Feasibility of oscillating and pulsed gradient diffusion MRI to assess neonatal hypoxia-ischemia on clinical systems

Abstract

Diffusion-time- (t_d) dependent diffusion MRI (dMRI) extends our ability to characterize brain microstructure by measuring dMRI signals at varying t_d. The use of oscillating gradient (OG) is essential for accessing short t_d but is technically challenging on clinical MRI systems. This study aims to investigate the clinical feasibility and value of t_d-dependent dMRI in neonatal hypoxic-ischemic encephalopathy (HIE). Eighteen HIE neonates and six normal term-born neonates were scanned on a 3 T scanner, with OG-dMRI at an oscillating frequency of 33 Hz (equivalent t_d ≈ 7.5 ms) and pulsed gradient (PG)-dMRI at a t_d of 82.8 ms and b-value of 700 s/mm². The t_d-dependence, as quantified by the difference in apparent diffusivity coefficients between OG- and PG-dMRI (ΔADC), was observed in the normal neonatal brains, and the ΔADC was higher in the subcortical white matter than the deep grey matter. In HIE neonates with severe and moderate injury, ΔADC significantly increased in the basal ganglia (BG) compared to the controls (23.7% and 10.6%, respectively). In contrast, the conventional PG-ADC showed a 12.6% reduction only in the severe HIE group. White matter edema regions also demonstrated increased ΔADC, where PG-ADC did not show apparent changes. Our result demonstrated that t_d-dependent dMRI provided high sensitivity in detecting moderate-to-severe HIE.

Introduction

Neonatal hypoxic-ischemic encephalopathy (HIE) is a major cause of neurodevelopmental disorders and death in neonates. About 1.5% of newborns worldwide suffer from hypoxic-ischemic brain damage to varying degrees.¹ More than 40% of the newborns with moderate-to-severe HIE suffer from cerebral palsy, epilepsy, and other neurological complications.² The time window for HIE treatment is extremely short, e.g. hypothermia treatment is known to have brain protection effect for moderate and severe HIE babies within 6 h after onset.³ The short time window emphasizes the importance of early diagnosis. Magnetic resonance imaging (MRI), as a non-invasive medical imaging technology, plays an important role in the diagnosis of neonatal HIE. At present, routine MRI examination for HIE babies includes T1-weighted, T2-weighted, and diffusion-weighted imaging, and spectroscopy.^4,5 These images provide rich information regarding the injury site, degree, and timing; however, their sensitivity and specificity do not fully meet the clinical needs.^6,7 It is necessary to develop early markers that provide specific information regarding the pathological changes of neuronal microstructure in addition to the global injury, in order to improve the diagnostic power and support clinical decisions.

Diffusion MRI (dMRI) has shown high sensitivity in early detection of ischemic injury, including HIE.^8,9 For instance, the apparent diffusion coefficient (ADC), which measures the extent of water molecule diffusion, drops within minutes after ischemic stroke.¹⁰ ADC reduction in the basal ganglia (BG) and ventrolateral thalamus (vThal) is a hallmark of neonatal HIE during acute injury.⁴ While it is assumed that the ADC reduction after ischemic insults is primarily associated with cell swelling,¹¹ a range of pathological events could contribute to the ADC reduction, including the change of ion concentration in the cytoplasm, astrocyte activation, neurite beading, etc.^12–14 An untapped potential of dMRI is to provide detailed microstructural information to identify the HI pathology. Recent progress on diffusion-time (t_d) dependent dMRI¹⁵ opened a new avenue to probe brain microstructure by measuring water diffusion at varying t_d, and thereby sensitizing the dMRI signals to microstructures at different spatial scales. The technique has shown to be useful in characterizing cellular microstructural features, such as cell size, intra/extra-cellular space, surface-to-volume ratio, and cellularity.^16–18 Several preclinical studies have used t_d-dependent dMRI to examine tumor and ischemic injury.^17,19–24 Wu et al.^19,20 demonstrated the advantages of this technique in a neonatal mouse model of HIE that the ADC measured at short t_d’s (1.25–2.5 ms) were more sensitive than conventional long t_d measurements to detect mild HI injuries that mostly consisted of subcellular microstructural changes. Aggarwal et al.²⁴ showed that t_d-dependent ADC measurements provided a contrast to detect HIE-induced neurodegeneration in the neonatal mouse hippocampus in an ex vivo study.

However, clinical translation of t_d-dependent dMRI remains challenging due to the difficulty of achieving short t_d with the limited gradient strength on clinical scanners. Often, oscillating-gradient (OG)²² was used to achieve shorter t_d (e.g. ≤10 ms) than previously attainable with conventional pulsed-gradient (PG). For OG-dMRI, the effective t_d was approximated as $\frac{1}{4f}$,²² and the b-value decreases rapidly with increasing oscillating frequency (b $\propto$ 1/f ³). Therefore, the performance of OG-dMRI is limited by the relatively low gradient strength on clinical systems, which leads to the limited oscillating frequency range, low b-value, and low signal-to-noise ratio (SNR). Nevertheless, OG-dMRI has been attempted in tumor,^25–28 stroke patients,^29,30 and healthy adult brain^31,32 given these technical limitations. Particularly, Baron et al.²⁹ performed OG and PG-dMRI scans in 11 stroke patients and the results suggested that the ADC differences between OG and PG scans, which represented the t_d-dependence, increased in the lesion areas. Another case study by Boonrod et al.³⁰ showed similar findings in patients with acute-to-subacute infarctions.

In this study, we tested the technical feasibility and clinical utility of t_d-dependent MRI with pulsed and oscillating gradients in neonatal HIE on a clinical 3 T system. Due to the higher ADC (1–1.5 µm²/ms)³³ and longer T₂ relaxation time (150–200 ms at 3 T)³⁴ in the neonatal brain than the adult brain, the technical limitations of OG-dMRI, e.g. low b-values and long TE, were partially mitigated in neonatal MRI. These advantages enabled us to investigate the value of t_d-dependent dMRI in the diagnosis of neonatal HIE. Here, we examined PG- and OG-ADC and the t_d-dependence measurements in mild, moderate, and severe HIE neonates in comparison with the normal controls.

Material and methods

Participants

The study was conducted in accordance with the principles of the Declaration of Helsinki and all research protocols were approved by the Institutional Review Board at the Children’s Hospital, Zhejiang University School of Medicine. Written and verbal informed consents were provided by the parents or legal guardians. Eighteen term-born HIE neonates were recruited from February 2019 to October 2019, including seven severe HIE, six moderate HIE, and five mild HIE neonates. The HIE neonates were enrolled due to an acute peripartum or intrapartum event of (1) Apgar score <5 at 5 min and 10 min after birth, and/or (2) umbilical artery blood pH < 7.0, or base deficit ≥12 mmol/L, or both, and/or (3) symptoms of the nervous system shortly after birth and lasted for more than 24 h, including changes in consciousness (excessive excitement, drowsiness, coma), changes in muscle tone (hypermyotonia or hypomyotonia), abnormal primordial reflexes (weakened or disappeared sucking and hug reflexes), seizure, central respiration failure, and pupil changes (dilated, diminished, or unequal). Acute encephalopathy due to other causes and non-acute brain injury occurred in utero were excluded. The severity of HIE was graded as severe (n = 7), moderate (n = 6), or mild (n = 5) according to the modified Sarnat scoring system.³⁵ Hypothermia was administrated to the majority of the severe (5/7) and moderate (5/6) HIE babies and selected mild HIE babies (2/5) according to the guideline.³⁶ The hypothermia was initiated as soon as the patients were administrated, and lasted for 72 h for all patients. The basic information of the HIE babies is listed in Table 1, including the severity, gender, gestational age at birth (GA), post-menstrual age at scan (PMA), hypothermia status, ventilation support, routine radiological report about the affected tissues, and clinical background of these patients.

Table 1.

Basic demographic and clinical characteristics of the neonatal HIE patients, including the severity, gender, gestational age at birth (GA), post-menstrual age at scan (PMA), hypothermia (HT) status, ventilation support, routine radiological report about the affected tissues, and clinical background of these patients.

	Severity	Sex	GA (weeks)	PMA (days)	HT	Venti-lator	Tissue affected (routine MRI)	Clinical background
1a	Severe	M	39.4	6	Yes	Yes	Basal ganglia (focal), thalamus, diffusive WM injury	Absent primitive reflexes, hypertonia, seizures, poor light reflex
2	Severe	M	37.6	6	Yes	Yes	Bilateral periventricular WM	Absent primitive reflexes, hypotonia, seizures
3b	Severe	M	40.8	8	No	Yes	Basal ganglia, thalamus (focal), frontoparietal lobe	Absent primitive reflexes, hypertonia, seizures
4	Severe	F	40.2	6	Yes	Yes	Periventricular WM, posterior limb of the internal capsule	Absent primitive reflexes, hypotonia, seizures
5	Severe	F	40.0	4	No	Yes	Basal ganglia (focal)	Absent primitive reflexes, hypotonia, poor light reflex
6b	Severe	M	38.7	10	Yes	Yes	Basal ganglia, thalamus, diffusive WM injury	Absent primitive reflexes, hypotonia, seizures
7b	Severe	M	40.8	22	Yes	Yes	Entire cerebrum	Absent primitive reflexes, hypotonia, seizures
8	Moderate	M	40.0	4	No	Yes	Subdural hemorrhage	Weak primitive reflex, hypertonia
9	Moderate	F	40.7	8	Yes	Yes	Not noticeable	Weak primitive reflexes, hypertonia, decreased spontaneous activity
10b	Moderate	M	38.4	7	Yes	No	Periventricular white matter	Poorly responsive, hypotonia, decreased spontaneous activities
11a	Moderate	F	39.2	17	Yes	No	Parieto-occipital subaponeurotic hematoma	Weak primitive reflexes, hypertonia, decreased spontaneous activities
12	Moderate	F	37.0	5	Yes	Yes	Not noticeable	Weak primitive reflexes, hypertonia, decreased spontaneous activities
13	Moderate	F	38.4	4	Yes	No	Parietal cortex	Weak primitive reflexes, hypotonia
14	Mild	F	40.2	6	No	No	Not noticeable	Slightly decreased spontaneous activities, hypertonia
15	Milda	F	40.0	6	Yes	No	Occipital subdural hemorrhage and left intraventricular hemorrhage	Slightly decreased spontaneous activities, hypertonia, weak primitive reflexes
16	Mild	M	40.7	5	No	No	Not noticeable	Lightly decreased spontaneous activities, low threshold to elicit motor reflexes
17	Mild	M	40.4	5	No	No	Globus pallidus, right frontal WM, and splenium of the corpus callosum	Slightly decreased spontaneous activities, hypertonia
18	Mild	F	39.0	15	Yes	Yesc	Enlargement of subarachnoid space	Mild distal flexion, hypertonic

^aPatient images that are presented in Figure 2.

^bPatient images that are presented in Figure 4.

^cThe baby was intubated because of lung disease (not for central respiratory failure).

In addition, six normal term-born neonates (post-menstrual age <28 days) were enrolled as controls. The control subjects were referred for MRI scanning due to febrile seizures, without known conditions that could alter the dMRI measurements. Radiological reports confirmed no evidence of neurologic abnormalities.

Data acquisition

All scans were performed on a 3 T Philips Achieva MRI scanner (Philips Healthcare, Best, The Netherlands) with a maximum gradient of 80 mT/m and a slew rate of 100 mT/m/ms, using an 8-channel head coil. The neonates were sedated using 10% chloral hydrate (0.5 ml/kg) via enema administration. A vacuum immobilization mat was used to minimize motion, and earmuffs were used to attenuate the scanner noise. MRI scans of the HIE neonates were performed at the earliest possible time, after hypothermia or other necessary interventions.

Routine T1-weighted images were acquired with a fast field-echo sequence with echo time (TE)/repetition time (TR) = 2/167 ms, flip angle = 80°, field of view (FOV) = 180 × 180 mm², in-place resolution = 0.42 × 0.42 mm, 15 slices with a slice thickness of 6.5 mm. T2-weighted images were acquired with a turbo spin-echo sequence with TE/TR = 80/3000 ms, FOV = 180 × 180 mm², in-place resolution = 0.34 × 0.34 mm, 15 slices with a slice thickness of 6.5 mm. A house-made OG-dMRI sequence with trapezoid-cosine gradients was implemented, and images were acquired with the following parameters: an oscillating frequency of 33 Hz (effective t_d = 7.5 ms) with gradient amplitude of 70 mT/m and slew rate of 46.7 mT/m/ms, two oscillating cycles, b-value of 700 s/mm², four tetrahedral diffusion directions of [1,1,1], [−1,1,1], [1,−1,1], and [1,1,−1], one non-diffusion-weighted image (b₀), TE/TR = 168/10,000 ms, FOV = 180 × 180 mm², two averages, in-plane resolution = 1.4 × 1.4 mm, and 10 slices with slice thickness of 4 mm. The slices were centered approximately at the mid-to-posterior BG section. PG-dMRI was performed with diffusion duration (δ)/diffusion separation (Δ) = 60/82.8 ms, and the other imaging parameters matched to those of the OG-dMRI scan. The entire OG- and PG-dMRI protocol took 3 min.

Data analysis

The diffusion-weighted images (DWIs) were first registered to the b₀ image within the same scan and then coregistered between PG and OG scans, by affine transformation to correct for subject motion and eddy current,³⁷ using Automatic Image Registration (AIR) version 5.1 (http://bishopw.loni.ucla.edu/AIR5/). ADC was calculated as –log (S/S₀)/b between the diffusion-weighted signal (S) and b₀ signal (S₀), and averaged over the four diffusion directions. The t_d-dependence was quantified as the difference between OG-ADC and PG-ADC, namely, ΔADC = OG-ADC–PG-ADC.

Regions of interest (ROIs) were manually delineated on the b0 images by an experienced pediatric radiologist (F.G.), which was checked by a MR physicist (D.W.). The putamen and globus pallidus of the BG (or lentiform nucleus), vThal, and anterior and posterior subcortical white matter (sWM) ROIs were defined on the center slice at the mid-to-posterior BG section (Figure 1(b)), and the procedure was blinded to the control and HIE groups. WM edema ROIs were defined on b0 images of four severe/moderate (n = 3/1) patients who had visible hypointensity on the T2-weighted images and b0 images. ROI-averaged OG-ADC, PG-ADC, and ΔADC were used for statistical analysis.

Figure 1.

PG- and OG-dMRI of the normal neonatal brain. (a) b₀ images, diffusion-weighted images (DWI), and mean apparent diffusivity (ADC) maps from the PG and OG measurements and their difference map (ΔADC) of a normal neonatal brain in three transverse views. (b) ROIs used in the analysis, including the anterior and posterior subcortical white matter (sWM), basal ganglia (BG), and thalamus. (c) ADC measured in the four ROIs using the PG- and OG-dMRI sequences. (d) ΔADC measured in the four ROIs. The bar plots in (c) and (d) represent mean ± standard deviation over six healthy neonates. *p ≤ 0.01 and **p < 0.001 by one-way ANOVA followed by post hoc t-tests.

Statistical analysis

In the control group, PG- and OG-ADC measurements in the different ROIs were compared by two-way analysis of variance (ANOVA) followed by post hoc t-test with Turkey’s correction. ΔADC was compared among the different ROIs with one-way ANOVA followed by post hoc t-test with Tukey’s correction. To compare the OG-ADC, PG-ADC, and ΔADC among severe, moderate, and mild HIE, and control groups while taking into account the hypothermia status (0 or 1) and PMA at scan, a three-way ANOVA was performed with the group, hypothermia, and PMA at scan as factors, followed by post hoc pairwise t-tests with Turey’s correction. The dMRI measurements were also compared between WM edema in four HIE patients and the controls using the Mann-Whitney rank–sum test. All statistical analysis was performed in Graphpad Prism (https://www.graphpad.com/scientific-software/prism/).

Results

The dMRI sequences were calibrated using a Philips water phantom. At the room temperature, ADC measured in the phantom was 2.24 ± 0.02 µm²/ms for the PG sequence and 2.25 ± 0.05 µm²/ms for the OG sequence, for 10 repeated measurements on different days. No statistical difference was found between PG and OG sequences in the phantom (p = 0.65).

We first examined the feasibility and reliability of OG- and PG-dMRI in the normal neonatal brains. Figure 1(a) showed representative b₀ images, DWIs with gradient direction of [1 1 1], and mean ADC maps measured using the OG and PG sequences. At a b-value of 700 s/mm², the DWI signal was reduced to approximately 30% of the b₀ signal, and tissue anisotropy can be well-appreciated in the DWI images. The OG-ADC was higher than PG-ADC in all ROIs (p < 0.001) (Figure 1(c)), and the percentages of ADC increase were 13 ± 2%, 11 ± 5%, 16 ± 1%, 15 ± 2% for the BG, vThal, anterior and posterior sWM, respectively. Consequently, the ΔADC was found to be significantly higher in the anterior and posterior sWM than that in the BG and vThal (p ≤ 0.01) (Figure 1(d)). Cortical ΔADC was not examined due to the lack of resolution to accurately delineate the thin cortex in the neonates.

Given the global insult, varying radiological manifestations were found across subjects, from no obvious injury in some of the mild-to-moderate HIE neonates to whole-brain injury in some of severe HIE neonates (Table 1). Figure 2 demonstrated OG- and PG-ADC maps of representative HIE neonates with severe, moderate, and mild injuries. It was found that PG-ADC was reduced in BG of the severe HIE patients compared with the control neonates, whereas OG-ADC did not show any apparent change. ΔADC increased markedly in the BG of the severe HIE patients, and also slightly increased in the BG of the moderate HIE patients, compared with the control subjects (red contours).

Figure 2.

T1-weighted, T2-weighted, PG-ADC, OG-ADC, and ΔADC maps of a normal neonate and neonates with severe, moderate, and mild HIE. The red dashed contours indicate the basal ganglia (BG) for the following analysis. The images were selected from representative patients, as denoted in Table1.

Focal or diffusive ADC reductions in the BG were visible in all severe HIE babies. Thereby, we focused on the analysis of BG (red contours in Figure 2), which is known to be selectively injured in moderate-severe HIE babies who underwent an acute hypoxic-ischemic event.³⁸ Three-way ANOVA indicated that only the group effect (control, mild, moderate, and severe) was significant for PG-ADC (p < 0.01), while all three factors, including group, hypothermia, and PMA at scan, had significant effects on ΔADC. Post hoc t-test showed that PG-ADC of the BG in the severe HIE group was significantly lower than that in controls (Figure 3(a)), but no differences were found in the moderate and mild HIE groups. OG-ADC did not show any significant differences among groups. A considerable elevation in ΔADC was observed in the BG of the severe and moderate HIE neonates, compared to the controls, and a gradual increase of ΔADC was observed from the mild to moderate and severe HIE groups (Figure 3(a)). Moreover, the extent of ΔADC difference between severe HIE patients and controls (23.7% increase) was larger than that of the PG-ADC (12.6% reduction). In other regions that we investigated, including the vThal and anterior/posterior subcortical WM, no significant changes were observed (Figure 4(b) to (c)), possibly due to the heterogeneity of injury across individuals.

Figure 3.

PG-ADC, OG-ADC, and ΔADC measurements in severe HIE, moderate HIE, mild HIE, and control neonates. (a) Measurements in the BG. (b) Measurements in the ventrolateral thalamus (vThal). (c) Measurements averaged over the anterior and posterior subcortical white matter (sWM). *p < 0.05 and **p < 0.01 by one-way ANOVA flowed by post-hoc t-tests.

Figure 4.

PG- and OG-dMRI of HIE neonates with white matter (WM) edema. (a) T1-weighted images, T2-weighted images, PG-ADC, OG-ADC, and ΔADC maps of four HIE neonates with WM edema. Arrows point to the anterior subcortical WM that showed edema based on the T2w-weighted images. (b) PG-ADC, OG-ADC, and ΔADC values in the WM edema regions with respect to the controls. *p < 0.01 by Mann–Whitney rank-sum test. The four patients are denoted in Table 1.

Four HIE neonates, including three severe HIE and one moderate HIE babies, exhibited WM edema in the anterior subcortical WM regions based on the T2-weighted images (arrows in Figure 4). PG- and OG-dMRI did not show ADC differences in the WM edema regions compared with the controls, whereas ΔADC was significantly increased in these regions (p = 0.009 by Mann–Whitney rank-sum test).

Discussion

In dMRI, diffusivity reflects the restrictive effects of multiple structural barriers encountered by water molecules within certain t_d. t_d-dependent dMRI with oscillating and pulsed gradients allows us to probe water diffusion at different length scales. The measured diffusivities reflect pathological changes at the corresponding scales according to $x=\sqrt{2Dt}$, e.g. ∼5.5 µm for OG-dMRI at 33 Hz and ∼18.2 µm for PG-dMRI at t_d of 82.8 ms in our study, assuming a diffusivity of 2 µm²/ms. The pattern of the t_d-dependent ADC change has been linked to several microstructural properties, as shown in a number of simulation and animal studies under normal and pathological conditions.^{17,19–24,39,40} Despite the technical challenges of OG-dMRI on clinical systems, a few human studies have demonstrated the potential of t_d-dependent dMRI, e.g. in tumor applications.^25,26,28 Here we demonstrated the technical feasibility and clinical utility of t_d-dependent dMRI in neonatal HIE, which is a potentially favorable application as the high ADC and T2 relaxation time in the neonatal brain mitigated the technical issues of OG-dMRI. We were able to reliably detect t_d-dependence from PG- and OG-dMRI measurements in the neonatal brain tissues, and we found higher ΔADC in the subcortical WM compared to deep GM. t_d-dependent dMRI has been used to examine ischemic injury,^{19,20,22,24,29,30} where drastic microstructural changes take place. In a mouse model of neonatal HIE, we previously demonstrated increased t_d-dependence in edema where neuronal swelling and astrocytic activation took place¹⁹; furthermore, the sensitivity to early HIE injury enhanced at ultra-short t_d, possibly owing to the unique sensitivity of short-t_d diffusion to changes of the dendritic and subcellular processes.²⁰ Two recent human studies of cerebral ischemia reported that the lesion contrasts reduced at short t_d using OG-dMRI^29,30; in other words, the ΔADC increased in the lesion areas, similar to our findings in neonatal HIE. According to previous simulation studies, the increase of ΔADC may be associated with an increase in cell/axonal size, cell density, or permeability,¹⁸ and the amount of ΔADC change depends on the t_d range under investigation. In the WM, the increased ADC could be mixed effects of axonal inflammation/beading.¹³ In our study, the observed increase of ΔADC in BG is likely associated with cell swelling and glial activation in the BG that eventually led to necrotic cell death²⁰ and neurological deficits.⁴¹ Moreover, the increase of ΔADC was not only found in severe HIE but also in moderate HIE and WM edema regions, whereas PG-ADC did not show changes in these cases. In addition, the effect size of ΔADC change in severe HIE was larger than that with PG-ADC. These evidences suggested ΔADC may provide higher sensitivity than the conventional PG-dMRI in detecting neonatal HIE.

It is known that diffusion MRI measurements in HIE neonates highly depend on the timing of examination. Barkovich et al.⁴ previously showed that ADC reduction ranged from 5 to 40% within the first three days after birth and then the ADC gradually increased in the edema regions, and pseudo-normalization may develop in some of the neonates after day 7. In our study, the patients could not be scanned during the acute stage when they were under hypothermia, and the PMA at scan varied from 4 to 22 days after birth. Although we have included the PMA at scan as a covariate in our analysis, it cannot replace the acute measurements. On the other hand, our results support that t_d-dMRI is less susceptible to pseudo-normalization, first reported in our previous study in the mouse model of HIE.²⁰

On a typical clinical MR scanner with gradient ≤80 mT/m, the oscillating frequency is limited to 30–60 Hz even with a relatively low b-value (<500 s/mm²),^42,43 and OG-ADC at the relatively low frequency was shown to be less sensitive to lesions in itself. In contrast, on preclinical scanners (gradient > 700 mT/m), higher oscillating frequencies (100–1000 Hz) could be reached, and the ADC at high oscillating frequency demonstrated high sensitivity to small microstructural changes in animal studies.^20,21 At an oscillating frequency of 200 Hz, the diffusion distance of water molecules is approximately 2 µm, which is likely to be selectively sensitive to subcellular microstructures (e.g., axons and dendrites). The diffusion distance is about 5 µm at an oscillating frequency of 33 Hz, and the measured signals may reflect the restrictive effects of extracellular, intracellular, and subcellular compartments, which may counteract each other, leading to the diminished lesion contrast. Advances in hardware^44,45 and imaging sequences^46,47 would be essential to push the limit of t_d-dependent dMRI in human studies.

Several other technical considerations are also worth noting for applications of OG-dMRI on clinical systems, including low SNR, high duty cycle, eddy current, and peripheral nerve stimulation. Due to the gradient constraint, multiple oscillating gradient cycles are needed to reach a reasonable b-value, and the echo time has to be prolonged in order to accommodate the long diffusion gradients, resulting in a relatively low SNR. Therefore, we used relatively thick slices (4 mm) to compensate for the SNR loss. Moreover, the application of strong oscillating gradients leads to high duty cycle and eddy current. A long TR is needed to reduce the duty cycle. Since the diffusion gradients are applied for every slice in a 2D multi-slice sequence, TR increases linearly with the number of slices. We used a relatively small number of slices (10 slices) with a TR of 10 s, as a whole-brain acquisition would require longer scan time. Also, eddy current induced image translation along the phase-encoding direction and distortion in the DWIs needs to be corrected by affine registration of the DWIs to the b₀ image.⁴⁸ Advanced eddy correction methods may also be used given certain additional information.^49,50 In addition, we used a low slew rate of 46.7 mT/m/ms to avoid peripheral nerve stimulation at the expense of a slightly longer gradient duration. Gradient optimization techniques⁵¹ could also help to reduce the stimulation. The design of oscillating gradient is also important for reliable measurement of the t_d-dependence, especially for the limited range of oscillating frequency on clinical scanners. It was shown that an oscillating frequency around 40 Hz with two cycles or more gave the best signal-to-noise ratio in the ΔADC maps,³¹ and our gradient waveform was among the optimal.

There are several limitations in the current study. First, our scan protocol only included one PG-dMRI and one OG-dMRI scan that took 3 min, and we were not able to examine the whole spectrum of t_d-dMRI at multiple Δ from PG sequence or multiple oscillating frequencies from OG sequence. Due to this reason, we could not employ biophysical models^17,23 to quantify the microstructural changes in HIE. Secondly, the study population was relatively small and the PMA at scan varied between individuals. Although we have included several clinical variables as covariates in the analysis, it may not entirely remove the influences. A larger cohort with more consistent PMA at scan and longitudinal follow-ups in the future would be important to validate the clinical significance of the current findings. In addition, we only examined a few selected GM and WM regions with manually delineated ROIs. A whole-brain analysis may provide more comprehensive information regarding the injury patterns. Nevertheless, as the first attempt, our study demonstrated the clinical feasibility and value of t_d-dMRI in neonatal HIE.

Conclusion

Here we demonstrated the feasibility of t_d-dependent dMRI on a clinical 3 T system for neonatal HIE, using oscillating and pulsed gradient sequences. The t_d-dependence, as characterized by ΔADC, was sensitive to the ischemic injury in the BG of moderate-to-severe HIE neonates, as well as the WM edema related pathology.