Microstructural Maturation of White Matter Tracts in Encephalopathic Neonates

Abstract

Purpose

To apply neurite orientation dispersion and density imaging (NODDI) to measure white matter microstructural features during early development.

Methods

NODDI parameters were measured in twelve newborns and thirteen 6 month infants, all with perinatal clinical encephalopathy.

Results

Between 0 and 6 months, there were significant differences in fractional anisotropy (FA) for all tracts; in neurite density for internal capsules, optic radiations, and splenium; and in orientation dispersion for anterior limb of internal capsule and optic radiations. There were no appreciable differences in NODDI parameters related to outcome.

Conclusion

NODDI may allow more detailed characterization of microstructural maturation than FA.

1. INTRODUCTION

MRI has proved an invaluable tool for assessing the developing neonatal brain. However, MRI may be apparently normal even in neonates with clinically obvious signs of encephalopathy [1], presumably reflecting the presence of microscopic injury, particularly in white matter tracts that are most susceptible to disruption by early neurological insult [2–7]. Better imaging tools are thus needed to identify neonatal brain injury and abnormal maturation at a microscopic scale.

Much of the work to develop such tools has centered on diffusion tensor imaging (DTI). Unfortunately, conventional processing of DTI data yields metrics such as fractional anisotropy (FA) that are sensitive to microstructure but cannot readily discriminate between the different components of microstructure [8–10]. A newer DTI method termed neurite orientation dispersion and density imaging (NODDI) can better discriminate microstructural features such as the density of axons and dendrites (collectively, “neurites”), fiber tract organization, and free water content [11–13].

In this study, we use NODDI to characterize white matter microstructural parameters in encephalopathic neonates born at term. Specifically, we measure neurite density (ND), orientation dispersion (OD), and free water fraction (FW) in major white matter tracts at birth and at six months of life, discuss how differences in NODDI parameters vary by tract, and compare NODDI parameters between infants with different neurological outcomes.

2. METHODS

2.1. Subjects

The study was initiated following approval by the institutional review board. Inclusion criteria for enrollment were: (a) birth at or after 36 weeks gestation; (b) moderate to severe encephalopathy identified by an attending neonatologist or pediatric neurologist; and (c) Apgar score less than 5 at 10 minutes of life, prolonged resuscitation at birth, pH of less than 7.0 in any blood sample taken within 60 minutes of birth, or base deficit greater than 12 mmol/L in any cord or arterial blood sample taken within 60 minutes of birth. Exclusion criteria were: (a) evidence of in utero or perinatal infection, (b) major anomalies of the brain or other major organ systems, or (c) evidence of congenital metabolic disease. For ethical and feasibility reasons, normal controls (i.e., infants born without encephalopathy) were not enrolled in the study.

Written informed consent was obtained for all subjects to be enrolled in a program of clinical neurological follow up and undergo brain MRI including research sequences. A total of twenty-five infants underwent brain imaging with a NODDI-compatible protocol (Table 1); twelve of these subjects were imaged in the first week of life and thirteen were imaged at six months of life. Clinical neuromotor development was assessed using the last documented neurological examination, with each infant being assigned a validated neuromotor score [14]—0 for normal development, 1 for abnormal tone or reflexes, 2 for abnormal tone and reflexes, 3 for functional deficit of power, 4 for cranial nerve involvement with motor abnormality, and 5 for spastic quadriparesis. All infants in this study had neuromotor scores of 0 or 1 an average of 318 days after birth. Longitudinal imaging evaluation with serial MRIs was not considered appropriate in these infants with no or mild neurological impairment.

Table 1.

Subject characteristics. Standard deviations are indicated in parenthesis.

	0 months	6 months
Corrected birth age (days)	−6.4 (10.5)	−4.4 (7.4)
Gender	55% male	46% male
5 minute Apgar	4.6 (1.6)	4.5 (2.2)
Seizures	18%	8%
Neuromotor score	0.36 (0.50)	0.23 (0.44)

2.2. Data acquisition and processing

MR imaging was performed on 3.0 Tesla GE magnets (Discovery MR750 or Signa HDxt, General Electric Medical Systems, Milwaukee, WI, USA) using a commercially available 8-channel head coil. Infants were sedated during image acquisition with intravenous propofol under the supervision of a trained anesthesiologist. Diffusion MRI data was acquired in each subject using a two-shell high angular resolution diffusion imaging (HARDI) protocol with ASSET acceleration factor 2, and spatial resolution of 2 × 2 × 2 mm. Inner shell data were acquired in 30 directions with b=700 s/mm², TR=5000 ms, and TE=62.1–68.3 ms in the 0 month group, and b=700 s/mm², TR=8000–12500 ms, and TE=68.8–80.5 ms in the 6 month group. Outer shell data were acquired in 55 directions with b=2000 s/mm², TR=6000 ms, and TE=77.1–85.9 ms in the 0 month group, and b=2000 s/mm² or 3000 s/mm², TR=11000–14500 ms, and TE=87.3–109.5 ms in the 6 month group. Each HARDI shell was supplemented by an additional b=0 s/mm² volume acquired using the same TR and TE. This supplemental volume was used to normalize the HARDI data in order to compensate for variable TR and TE. Total HARDI acquisition time was approximately 3 minutes for the inner shell and 5 minutes for the outer shell.

Diffusion data were corrected for motion and eddy current artifacts using the FMRIB Software Library [15]. Inner shell HARDI data were used to calculate FA maps, after which the complete set of two-shell HARDI data were processed using the NODDI Matlab toolbox (University College London) [11]. NODDI analysis yielded whole-brain maps of OD, ND, and FW.

2.3. ROI analysis

ROIs were manually drawn on axial and sagittal FA maps by a neuroradiologist within the anterior and posterior limbs of the internal capsule (ALIC and PLIC); optic radiations; and genu, body, and splenium of the corpus callosum (Figure 1). This neuroradiologist was blinded to the NODDI maps during ROI selection. ROIs could not be drawn in one neonate due to excessive motion artifact, reducing to twenty-four the number of subjects included in the analysis. In these remaining subjects, ROIs were automatically translated to co-registered NODDI parameter maps, and mean values of FA, OD, ND, and FW were calculated within each ROI for each subject.

Figure 1.

Fractional anisotropy maps with superimposed ROIs corresponding to (a) anterior and posterior limbs of the internal capsule, (b) optic radiations, (c) genu and splenium of corpus callosum, and (d) body of corpus callosum.

Statistical analysis of ROI data was performed in Matlab 2013a (The Mathworks Inc., Natick, MA). Groupwise analysis of differences between was performed using Student’s two-tailed t test. Statistical significance in the face of multiple comparisons was determined using the Benjamini-Hochberg method and a false discovery rate of 5% over 24 independent comparisons (4 parameters in each of 6 ROIs) [16]. Statistical comparison of the corpus callosum to ALIC, PLIC, and optic radiations at birth was also performed using Student’s two-tailed t test.

3. RESULTS

Microstructural maps for two typical subjects are shown in Figure 2. Qualitatively, maps of OD depict the known locations of highly organized white matter tracts and show correspondingly low values in these regions even at birth, as would be expected given the reciprocal relationship between OD and FA. ND demonstrates comparatively less heterogeneity across each image, though there is slightly higher ND in the subcortical white matter shortly after birth that disappears by six months of age. Maps of FW show large values within the ventricles and subarachnoid spaces, with very low values within the brain substance at term.

Figure 2.

NODDI parameter maps in two representative subjects, one shortly after birth (top row) and one at 6 months (bottom row). Images are scaled to similar size. FA = fractional anisotropy, OD = orientation dispersion, ND = neurite density, FW = free water fraction.

Microstructural indices—FA, OD, ND, and FW—were measured within each selected white matter tract (Table 2). At 0 months, the corpus callosum collectively demonstrated significantly higher FA than the ALIC or optic radiations (p<0.05). Moreover, the corpus callosum demonstrated significantly lower OD and higher FW than the PLIC, ALIC, or optic radiations at birth (p<0.05).

Table 2.

NODDI microstructural parameters at 0 months and 6 months. Standard deviations are indicated in parenthesis. Asterisks denote statistical significance.

	Tract†	0 months	6 months	p
FA	PLIC	0.443 (0.049)	0.602 (0.049)	<0.001*
	ALIC	0.306 (0.042)	0.445 (0.084)	<0.001*
	OR	0.347 (0.042)	0.447 (0.047)	<0.001*
	CC genu	0.511 (0.060)	0.672 (0.048)	<0.001*
	CC body	0.492 (0.060)	0.578 (0.035)	<0.001*
	CC splenium	0.597 (0.053)	0.770 (0.044)	<0.001*
ND	PLIC	0.287 (0.033)	0.435 (0.037)	<0.001*
	ALIC	0.208 (0.038)	0.354 (0.028)	<0.001*
	OR	0.179 (0.063)	0.329 (0.039)	<0.001*
	CC genu	0.270 (0.107)	0.302 (0.054)	0.392
	CC body	0.221 (0.060)	0.255 (0.049)	0.149
	CC splenium	0.315 (0.104)	0.402 (0.042)	0.021*
OD	PLIC	0.097 (0.027)	0.109 (0.021)	0.093
	ALIC	0.131 (0.052)	0.168 (0.043)	0.014*
	OR	0.108 (0.051)	0.160 (0.025)	<0.001*
	CC genu	0.043 (0.048)	0.054 (0.024)	0.492
	CC body	0.048 (0.048)	0.048 (0.024)	0.982
	CC splenium	0.049 (0.060)	0.042 (0.019)	0.717
FW	PLIC	0.000 (0.000)	0.000 (0.000)	1.000
	ALIC	0.000 (0.000)	0.000 (0.000)	0.163
	OR	0.001 (0.001)	0.000 (0.000)	0.011*
	CC genu	0.127 (0.097)	0.015 (0.023)	0.003*
	CC body	0.051 (0.048)	0.039 (0.052)	0.572
	CC splenium	0.080 (0.071)	0.011 (0.012)	0.009*

^†OR = optic radiation, CC = corpus callosum.

Measured FA was significantly higher at 6 months than at 0 months within all measured white matter tracts. In contrast, ND was significantly higher only within the PLIC, ALIC, optic radiations, and splenium. Furthermore, OD was significantly higher only within the ALIC and optic radiations. Measured FW was significantly lower at 6 months than at 0 months in the optic radiations, genu, and splenium.

To identify potential relationships between imaging and neurological outcomes, FA and NODDI indices at 6 months were further assessed using subgroup analysis based on neuromotor outcomes (Table 3). There were no statistically significant differences in FA or NODDI parameters related to neuromotor scores.

Table 3.

NODDI microstructural parameters at 0 months in infants with neuromotor scores (NMS) of 0 or 1. Standard deviations are indicated in parenthesis.

	Tract†	Normal‡	Encephalopathic
			NMS = 0	NMS = 1	p
FA	PLIC	0.50 (0.02)	0.45 (0.04)	0.42 (0.05)	0.150
	ALIC	0.37 (0.04)	0.30 (0.04)	0.31 (0.04)	0.533
	OR	0.32 (0.03)	0.36 (0.04)	0.33 (0.04)	0.083
	CC genu	0.54 (0.05)	0.52 (0.05)	0.49 (0.07)	0.466
	CC body	0.46 (0.05)	0.50 (0.06)	0.47 (0.06)	0.441
	CC splenium	0.61 (0.03)	0.60 (0.05)	0.59 (0.06)	0.714
ND	PLIC	0.39 (0.03)	0.28 (0.03)	0.29 (0.05)	0.710
	ALIC	0.30 (0.03)	0.21 (0.03)	0.20 (0.05)	0.793
	OR	0.23 (0.03)	0.17 (0.03)	0.19 (0.10)	0.765
	CC genu	0.38 (0.03)	0.28 (0.09)	0.26 (0.15)	0.859
	CC body	0.45 (0.06)	0.20 (0.04)	0.26 (0.08)	0.226
	CC splenium	0.45 (0.03)	0.29 (0.04)	0.35 (0.17)	0.566
OD	PLIC	0.17 (0.01)	0.09 (0.02)	0.11 (0.03)	0.040
	ALIC	0.22 (0.02)	0.13 (0.05)	0.13 (0.06)	0.809
	OR	0.26 (0.03)	0.10 (0.02)	0.12 (0.08)	0.459
	CC genu	0.15 (0.02)	0.03 (0.01)	0.07 (0.08)	0.438
	CC body	0.20 (0.04)	0.04 (0.04)	0.06 (0.06)	0.600
	CC splenium	0.13 (0.02)	0.03 (0.02)	0.08 (0.10)	0.385
FW	PLIC	0.01 (0.01)	0.00 (0.00)	0.00 (0.00)	1.000
	ALIC	0.01 (0.01)	0.00 (0.00)	0.00 (0.00)	0.307
	OR	0.02 (0.02)	0.00 (0.00)	0.02 (0.03)	0.205
	CC genu	0.18 (0.05)	0.17 (0.09)	0.05 (0.03)	0.015
	CC body	0.26 (0.11)	0.06 (0.05)	0.03 (0.02)	0.161
	CC splenium	0.17 (0.04)	0.11 (0.07)	0.03 (0.02)	0.035

^†OR = optic radiation, CC = corpus callosum.

^‡Normal reference data from Ref. [37].

4. DISCUSSION

Previous studies have attempted to characterize the microstructural changes of maturation by measuring the evolution of FA or related anisotropy metrics that are known to be dependent on tissue microstructure [2, 17–20]. However, the relationship between FA and underlying microstructure is complex and influenced by a number of factors, including fiber orientation, axonal diameter, myelination, free water content, and cellular infiltrates in response to infection or injury [21–24]. As such, the insights that can be attained through FA alone are limited.

Several groups have developed alternate DTI-based metrics that have different sensitivities to various microstructural features compared to FA [19, 24, 25]. In combination with FA, these metrics confer an improved ability to disentangle the competing effects of multiple underlying microstructural changes on the overall diffusion signal. In so doing, they have clarified the timing of maturational changes, including identification of a “premyelination” phase that precedes myelin-related changes on T1- or T2-weighted imaging [7, 26]. By comparison, NODDI uses a three-compartment tissue model to map the observed diffusion signal directly into parameters of OD, ND, and FW [11]. These parameters are more intuitively appealing as markers of tissue architecture, and emerging evidence suggests that they accurately reflect histological features of the underlying brain [12, 13, 27].

To our knowledge, no other studies have yet described changes in NODDI parameters in the immediate postnatal phase. Despite significantly higher FA in all measured white matter tracts at the end of the first six months of life, we found corresponding significant ND differences only within the PLIC, ALIC, optic radiations, and splenium, and significant OD differences only within the ALIC and optic radiations. Observed differences in FW in the corpus callosum are presumed to reflect smaller partial volume effects in the larger, 6 month old brain.

We speculate that these tract-to-tract differences in NODDI parameter changes reflect more basic processes of physiological and structural development. MRI and histological studies have consistently shown an ordered sequence of white matter myelination, involving the PLIC and optic radiations at or near term, the splenium and ALIC within 1–4 months post-term, and the body and genu of the corpus callosum beginning at approximately 6 months post-term [28–32]. The magnitude of ND increase we measured in each tract—large in the PLIC, optic radiations, and ALIC; intermediate in the splenium; and small in the genu and body—matches this known myelination sequence. In this context, the small ND change in the genu and body presumably reflects the comparatively early stage of myelin maturation in these structures at 6 months, which marks the end of our observational period. These observations also corroborate earlier results that suggest that FA increases in the splenium occur earlier than in the remainder of the corpus callosum, presumably due to relatively later maturation of myelin in the genu and body [33, 34].

One of the primary motivations for studying NODDI in encephalopathic infants is the hypothesized ability to more accurately detect and characterize subtle microstructural abnormalities that may occur in this population. Applying NODDI in these patients shortly after birth may thus improve our understanding of the biophysical basis of perinatal injury and ability to predict eventual neurological outcome. Recently, Lally and colleagues showed that term infants with moderate or severe encephalopathy at birth exhibited significantly lower FA and radial diffusivity in white matter tracts compared to term infants with no or mild encephalopathy [35, 36]. However, using a single-shell NODDI model, the authors demonstrated that these differences in FA and radial diffusivity were the result of lower white matter ND in encephalopathic infants, as there was no appreciable difference in OD between these groups [35, 36]. Thus, while DTI-based metrics such as FA may indicate the presence of a microstructural abnormality, NODDI-derived parameters can add important new insights by resolving these abnormalities into tissue component features (e.g., ND and FW) and geometrical organization (e.g., OD) [12, 37, 38]. The importance of decreased ND in encephalopathic infants has been underscored by subsequent work demonstrating that decreased ND—but not OD—is predictive of poorer neurological outcomes [39]. Aside from elucidating the nature of injury, these results offer promise for improved neurodevelopmental prognostication shortly after birth.

Our outcomes-based subgroup analysis did not demonstrate any discernible differences in NODDI parameters between encephalopathic infants with normal neurological development (neuromotor score of 0) and encephalopathic infants with mild neurological impairment (neuromotor score of 1), suggesting that microstructural differences between these two groups is small. Encephalopathic infants with more severe neurological impairment (neuromotor scores of 2–5), which were not represented in our study, may exhibit more obvious differences in NODDI parameters—presumably decreased ND without significantly altered OD, as in recent work using a single-shell NODDI model [35, 36, 39]—and thus may be a useful target population for future studies.

Curiously, our data provide some evidence that the presence of encephalopathy may itself affect NODDI parameters, regardless of ultimate neurological outcome. In particular, the OD and ND values that we measure in infants with encephalopathy appear considerably lower than in normal infants [37], while corroborating the results of Lally and colleagues suggesting that OD in encephalopathic neonates is likely to be less than 0.1 in areas such as the corpus callosum with FA above 0.4 [36]. Though minor differences in ROI selection and DTI acquisition can influence the accuracy of NODDI and FA estimates in different patient groups, the fact that our measurements of FA in encephalopathic patients—measurements which are derived from the same DTI source data and ROIs used for NODDI analysis—are in good agreement with published FA values from non-encephalopathic infants [18, 19, 34, 37] implies that the actual influence of these technical factors is minimal, and therefore, that observed differences in NODDI parameters are reliable.

A limitation of this study is that the ages of the subjects are not uniformly distributed between 0 and 6 months, but rather are clustered around these two time points. This approach is appropriate to gain a general understanding of microstructural development but obscures the multiphasic temporal evolution that has been noted in other studies [25, 34, 40, 41]. A second limitation was that imaging was performed with acquisition parameters chosen to maximize signal-to-noise ratio and minimize acquisition time. For example, we changed the outer shell b value partway through the study in order to increase diffusion signal. This change was considered unlikely to affect NODDI estimates, as Zhang et al. have shown that different two-shell acquisition protocols with outer shell b value ranging from 2000–3000 s/mm² yield very similar NODDI results [11]. In addition, we preferred to use the minimum TE for each diffusion acquisition and allowed minor variation of this parameter between patients. The effect of these variations on the resulting diffusion signal were offset by normalizing each diffusion volume by a b=0 s/mm² volume acquired with the same TE. Variations in TR were related to the number of slices needed for complete coverage of the brain. A third limitation is that encephalopathic infants with moderate or severe neurological impairment were not studied, which may obscure microstructural abnormalities that are normal or only very minimally abnormal in the less impaired population of infants included in our analysis.

These limitations notwithstanding, we have demonstrated differences in NODDI microstructural parameters during the first six months of life. Regional variation in observed NODDI parameters differences may correspond to known differences in the timing of myelination between these structures. Thus, NODDI or other related methods that can accurately characterize tissue microstructure may prove useful for characterizing subtle brain injury, and in so doing may improve our ability to predict neurological outcomes in infants that have experienced early neurological insult.

5. CONCLUSION

We have characterized differences in NODDI microstructural indices in major white matter tracts of encephalopathic term neonates during the first six months of life. Despite higher FA in all measured white matter tracts at 6 months of life, differences in NODDI indices between 0 and 6 months demonstrate tract-to-tract variability that may reflect regional variation in myelination over this time period. NODDI parameters were similar in encephalopathic infants with normal neurological development and those with mild neurological impairment. Methods such as NODDI may hold promise for detecting and characterizing abnormal microstructural maturation and may lead to improved understanding of the nature of early neurological insult.

Highlights.

• In all white matter tracts, fractional anisotropy changed between 0 and 6 months.

• Neurite density and orientation dispersion only changed in some tracts.

• No parameter differences were found between mildly impaired and unimpaired infants.

• These patterns of change may reflect the known timing of myelination in each tract.