for corrected gestational age at time of scan. The thalamus, basal ganglia, and posterior limb of the internal capsule (PLIC), areas known to be predictive of motor outcome ¹⁴, were assessed systematically.

Follow up procedures

Infants were seen for a follow-up visit at the age of 22–26 months in the NICU Follow-up Clinic. A certified examiner blinded to study MRI results performed a standardized neurologic exam (the NICHD Neonatal Research Network standard neurological exam)¹⁵ and diagnosed CP based on abnormalities in tone, reflexes, coordination and movement and delay in motor milestones. For participants diagnosed with CP, Gross Motor Function Classification Level (GMFCS) was determined ¹⁶ and cerebral palsy was classified as mild (GMFCS 1), moderate (GMFCS 2 or 3), or severe (GMFCS 4 or 5). Infants were classified into 2 groups, CP and non-CP, based on the results of the neurologic exam.

DTI analysis

DTI data were pre-processed in FSL software (FMRIB Software Library, FMRI, Oxford, UK)¹⁷ including correction for eddy current and head motion¹⁸ and skull stripping¹⁹. Preprocessed images were subjected to tensor decomposition for generating FA maps in FDT (FMRIB’s Diffusion Toolbox) ²⁰. All subjects’ FA maps were aligned into the JHU_neonate_nonlinear_fa template²¹ using a non-linear registration in FNIRT (FMRIB’s Non_linear Registration Tool)²² and then the affine transform to MNI space was applied. After transformation into MNI space, a mean FA image was created and thinned to generate a mean FA skeleton of white matter tracts. Tract-based spatial statistics (TBSS)²³ was then used to test group differences of DTI measures. Voxel-wise statistical analysis across subjects was carried out for voxels with FA>0.15 to include only major fiber bundles. Non-parametric permutation based statistical analysis was performed with Randomise ²⁴, and statistical results are reported as corrected p-values at p<0.05 after controlling for family-wise error rate.

To investigate the diffusion changes in specific white matter tracts, the JHU neonatal white matter atlas²¹ was used to parcellate the white matter into 27 white matter regions of interest (ROIs)²⁵. FA was calculated by averaging values within each region based on the white matter atlas. We specifically chose to focus on the tracts related to motor function for the ROI analysis, namely: the entire corpus callosum, body of the corpus callosum, genu of the corpus callosum, splenium of the corpus callosum, right and left cingulate gyrus, right and left corticospinal tracts, the tract between the premotor cortex and the primary motor cortex, and the tract between the thalamus and the primary somatosensory cortex. A two sample t-test analysis was performed to compare the results of ROI-based data between the CP and normal motor outcome groups. A false discovery rate (FDR) p<0.05 was applied to correct for multiple comparisons.

fMRI analysis

For the motor functional connectivity analysis, all image processing was performed using SPM8 (Welcome Department of Imaging Neuroscience, London, UK, http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). The functional image series were corrected for head motion by realigning all images to the mean of all functional volumes. The mean functional image was co-registered to the corresponding T1-weighted high-resolution image from the child during the same imaging session. T1-weighted images were segmented using neonatal tissue probability maps²⁶. Finally, the images were normalized to the neonatal template, resliced to a voxel size 2×2×2mm, and smoothed with a full width at half maximum (FWHM) Gaussian Smoothing kernel of 6mm. After pre-processing for each subject, a GLM was used to regress the time course of the motor task versus rest as the main effect and the realignment parameters as regressors of no interest. SPM.mat files of each subject were imported into CONN (Functional Connectivity Toolbox v.14, https://www.nitrc.org/projects/conn)²⁷ to process the ROI-to-ROI connectivity analyses. This toolbox implements a noise reduction method (aCompCor)²⁸ that accounts for physiological noise, and the covariates were included with a principal components analyses (PCA) reduction of the signal from subject specific white matter and CSF masks. The residual BOLD timeseries was then low-pass filtered (.008<f). Additionally, 6 motion parameter timeseries were condensed into a single vector and entered into second level analysis to reduce further motion effects. Functional connectivity analysis was performed by grouping voxels into 223 ROIs based upon a neonatal functional brain parcellation atlas²⁹. Bi-variate correlations were calculated between each pair of ROIs. All ROIs were imported as possible connections for our selected ROIs. The unpaired t-test was used with a threshold set at p<0.05 false discovery rate (FDR) corrected to determine whether significant differences existed between the groups (CP versus non-CP).

Other statistical analyses

Statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary NC). Because the data had a non-normal distribution, we performed non-parametric statistical testing to compare the 2 groups (CP and non-CP). Clinical and socioeconomic factors were compared using the Wilcoxon rank sum test for continuous variables and Fisher’s exact test for binary variables. Additional analyses were performed to determine the sensitivity and specificity of the median FA value in each tract as a cut point to predict CP. Sensitivity was defined as the ability of FA below the median value to correctly classify an infant as later developing CP. Specificity was defined as the ability of FA above the median value to correctly classify as infant as NOT developing CP.

Results

Participants and structural MRI

Complete good quality DTI and motor task fMRI data sets were collected in 12 of the 18 infants with perinatal brain injury imaged for this study. The remaining 6 infants had too much motion during the fMRI and/or DTI sequences (n=3) or were too fussy to complete the sequences (n=3). Of these 12 infants, 5 were diagnosed with CP by 2 years (CP group) and 7 infants had a normal motor exam at 2 years (non-CP group). Demographic information and information about brain injury and outcome is shown in Table 1. The CP group and non-CP were compared for various clinical and sociodemographic factors with no differences seen, as shown in Table 2.

Table 1:

Demographic and brain injury information

Subject	Gestational age (weeks)	BW	Sex	NICU days	Type of injury	PLIC involvement	Basal ganglia involvement	Thalamic involvement	Other	Cerebral palsy
1	40	4170	F	22	HIE	none	none	none	Moderate bilateral watershed injury	no
2	36	2480	F	17	IVH	none	none	none	Grade 3 IVH bilaterally	no
3	29	1340	F	42	PVL	Abnormal signal and lack of myelination	none	Mild volume loss		Yes (severe)
4	38	3680	M	11	stroke	Mildly delayed myelination	none	Moderate signal abnormality		Yes (moderate)
5	39	3518	M	19	HIE	none	none	none	Diffusely abnormal white matter signal with small periatrial lesions	no
6	37	2990	F	44	HIE	Abnormal signal and lack of myelination	Abnormal signal bilaterally	Severe signal abnormality		Yes (severe)
7	37	3010	M	5	HIE	Abnormal signal and lack of myelination	None	Mild signal abnormality	Diffusely abnormal white matter signal	no
8	41	3390	F	47	HIE	Abnormal signal	Abnormal signal bilaterally	Severe signal abnormality	Severe watershed injury	Yes (severe)
9	34	3260	M	70	HIE	none	Abnormal signal bilaterally	none		no
10	39	2310	M	68	HIE	Abnormal signal	Abnormal signal bilaterally	Moderate signal abnormality	Abnormal white matter signal	no
11	36	2740	F	17	HIE	Abnormal signal	Abnormal signal bilaterally	Severe signal abnormality		Yes (mild)
12	39	3150	M	77	HIE/stroke	none	none	none	Occipital ischemia	no

HIE = hypoxic-ischemic encephalopathy, IVH = intraventricular hemorrhage, PVL = periventricular leukomalacia, PLIC = posterior limb of the internal capsule

Table 2:

Comparison of demographic variables

	CP group	No CP group	P value
Median (range) gestational age in weeks	37 (29–41)	39 (34–40)	0.631
Median (range) birth weight in grams	2990 (1340–3680)	3150 (2310–4170)	0.751
Male sex	20%	71%	0.242
Median (range) NICU stay in days	42 (11–47)	22 (5–77)	0.691
Public insurance	80%	71%	1.002

¹Using Wilcoxon rank sum test

²Using Fisher’s exact test

The sensitivity and specificity of structural abnormalities based on T1 and T2 weighted images to predict outcome was calculated. Moderate/severe PLIC involvement (as assessed by a blinded pediatric neuroradiologist) had a sensitivity of 80% and a specificity of 71% to predict later CP. Moderate/severe basal ganglia involvement had a sensitivity of 60% and a specificity of 71% to predict CP.

Structural connectivity

TBSS showed widespread reduction of FA in all white matter tracts related to motor function in the infants who later developed CP as compared to the group with normal motor outcome (Figure 1). ROI-based analysis also showed differences in FA in all white matter tracts related to motor function between groups (Figure 2).

Figure 1.

Tract-based spatial statistics (TBSS) map showing the DTI skeleton based on peak fractional anisotropy (FA) value for each participant in the group in green, and difference in FA between the cerebral palsy (CP) group and the non-CP group in red. Areas in red represent the tracts where the non-CP group had higher FA than the CP group, using a threshold of p<0.05 corrected for multiple comparisons using the family-wise error rate.

Figure 2.

Results of region-of-interest analysis showing differences in FA in motor-related tracts between the cerebral palsy (CP) group (light blue) and the non-CP group (dark blue). All tracts shown were significantly different between groups (p<0.05, false discovery rate corrected). BCC = body of corpus callosum, CC = corpus callosum, CG = cingulate gyrus, CST = corticospinal tract, GCC = genu of corpus callosum, preprimaryMC = tract connecting premotor cortex to primary motor cortex, SCC = splenium of corpus callosum, thal-PSC = tract connecting the thalamus to the primary somatosensory cortex.

Prediction of outcome using FA

FA value below the median in most white matter tracts had good sensitivity and specificity to predict CP. Namely, FA below the median in the corpus callosum, genu of the corpus callosum, body of the corpus callosum, left corticospinal tract, bilateral tracts connecting premotor-primary motor cortex tracts, and bilateral tracts connecting thalamus-primary sensory cortex correctly predicted all 5 infants who developed CP and falsely predicted one normal baby as developing CP, leading to a sensitivity of 100% and specificity of 86%.

Functional connectivity

During the motor task, the CP group had lower functional connectivity from the right supplemental motor area to other ROIs than the non-CP group (Figure 3). No connectivity differences were seen in other areas of the brain during the motor task.

Figure 3.

Functional connectivity map derived from BOLD time courses during the motor task. Using a significance threshold of p<0.05 (false discovery rate corrected), only the seed in the right supplemental motor area (SMA) was significant, with the non-cerebral palsy (CP) group having significantly increased connectivity from the right SMA to other areas of the brain as compared to the CP group. SFGdor-L = left dorsolateral superior frontal gyrus, HIP-R = right hippocampus, AMYG-R = right amygdala, MFG-R = right middle frontal gyrus.

Motor task fMRI

There were no differences in the brain activation during the motor task between the 5 infants in the CP group and the 7 infants with normal motor outcome in the control group.

Discussion

Motor function is complex and requires intact pyramidal and extrapyramidal systems in the brain, as well as an intact spinal cord and peripheral nervous system. The pyramidal pathway starts in the primary motor cortex (precentral gyrus) with axons extending down the corticospinal tracts, posterior limb of the internal capsule, cerebral peduncles, pyramids of the medulla, to the spinal cord. The extrapyramidal system consists of the basal ganglia (globus pallidus, putamen, caudate), thalamus, and cerebellum and modifies neural impulses originating in the primary motor cortex. In the term infant, injury to the motor system is usually caused by hypoxic-ischemic encephalopathy (injury to the deep gray matter in the basal ganglia/thalamus) or stroke (injury to the motor cortex). In the preterm infant, injury to the motor system is usually caused by white matter injury to the corticospinal tracts or the thalamocortical radiations from periventricular leukomalacia or periventricular hemorrhagic infarction³⁰. In our study, we evaluated the correlation of neonatal DTI and fMRI with later CP in infants with perinatal brain injury from both of these causes.

Structural MRI is the standard of care to evaluate brain injury in both term infants with suspected brain injury due to hypoxic ischemic encephalopathy (HIE) or stroke and preterm infants with suspected brain injury seen on head ultrasound. Several scores based on structural MRI have been developed for both term ^31-33 and preterm ^34-36 and patterns of injury to predict later outcomes, although not many have focused on motor outcomes specifically. For term infants with HIE, meta-analysis showed structural MRI had a pooled sensitivity of 91% and specificity of 51% to predict death or moderate/severe disability ³⁷. In cooled infants with HIE, major MRI abnormalities predict death or major disability with a sensitivity of 88% and a specificity of 82% ³⁸. In preterm infants, Hintz et al³⁹ found a sensitivity of 77% and a specificity of 81% of moderate-severe white matter abnormalities to predict CP and Woodward et al³⁴ found a sensitivity of 65% and a specificity of 84% of moderate-severe white matter abnormalities to predict CP. Van’t Hooft, in a meta-analysis, found sensitivity of 77% and specificity of 79% of term equivalent MRI to predict CP⁴⁰. In our cohort, macrostructural injury to motor regions, exhibited sensitivity and specificity values that were similar to the published literature for structural MRI for prediction of CP. Clearly, better predictive biomarkers are needed.

DTI evaluates microstructural changes in white matter tracts. FA reflects the degree of directionality in diffusion (anisotropy) and is therefore is higher in regions of greater organization. Higher FA values are therefore usually interpreted as reflecting a more organized and intact white matter structure, but may also reflect the presence of fewer crossing fibers within a particular voxel. Several studies have related DTI in the neonatal period and later motor outcomes. Massaro et al ⁵ found a strong positive association between FA in the corticospinal tracts and Bayley motor score at age 1–2 in 42 infants with HIE using DTI tractography. Tusor et al ⁶ used TBSS in 32 infants with HIE and evaluated motor outcome at 12–28 months. FA values correlated significantly with developmental quotient, and those infants with unfavorable later outcomes had significantly lower FA values in multiple white matter tracts, including the corpus callosum and anterior and posterior limbs of the internal capsule. Ancora et al ⁷ evaluated 15 infants with HIE treated with selective head cooling and followed the cohort to 2 years. They found that axial diffusivity and radial diffusivity, other DTI-derived measures of white matter organization, in the corpus callosum and frontal/parietal white matter respectively, were predictive of developmental (Griffiths) score at age 2. None of these authors reported the sensitivity or specificity of DTI measures to predict CP.

In preterm infants with periventricular hemorrhagic infarction, Roze et al ⁴ assessed the symmetry of FA in posterior limb of the internal capsule bilaterally and found this asymmetry index to predict later hemiplegia with a sensitivity of 100% and specificity of 87%, similar to our findings. Kim et al ² studied 32 preterm infants with DTI using ROI-based analysis and followed them to determine motor outcome at >15 months of age. The mean FA in the genu of the corpus callosum, the PLIC, and the centrum semiovale was lower in the 5 infants with white matter abnormalities who developed CP compared with the 5 infants with white matter abnormalities who did not develop CP. Using TBSS, van Kooij et al ³ studied 63 preterm infants at term equivalent age and at age 2, and found that gross motor scores were associated with FA and radial diffusivity throughout the WM. In our cohort, we found that FA value below the median for the cohort in many white matter tracts, including the corticospinal tracts and the corpus callosum, was a better predictor of later CP than structural MRI.

We also evaluated brain activation and functional connectivity during a passive motor task using fMRI. Performing neonatal task-based fMRI is challenging due to limitations of passive tasks available in the neonatal population and the necessity for the neonate to remain asleep and motion-free during the MRI. Several groups have attempted to use passive sensorimotor experiments in neonates. Erberich ⁴¹ found that BOLD activation was inconsistent in preterm infants studied with the passive sensorimotor task. Two infants had activation in the primary sensory and motor areas and the remaining four infants showed activation in the early supplemental motor area. Using a custom-made device to allow for precise sensorimotor stimulation of the hand, Arichi et al ⁴² found well-localized positive BOLD signal activation in the primary somatosensory cortex in the majority of preterm infants scanned, even those scanned prior to term equivalent age. Heep et al ⁴³ studied 8 preterm infants at term-equivalent age with a sensorimotor stimulation paradigm using a passive forearm extension-flexion movements by manual traction. They found bilateral, predominantly negative, activation in the sensorimotor cortex in 9 of 10 trials. Arichi et al also evaluated infants with periventricular hemorrhagic infarction who later developed hemiparesis as well as 3 control preterm infants with normal motor outcomes with a passive sensorimotor task. In infants with periventricular hemorrhagic infarction, afferent thalamocortical tracts appeared to have developed compensatory trajectories which circumvented areas of damage. In contrast, the corticospinal tracts showed marked asymmetry following focal brain injury. Sensorimotor network analysis suggested that interhemispheric functional connectivity was largely preserved despite brain injury.

In the present study, we did not find consistent group activation during the sensorimotor task, perhaps due to the technical limitations of the task (e.g. subject movement during the bulb suction stimulation). We also used bilateral stimulation instead of unilateral stimulation due to the heterogeneity of our population (some infants with unilateral stroke) and this may have affected the overall group activation. We did find differences in functional connectivity between groups after we regressed task-related activation out of the fMRI data and then evaluated temporal correlations in the residual data, which is more analogous to evaluating resting-state functional connectivity. Linke et al¹³ evaluated resting-state functional connectivity in 26 term and preterm infants with perinatal brain injury and 27 infants without injury. They found that disruption to connectivity of the somatomotor and frontoparietal executive networks predicted motor impairment at 4 and 8 months better than other clinical measures, including structural MRI score. This study and our results suggest that functional connectivity MRI may be a promising predictor of later motor outcomes.

The strengths of our study include high quality data acquired at 3T using DTI with 61 directions, robust image processing methods, and objective assessment of microstructure with TBSS. The main limitations of our study include small sample size and heterogeneous etiology of brain injury. Given the small sample size, we were unable to control for potentially confounding perinatal variables in the analysis, although the CP and non-CP groups did not differ in the variables we evaluated. Despite the heterogeneity of the group, we still found significant differences in both DTI and functional connectivity, suggesting that these MRI methods may prove to be powerful tools for predicting later motor outcome independent of the etiology of injury. In conclusion, we have shown that infants with perinatal brain injury who later develop CP have altered structural and functional connectivity within the motor networks in the neonatal period as compared to infants with brain injury with normal motor outcomes.