Differentiating T2 hyperintensity in neonatal white matter by two-compartment model of diffusional kurtosis imaging

In conventional neonatal MRI, the T2 hyperintensity (T2h) in cerebral white matter (WM) at term-equivalent age due to immaturity or impairment is still difficult to identify. To clarify such issue, this study used the metrics derived from a two-compartment WM model of diffusional kurtosis imaging (WM-DKI), including intra-axonal, extra-axonal axial and radial diffusivities (Da, De,// and De,⊥), to compare WM differences between the simple T2h and normal control for both preterm and full-term neonates, and between simple T2h and complex T2h with hypoxic-ischemic encephalopathy (HIE). Results indicated that compared with control, the simple T2h showed significantly increased De,// and De,⊥, but no significant change in Da in multiple premyelination regions, indicative of expanding extra-axonal diffusion microenvironment; while myelinated regions showed no changes. However, compared with simple T2h, the complex T2h with HIE had decreased Da, increased De,⊥ in both premyelination and myelinated regions, indicative of both intra- and extra-axonal diffusion alterations. While diffusion tensor imaging (DTI) failed to distinguish simple T2h from complex T2h with HIE. In conclusion, superior to DTI-metrics, WM-DKI metrics showed more specificity for WM microstructural changes to distinguish simple T2h from complex T2h with HIE.


Results
Subjects. The study design flowchart is shown in Fig. 2. Of 423 neonates reviewed, 132 neonates with the PMA of 37-42 weeks had received conventional MRI and DKI. Further 13 neonates were excluded from the Considering their critical roles in WM development and injury, the axons, myelin sheath and oligodendrocyte precursors (pre-OLs) were determined as the main elements of our hypothesis model. As shown in left column, myelination process is mainly depicted by the proliferation of oligodendrocytes lineage precursors and ensheathment of oligodendroglial processes around the axons. As shown in middle column, in case of simple T2h, delayed myelination occurred (presented as the loss of pre-Ols). Thus, compared with control neonates, changes of WM-DKI metrics only presented in extra-axonal spaces rather than intra-axonal spaces at premyelination stage. As shown in right column, in condition of complex T2h with HIE, we hypothesized there existed additional axonal damages, which could lead to the intra-axonal diffusivity decrease in WM at both myelinated and premyelination stages. Beyond, HIE may induce the destructive axons, pre-OLs loss, vasogenic edema and aggregation of microglia and astrocytes; the compound of such changes might lead to the increased extra-axonal radial diffusivity and unchanged or slightly increased extra-axonal diffusivity.
analysis due to congenital metabolic disorder (n = 2), congenital infection (n = 3) or motion artifact (n = 8). Thus, 92 full-term neonates with gestational age (GA) of 39.18 ± 1.28 weeks and PMA at MRI scan of 40.55 ± 1.45 weeks and 27 preterm neonates with GA of 35.08 ± 2.07 weeks and PMA at MRI scan of 38.58 ± 1.73 weeks were enrolled for MRI interpretation. Following discussion and consultation, 21, 41 and 27 full-term neonates, and 8, 11 and 8 preterm neonates, were allocated respectively to the normal control group, simple T2h group and complex T2h with other abnormality group (detailed gestational ages and PMA at MRI scan in each group were shown on Table 1); 5 full-term neonates with GA of 39.20 ± 1.87 weeks and PMA at MRI scan 40.51 ± 2.01 weeks from complex T2h with other abnormality group were further allocated to the complex T2h with HIE group. The rest 22 full-term neonates and 8 preterm neonates in complex T2h with other abnormality group, and 3 full-term neonates presenting with punctate WM lesions without T2h were excluded from DKI data analysis.
In preterm and full-term neonates, the proportions of T2h were 70.4% (19 of 27) and 73.9% (68 of 92), and the proportions of simple T2h were 40.7% (11 of 27) and 44.6% (41 of 92), respectively. The proportions of T2h and simple T2h did not differ significantly between preterm and full-term neonates (P = 0.81, 0.83, respectively). The demographic parameters and clinical details of neonates in control group, simple T2h group and complex T2h with HIE group are shown in Table 1. There were no significant differences in the GA, PMA, birth weight (BW), age at the time of MR scan, presence of small size for GA, gender, or multiple births between the preterm simple T2h and control groups, full-term simple T2h and control groups, or full-term simple T2h and complex T2h with HIE groups. Compared with the full-term simple T2h group, the preterm simple T2h group had significantly lower mean GA, PMA and BW, and a significantly higher age at the time of MR scan.

Tract-based spatial statistics (TBSS) analysis. Part I. Differences in WM-DKI metrics between
the simple T2h and control groups. As shown in Fig. 3, the preterm simple T2h group showed significantly decreased FA, significantly increased MD, AD, RD, D e,// and D e,⊥ in the superior corona radiata (SCR) and subcortical deep WM regions of superior frontal gyrus, superior occipital gyrus, precentral gyrus and postcentral gyrus, etc , and no change in D a compared with the control group. In comparisons between the full-term simple T2h and control groups; besides of above regions, similar results were observed in more widespread WM regions, including posterior thalamic radiation (PTR), corpus callosum (CC) and external capsule (EC), etc. However, for both preterm and full-term simple T2h groups compared with corresponding control groups, no differences were observed mainly in the myelinated WM regions at TEA, such as the cerebral peduncle (CP) and posterior limb of the internal capsule (PLIC). The metric changes in the simple T2h group reflect increased diffusivity from the extra-axonal space, which was compatible with delayed myelination (illustrated in Fig. 1). When these diffusional metrics were compared directly between the preterm and full-term simple Regions colored light blue-blue represent significantly decreased voxels (P < 0.05) in the simple T2h group compared with the control group, while regions colored yellow-red represent significantly increased voxels (P < 0.05) in the simple T2h group compared with the control group. These have been overlaid on the mean FA template with a mean skeleton (green). Compared with the control group, the simple T2h group showed decreased FA and increased MD, AD, RD, D e,∥ and D e,⊥ in multiple premyelination regions in both preterm and full-term neonates, but no differences were observed in myelinated regions (such as the cerebral peduncle and posterior limb of the internal capsule). Moreover, D a showed no significant change.
Scientific RepoRts | 6:24473 | DOI: 10.1038/srep24473 T2h groups (with GA, PMA and BW as covariates for correction), no differences were found (TBSS analysis not shown), indicating no essential differences between preterm and full-term neonates with the simple T2h.
Part II. Differences in WM-DKI metrics between the simple T2h and complex T2h with HIE groups. Compared with the full-term simple T2h group, the complex T2h with HIE group exhibited significantly decreased D a and FA in the CP, PLIC, PTR, SCR, CC and EC, with mainly increases in MD, RD, and D e,⊥ (Fig. 4). The involvement of the CP and PLIC in these changes indicates the extensive axonal damage occurring in both myelinated and premyeliantion WMs in the neonates of HIE, which were different changes from the simple T2h (illustrated in Fig. 1).

ROI analysis.
The PLIC, SCR, CC and EC, which respectively represented projection fibers, commissural fibers and association fibers respectively, were selected as target ROI regions. The results of the ROI analysis (Tables 2-5) were almost entirely consistent with those of the TBSS analysis. D a was identified as distinguishing between the simple T2h and complex T2h with HIE groups in regions that represented myelinated (the PLIC) and premyelination fibers (the SCR). In addition, all parameters for the PLIC showed no significant differences between the simple T2h and control groups in both preterm and full-term neonates.

Discussion
In this study, an advanced technique employing a two-compartment WM-DKI model was used to explore the diffusion variation underlying T2h in the neonatal WM. The results provide the novel insights into the possible abnormal changes underlying simple T2h, which were similar between preterm and full-term neonates. Furthermore, above findings of WM-DKI metrics suggest that simple T2h and complex T2h with HIE may originate from differing WM microstructural changes. With respect to simple T2h, previous DTI studies 7,23-25 have also observed lower FA and higher MD, AD and RD in neonates. However, due to the non-specificity of DTI measures for the intra-and extra-axonal diffusion, it is still difficult to determine what structure alters on earth. Our results regarding the simple T2h group observed the changes of DKI-WM metrics mainly in multiple premyelination regions hinting that the elevated diffusivity was from the extra-axonal space rather than the intra-axonal space; besides, such abnormality was absent in highly anisotropic regions (such as the PLIC and CP, etc ) where show almost complete myelination at the TEA 32 . A more recent DTI study 25 about T2h (DEHSI) also observed similar patterns of spatial distribution. Specifically, higher diffusivity values and lower FA were found in centrum semiovale and OR, while CP and CST showed no difference between those with or without DEHSI at TEA. It is well known that the normal premyelination process is closely linked to a decrease in brain water content and the proliferation of oligodendrocyte lineage precursors, which showed an overall decrease in diffusivity and increase in FA with age 14 . Hence, the extra-axonal diffusional changes in the opposite way (increase in extra-axonal diffusivity and decrease in FA ) observed in simple T2h were just compatible with the delayed myelination, which might result from a sparsity of oligodendrocytes and Regions colored light blue-blue represent significantly decreased voxels (P < 0.05) in the complex T2h with HIE group compared with the full-term simple T2h group, while regions colored yellow-red represent significantly increased voxels (P < 0.05). These have been overlaid on the mean FA template with a mean skeleton (green). D a and FA were significantly decreased in both myelinated and premyelination WM regions in the complex T2h with HIE group compared with the full-term simple T2h group, and MD, RD and D e,⊥ were increased to differing extents.
Scientific RepoRts | 6:24473 | DOI: 10.1038/srep24473 a relatively elevated free water content in the extracellular matrix (as illustrated Fig. 1). This speculation was also supported by an animal experiment 22 , in which they found non-necrotic WM injury mainly led to myelination failure but without axonal degeneration. Additionally, we found no obvious differences in the microstructural changes of preterm and full-term neonates with simple T2h, suggesting that these two populations possess similar microstructural alterations.
Targeting the complex T2h with HIE, lower FA and slightly higher MD and RD were found than those with simple T2h. Although these, it was not sufficiently specific to differentiate these two groups since similar diffusivity changes were observed between simple T2h and controls. Notably, being different from simple T2h, D a was markedly decreased in both myelinated and premyelination WM regions in complex T2h with HIE. These observations may hint an extensive breakdown of axons due to these severe WM damage, which was also demonstrated by previous histological studies 22,33 . All these may suggest D a to be a specific biomarker for distinguishing these two abnormalities. A similar change in D a has been reported in a study of acute and subacute ischemic lesions 31 , supporting D a as a sensitive and specific biomarker of axonal abnormalities. Furthermore, these findings suggest that the microstructural changes can be more severe in complex T2h, partly explaining the varied neurodevelopmental outcomes observed previously 1-6 .
There were some limitations to this study. First, there is unavoidable subjectivity in the screening of T2h in T2WIs, as reported previously 11 . Although ADC and T2 values have been proposed to improve the accuracy of diagnosing T2h 23,24,34 , we did not use these approaches due to the lack of a definite and unified standard. However, the inter-and intra-observer agreement in this study (Kappa values 0.74 and 0.82) were receivable in repeatability test. Second, the sample size for neonates with HIE was small due to the low incidence of this injury type at our institution. Moreover, the number of preterm neonates with scans at TEA was also low due to parental concerns. Therefore the inclusion criteria for PMA can be easily met in full-term neonates, the high incidence of T2h in     this population may reflect selection bias rather than the real situation. Third, the absence of follow-up for our enrolled neonates necessitates to further validation of our hypothesis.
In conclusion, through using two-compartment WM-DKI metrics, this study provided series of interesting findings about the underlying microstructural changes of T2h: (1) for both preterm and full-term neonates with simple T2h, increased extra-axonal diffusivity and unchanged intra-axonal diffusivity in multiple premyelination WM regions may hint the delayed myelination; (2) while, T2h with HIE showed markedly decreased intra-axonal diffusivity which may be related to axonal damage; (3) superior than conventional DTI metrics, WM-DKI metrics are more specific for identifying the WM microstructural changes (e.g. intra-and extra-axonal diffusivity) in developmental brain.

Materials and Methods
This single-center retrospective study was approved by the Ethics Committee of the First Affiliated Hospital of Xi'an Jiaotong University, and the written parental consent was obtained prior to scanning. This study was conducted in accordance with the Declaration of Helsinki.
Subjects. All neonates were consecutively collected from our institution's neonatal intensive care unit, from December 2010 to February 2014. The neonates that had received conventional MR imaging and DKI at the PMA of 37-42 weeks were included. Any neonates who diagnosed as congenital metabolic disorder, malformation or infection, or had poor MR image quality were excluded from this study. Demographic parameters and clinical details of all the neonates were reviewed and recorded by an experienced neonatal radiologist (Q.L.S. with 10 years of related experience ).   Table 4. Comparisons of DKI-WM metrics of the corpus callosum among study groups. Note: data presented as means ± SDs. AD, axial diffusivity; D a , intra-axonal diffusivity; D e,// , extra-axonal axial diffusivity; D e,⊥ , extra-axonal radial diffusivity; FA, fractional anisotropy; MD, mean diffusivity; RD, radial diffusivity.   Table 5. Comparisons of DKI-WM metrics of the external capsule among study groups. Note: data presented as means ± SDs. AD, axial diffusivity; D a , intra-axonal diffusivity; D e,// , extra-axonal axial diffusivity; D e,⊥ , extra-axonal radial diffusivity; FA, fractional anisotropy; MD, mean diffusivity; RD, radial diffusivity.
Scientific RepoRts | 6:24473 | DOI: 10.1038/srep24473 MR data acquisition. All MR images were obtained using a 3.0 Tesla scanner (Signa HDxt, General Electric Medical System, Milwaukee, WI, USA) equipped with an 8-channel phase array radio-frequency head coil. Chloral hydrate (50 mg/kg) was administered orally for sedation in 30 minutes before the scan. The neonate was kept warm with swaddle and the head immobilized by molded foam. Micro-earplugs and earmuffs were used to protect the hearing. During imaging, the neonate was monitored by the attending pediatrician, and vital signs were monitored continuously. Three-dimensional fast spoiled gradient-recalled echo T1WIs (repetition time /echo time, 10/4.6 ms) and fast spin-echo T2WIs (repetition time/echo time, 4200/116.4-118.9 ms) were obtained. DKI was acquired using the following parameters: 18 directions; b value = 0, 500, 1000 and 2000 s/mm 2 ; repetition time/echo time= 8000/100.2-117.7 ms; slice thickness of 4 mm; field of view, 180 × 180 mm 2 ; matrix, 256 × 256; and acquisition time of 8 minutes 42 seconds. The total scan time was less than 30 minutes. MR interpretation. MR abnormalities (i.e. T2h and other abnormalities) were screened according to the clinical MR standards 2 . Wherein, T2h was defined as visually higher signal intensity in periventricular and/or subcortical WM than in normal unmyelinated WM on T2WIs; while the "anterior caps" and "posterior arrowheads" would be excluded due to its common high signal intensity 12 . The other abnormalities included cystic encephalomalacia, punctate lesions, a loss of volume of the periventricular WM and corpus callosum, the change of gray matter signal intensity, the widening of the subarachnoid space and intraventricular hemorrhage.
Neonates without any MR abnormalities and no evidences of any clinical episodes that might cause cerebral damage or delayed maturation were allocated to the control group. Neonates with T2h but no above abnormalities were allocated to the simple T2h group. Meanwhile, neonates with both T2h and above abnormalities were allocated to the complex T2h group. Furthermore, in the complex T2h group, the full-term neonates fulfilled both MRI 35 and clinical diagnostic criteria for HIE 36 would be allocated to the complex T2h with HIE group. Specifically, the MR diagnostic criteria refers to the focal or diffuse abnormal signal intensities in the bilateral globus pallidus, putamen or thalamus on conventional MRI 35 . And the clinical diagnostic criteria for HIE 36 includes: A pH ≤ 7.0 or a base deficit ≥ 16 mmol/L on umbilical cord blood or any postnatal blood sample within 1 hour of age; or history of an acute perinatal event and either no blood gas available, or a pH from 7.01 to 7.15 or a base deficit from 10 to 15.9 mmol/L, with a 10-minute Apgar score ≤ 5, or assisted ventilation initiated at birth and continued for at least 10 minutes. The rest neonates in the complex T2h group, and neonates without T2h but with any other abnormalities were excluded.
Two experienced pediatric radiologists (B.L.Y and J.G., respectively with 35 and 7 years of related experience), blinded to the neonatal/perinatal history, evaluated all MR images independently. They reviewed the published written and visual descriptions of the T2h appearances and agreed on its appearances prior to image analysis. The MR images were anonymized and reviewed on the same workstation, with the same window width and level. One month after the initial review, the MR images were reviewed for a second time by one of the observers (J.G.). Disagreements regarding image findings were resolved by discussion and mutual agreement. In the end, the intraand inter-observer agreements were evaluated. MR data processing. Rigid registration and distortion correction were performed after brain extraction 37,38 .
Artifact-corrupted DWIs were excluded by using an automated method 39 . Diffusion tensor and kurtosis tensor were estimated by using constrained weighted linear least squares (CWLLS) 40,41 , according to the following equation 42 : where S(b) was the diffusion weighted signal at a particular b value, and S(b) the signal without applying any diffusion gradient. D, K were the apparent diffusion coefficient and diffusional kurtosis. Parametric maps of FA, MD, AD, and RD were derived from the diffusion tensor 27,43 . The kurtosis tensor (KT) was transformed by the three eigenvectors of the diffusion tensor (DT) 43 : The kurtosis along an individual DT eigenvector can be computed from the transformed KT 43 : where λ i was the eigenvalues of the DT. RK can be derived by using the eigenvectors on the radial directions 43 . In the WM model for DKI, let D a and D e represent diffusion tensors in the intra-axonal and extra-axonal spaces. The axonal water fraction (AWF) was denoted by the symbol f. The DW signal was described as a function of the b value by the equation 27  The AWF was estimated by the equation 27 : The diffusion coefficients in the intra-axonal and extra-axonal spaces were calculated by the equations 27 : e,i i i The intra-axonal diffusivity 27 : The axial diffusivity in the extra-axonal space 27 : e,// e,1 where λ e,1 was the primary eigenvalue of D e .
The radial diffusivity in the extra-axonal space 27 : e, e,2 e ,3 where λ e,2 , λ e,3 were the 2nd, 3rd eigenvalues of D e . Artifacts rejection and tensor estimation were performed by using the in-house software implemented in MATLAB version 7.11.0 (Math Works, Natick, MA, USA). 44 was performed by using the optimized pipeline for neonates 45 . All the FA images were normalized to the neonatal Johns Hopkins template 46 . The aligned FA image of each subject was projected onto the mean FA skeleton (threshold = 0.15). Inter-group comparisons of the above metrics were tested with adding covariates (including GA, PMA and BW) in the general linear model. The number of permutations was set at 5000. All tests were taken to be significant at P < 0.05 after family-wise error rate (FWE) correction with threshold-free cluster enhancement (TFCE). ROI analysis was also performed based on the Johns Hopkins University WM label atlas 46 . The PLIC, SCR, CC and EC, which respectively represented projection fibers, commissural fibers and association fibers respectively, were selected as target regions.

TBSS and Region of interest (ROI) analysis. TBSS
Statistical analysis. Statistical analysis was conducted using SPSS for Windows version 17.0 (SPSS, Chicago,