MRI predictors of long-term outcomes of neonatal hypoxic ischaemic encephalopathy: a primer for radiologists

Abstract

This review aims to serve as a foundational resource for general radiologists, enhancing their understanding of the role of Magnetic Resonance Imaging (MRI) in early prognostication for newborns diagnosed with hypoxic ischaemic encephalopathy (HIE). The article explores the application of MRI as a predictive instrument for determining long-term outcomes in newborns affected by HIE. With HIE constituting a leading cause of neonatal mortality and severe long-term neurodevelopmental impairments, early identification of prognostic indicators is crucial for timely intervention and optimal clinical management. We examine current literature and recent advancements to provide an in-depth overview of MRI predictors, encompassing brain injury patterns, injury scoring systems, spectroscopy, and diffusion imaging. The potential of these MRI biomarkers in predicting long-term neurodevelopmental outcomes and the probability of epilepsy is also discussed.

Introduction

Hypoxic-ischaemic encephalopathy (HIE) is a significant source of mortality and disability in neonates caused by hypoxia and ischaemia to the neonate’s brain.¹ Neonatal brain hypoxia and ischaemia can occur during the antenatal, intrapartum, or postnatal period, and may result in severe neurological complications, including cerebral palsy, epilepsy, developmental delays, and cognitive impairment.¹ Therapeutic hypothermia (TH) is an effective and safe intervention, significantly mitigating mortality and disability in moderate to severe HIE cases. Nonetheless, 30%-70% of affected neonates have an early death or survive with neurological impairments.^2–5 Therefore, it is imperative to establish reliable parameters to assess the extent and severity of brain injury during the subacute period following hypothermia. These parameters could guide management of required adjuvant therapies, like duration of seizure management in patients with extensive injury on MRI or the choice of appropriate neurorehabilitation for specific brain injuries. Additionally, they could serve as early indicators of the effectiveness of TH and aid in the assessment of preclinical and clinical trials for emerging neuroprotective treatments, such as erythropoietin and allopurinol.⁶ Early and accurate predictors of long-term neurodevelopmental outcomes (LTNO) are essential for prognostication and comprehensive family counseling post-HIE.

MRI, by virtue of its non-invasive, high-resolution imaging capabilities, can elucidate pathological changes otherwise undetectable via clinical assessment or alternative imaging techniques. A plethora of studies have underscored the predictive merit of MRI in determining neonatal HIE outcomes.^7–9 Nonetheless, the widespread clinical applicability of MR’s prognostic value is curtailed by heterogeneity in imaging techniques, timing of assessments, follow-up durations, and thresholds for unfavourable outcomes of the index test.⁷

The objective of this review is to offer a foundational resource for radiologists, detailing the most contemporary and frequently utilized semi-quantitative and quantitative methodologies for evaluating HIE injuries in term and near-term infants (gestational age ≥35 weeks) after TH and their prognostic implications. This review emphasizes the associations between MRI and long-term follow-ups, wherein neonates are followed until they reach or exceed 18 months of age, given the challenges in accurately assessing neurological sequela in neonates younger than 12 months and the observations that mental and behavioural disabilities may present even later.¹⁰

Long-term neurodevelopmental outcomes

Research on the long-term neurodevelopmental impacts in infants suffering from HIE frequently utilizes composite outcomes as the outcome measures. These composite outcomes are categorized into favourable or unfavourable/adverse groups, but definitions can vary markedly across studies, primarily depending on the neuropsychiatric assessment instruments used. Adverse or unfavourable outcomes are generally identified by one or more of the following: mortality, cerebral palsy, detectable developmental delay using validated tools such as IQ tests or the Bayley Scales of Infant and Toddler Development, intractable epilepsy, or vision and hearing impairments. Notably, the exact criteria for these outcomes can fluctuate among different studies (Table 1). Moreover, several meta-analyses have consistently affirmed the reliability of MRI in predicting these adverse outcomes.^7,9

Table 1.

Characteristics of studies using MRI predicting long-term neurodevelopmental outcomes (modified from Ouwehand et al⁹).

Study	Timing of MRI, days†	n/N*	HIE grade**	Follow-up, months	Adverse outcome, %	Definitions of adverse outcome	Predictors
Shankaran et al17	14.7 ± 10.6	73/73	0/54/18	18-22	30	Death, BSID-II < 85, GMFCS 2-5, HI, VI, epilepsy	NICHD NRN score
Shankaran et al54	15.3 ± 12.3	69/69	NR	72-84	32	Death, IQ < 70	NICHD NRN score
Alderliesten et al39	4.7 ± 1.3	65/65	6/47/12	18	28	Death, GMDS < 85, CP	ADC of posterior corpus callosum
Charon et al30	4 (3-6) (first MRI), 11 (7-21) (second MRI)††	38/43	0/17/14	24	21	Death, revised Brunet-Lezine score < 70, GMFCS 3-5	Signal abnormalities of PLIC on first or second MRI
Weeke et al55	6.4 ± 2.9	22/26	0/17/7	24	50	Death, BSID-III < 85, GMFCS 2-5, epilepsy, moderate to severe disability	Adapted Barkovich score
Alderliesten et al27	Within 7 days†††	88/88	10/62/16	24	33	Death, BSID-III < 85, CP, VI, HI	ADC of BGT, Lactate/NAA ratio of left BGT (> 0.4)
Heursen et al31	4-6†††	54/54	10/29/15	24	33	Death, BSID-III < 70, GMFCS 2-5	ADC of PLIC and thalami
Barta et al38	Within 4 days†††	51/68	0/5/46	18-26	31	Death, BSID-II < 70	MRS (left thalamus); mIns/NAA (≥ 0.6798) is the best predictor
Weeke et al51	Within 7 days†††	173/173	16/129/25	24, 66-96 (school age)	26, 21	(24 m) Death, BSID-III < 85, GMFCS 2-5 (school age) Death, GMFCS 2-5, IQ < 85	Weeke score, grey matter subscore
Chang et al32	NR	107/120	11/80/16	18-24	29	BSID-III < 85	DWI lesion sizes and number
Bach et al18	4 [3,5] ††††	155/204	NR	30	NR	BSID—II < 70 or BSID—III < 85	Barkovich score
Thoresen et al19	8 [6.5, 9] ††††	168/178	NR	18-24	21	Death, BSID-III < 85, GMFCS 3-5, SI, VI	Rutherford score, WMxBGT
Steiner et al80	8 [4,9] ††††	54/56	0/45/11	24	25	Death, BSID-III < 70	NICHD NRN score, Rutherford score
Troha Gergeli et al81	5.7 ± 2.6	50/50	0/21/29	18, 60-72	37.8, 34	(18 m) Death, BSID-III < 85 (60-72 m) Death, moderately to severely abnormal ATNA, intractable epilepsy, GMFCS 2-5	Rutherford score
Aker et al82	5 ± 1	25/25	0/24/1	18	21	Death, BSID-III < 85, GMFCS 3-5, VI, HI, seizures	Rutherford score

^†Mean ± SD if not specified.

^††Median (range).

^†††Range.

^††††Median [interquartile range].

^*Number of infants receiving hypothermia with neurodevelopmental follow-up and the total infants enrolled into this study.

^**Reported as mild, moderate, or severe.

Abbreviations: ADC = apparent diffusion coefficient, ATNA = Amiel-Tison Neurological Assessment, BSID-II = Bayley Scales of Infant Development II, BSID-III = Bayley Scales of Infant and Toddler Development third edition (Bayley-III), CP = cerebral palsy, GMDS = Griffiths mental developmental scales, GMFCS = Gross Motor Function Classification System, HI = hearing impairment, mIns = myo-inositol, NAA = N-acetylaspartate, NR = not reported, PLIC = posterior limb of internal capsule, VI = visual impairment.

Methods for grading and quantification of brain injury in HIE

Numerous qualitative and quantitative techniques are available for evaluating brain damage in MRI scans of neonates suffering from HIE. The primary imaging modality for discerning regions of brain injury in newborns with HIE combines conventional MRI methodologies, specifically T1-, T2-, and diffusion-weighted imaging (DWI).^11,12 These established MRI techniques yield qualitative details, such as the locations and extent of brain lesions.¹¹ Further, DWI facilitates early recognition of acute ischaemic injuries by measuring water molecule diffusion within brain tissue while providing quantifiable data through the apparent diffusion coefficient (ADC), which gives insights into the extracellular matrix.¹³ In addition, more sophisticated MRI techniques, including diffusion tensor imaging (DTI) and proton MR spectroscopy (¹H-MRS), provide quantitative data on the brain’s microstructure and function.¹⁴

Brain injury patterns

The degree and location of brain injury in infants suffering from HIE can be categorized into three primary patterns. These classifications rely on the severity and duration of the hypoxic-ischaemic event, as well as the brain's level of maturation at the time of the event¹⁵ (Figures 1-4). It is important to recognize that neonatal brain injuries can manifest as a combination of injury patterns, and that these patterns and the severity of injuries can occasionally shift over time.¹⁶

Figure 1.

Normal. MRI performed on day 4 in a newborn who received therapeutic hypothermia for HIE. The signal intensity of the posterior limb of the internal capsule appears normal on T1-weighted imaging (A), T2-weighted imaging (B) and diffusion-weighted imaging (C).

Figure 4.

Near-total injury Pattern. MRI performed within the first week after birth, showing a near-total injury pattern characterized by extensive damage involving the cortex, white matter, basal ganglia, and thalami. These injured areas appear slightly hyperintense on T1-weighted imaging (A, B) and T2-weighted imaging (C, D) and are more easily visualized using diffusion weighted imaging (E, F).

Figure 2.

BGT predominant Pattern. MRI performed within the first week after birth, showing injury involving bilateral posterior putamina and thalami (BGT) as well as bilateral optic radiations, which appear T1 hyperintense on T1-weighted (A) and T2-weighted imaging (B) and more evident on diffusion-weighted imaging (DWI) (C).

Figure 3.

White matter/watershed predominant pattern. White matter/watershed predominant pattern. MRI performed on day 4 demonstrates a watershed predominant pattern of injury. T1-weighted inversion recovery imaging (A) and T2-weighted imaging (B) appear near completely normal with diffusion-weighted imaging (C) showing restricted diffusion in the watershed areas of frontoparietal lobes. Right frontoparietal scalp haematoma is also noted.

Basal ganglia and thalamus predominant pattern

The BGT pattern is often observed in circumstances that present a sudden disruption of placental perfusion or oxygen supply in the umbilical cord, which can arise due to placental abruption, umbilical cord prolapse, shoulder dystocia, or uterine rupture. This scenario typically results in acute, profound asphyxia, primarily affecting the ventrolateral thalami, posterior putamina, posterior limb of internal capsule (PLIC), and perirolandic cortex.

White matter/watershed predominant pattern

The white matter/watershed (WM/WS) pattern typically arises from mild to moderate asphyxia, also known as partial prolonged asphyxia, leading to a primary manifestation in the watershed areas. The lesions follow a parasagittal distribution along the vascular border zones and are also known as the “parasagittal pattern of injury”.

Near-total injury pattern

The Near Total Injury pattern, also labelled as “global injury”, characterizes infants subjected to very severe and prolonged asphyxia. This pattern is distinguished by widespread injury involving the entire or near entire hemispheres.¹⁷

Predicting LTNO based on injury patterns

Research studies have demonstrated that specific brain injury patterns, as well as MRI scoring systems, can effectively predict unfavourable neurodevelopmental outcomes in infants with HIE. Notably, the BGT injury pattern and near total injury pattern have been linked to adverse outcomes, while infants with only punctate white matter lesions did not exhibit a significantly increased risk.^18–20 The BGT pattern has been associated with a higher risk of abnormal cognitive, motor, and composite neurocognitive outcomes than the WM/WS pattern at 2.5 years.^18,21 On the contrary, the WM/WS pattern of injury has been associated with milder and fewer impacts on neurodevelopmental outcomes.²² Instances of the Near-Total Injury pattern are reported less frequently, often because these infants may be too medically unstable for scans after birth, and death may occur before an MRI can be performed.²³

Quantitative diffusion imaging

Diffusion-weighted imaging and DTI are techniques for measuring diffusion-driven displacements of water molecules over time and probing the microstructural anomalies of the brain following an HIE event, affording quantitative detection of pathologies that are not discernible by conventional MRI. Both DWI and DTI can measure the mean diffusivity, while DTI provides a deeper characterization of water molecule movement, including fractional anisotropy (FA) as well as axial and radial diffusivity—longitudinal and transversal to the fibre direction, respectively.

Diffusion imaging is more sensitive, and the abnormalities can occur earlier—during the first hours to days—than conventional T1- and T2-weighted MRI after hypoxic ischaemic injury. The mean diffusivity or ADC is initially reduced, correlating with the severity of apoptotic injury.²⁴ The mean coefficient reduction gradually recovers and reaches normal values (“pseudonormalization”) by the end of the first week following injury in normothermic patients, which can lead to underestimation of brain injury.²⁵

One common analytic approach of diffusion imaging is the region of interest approach, wherein diffusion measures are collected from specified brain regions, determined either by manual delineation or automatic parcellation or segmentation. Multiple studies have noted that a lower ADC across various brain regions within the first week to 10 days is associated with adverse outcomes by using the region of interest approach.⁹ Beyond these, there are a few studies using more computationally demanding quantitative methods, including tract-based spatial statistics (TBSS) and a structural connectivity network approach. These methods are automated, observer-independent approaches that allow for robust comparisons of DTI without the need for subjective pre-specification of regions of interest.

Tract-based spatial statistics enables a comprehensive whole-brain analysis of aligned FA images derived from multi-subject diffusion data.²⁶ Structural connectivity refers to the white matter tracts, or anatomical connections, between various brain regions. After these connections have been reconstructed, a connectivity matrix is established and examined using network analysis parameters. These include the node degree, clustering coefficient, and shortest path length, among other properties. However, while these analytic methods show promise, it is important to note that they are currently not standardized and often require advanced computation skills. As a result, their direct applicability in clinical practice remains limited currently.

Predicting LTNO based on diffusion imaging

Diffusion imaging, specifically ADC values in different brain structures, have been shown to predict adverse outcomes well. Multiple studies have noted that a lower ADC across various brain regions within the first week to 10 days is associated with adverse outcomes by using the region of interest approach.⁹ Key areas studied include the centrum semiovale, basal ganglia,^27–30 thalami,^28–31 and the PLIC.^28–31 However, ADC values in the cerebellum and brainstem have not been associated with outcomes.³¹ Among the anatomic regions, the ADC values of PLIC (0.82 × 10⁻⁹/mm²/s³⁰; 0.94 × 10⁻³/mm²/s³¹) was reported with the highest predictive value.^30,31 The higher size and number of lesions on DWI are also associated with worse outcomes at 1.5-2 years, although the area under curve (AUC) did not surpass that of the NICHD score.³² The FA values of basal ganglia and PLIC are more specific but less sensitive than ADC values in predicting LTNO.^9,33,34 Tusor et al utilized TBSS to illustrate significant associations between lower FA values in diverse brain regions and outcomes at the age of 2 years.³⁵ Spencer et al used structural connectivity network approach to predict motor function in children aged 6-8 years, who were treated with TH and did not develop cerebral palsy.³⁶

Proton MR spectroscopy

Proton nuclear magnetic resonance spectroscopy has the unique capability to discern metabolic changes and detect perturbations in neurochemistry instigated by perinatal asphyxia.³⁷ Multiple studies have deployed ¹H-MRS to forecast neurodevelopmental outcomes, utilizing a variety of metabolites and defining regions of interest.^{9,20,29,33,34,38–40} N-acetylaspartate (NAA) is the second most prevalent amino acid in the brain, found predominantly in neurons. It is widely recognized as a surrogate biomarker for neuronal integrity and function. Studies indicate that NAA levels tend to decrease following neuronal injury or dysfunction. Lactate, on the other hand, is a byproduct of anaerobic metabolism, so its levels increase in areas affected by hypoxic-ischaemic injuries. The lactate concentration associated with HIE fluctuates throughout the disease progression. In infants with moderate to severe HIE receiving TH, cerebral lactate concentrations are initially markedly elevated, show a slight decrease post-hypothermia, continue to decrease over time, and can still be detectable up to 2 weeks after birth.⁴¹ Conversely, lactate concentrations remain low in infants experiencing no or only mild HIE.⁴¹ The most common analytic approach involves using metabolite ratios in the basal ganglia and thalamus (BGT).^29,34,38,39 The concentrations of metabolites in the other regions of grey matter and white matter have also been studied⁴⁰ (Figure 5). Metabolite concentrations in newborns are age-dependent and vary from those in adults. Specifically, NAA and glutamate levels are lower during the neonatal period and rise during the first three months of life while myo-inositol levels fall. Beyond three months, these changes decelerate and tend to stabilize by around five years of age.⁴² Regarding the MRS acquisition techniques, single voxel and multi-voxel are the two main approaches. Single voxel MRS is the conventional choice for clinical MRS, allowing the acquisition of a single spectrum from a specific brain region in a relatively short time. Multi-voxel MRS, however, captures spectra from a larger tissue volume, providing the advantages of broader coverage and higher spatial resolution. However, multi-voxel MRS has the disadvantages of longer scan times and lower spectral quality due to less uniform shimming and lower signal-to-noise ratios in individual spectra.⁴² Currently, there is a lack of comparative studies evaluating the diagnostic accuracy between single voxel MRS versus multi-voxel MRS in neonatal HIE. In a large prospective multicenter study of infants who all underwent MRI and single voxel MRS in the first week of life, MRS was found to surpass structural MRI and DTI in predicting neurodevelopmental outcomes.³⁴ However,¹H-MRS is not without limitations: the necessity for predefining the investigative region could theoretically bypass injury in non-specified areas. Another constraint lies in its protracted acquisition time, although this can be mitigated by state-of-the-art imaging acceleration techniques.

Figure 5.

Magnetic resonance spectroscopy (MRS). Two proton magnetic resonance spectroscopy (1H-MRS) studies were performed on the basal ganglia using an intermediate TE of 144 ms within the first week post hypoxic ischaemic encephalopathy in two infants. The first MRS from an infant with normal MRI findings (A) demonstrated a typical N-acetylaspartate (NAA) peak at 2.0 ppm, a creatine peak at 3.0 ppm, and a choline peak at 3.2 ppm. Conversely, the second MRS from an infant with near-total brain injury (B) showed a diminished NAA peak at 2.0 ppm and a lactate peak, which was observed as an inverted peak at 1.33 ppm.

Predicting LTNO based on ¹H-MRS

Proton nuclear magnetic resonance spectroscopy has been shown to offer high sensitivity and specificity in predicting LTNO. However, it should be noted that the cut-off values depend on the parameters and scanners. In addition, some studies only performed group comparison without defining cut-off values. Several investigations have consistently found an association between the lactate/NAA ratio (>0.22³⁴) or the (lactate + threonine)/NAA ratio (>0.39³³) in the BGT and adverse outcomes.^9,20,33,34 Additional metabolite ratios tied to negative neurodevelopmental outcomes include decreased NAA/Choline (≤1.82), NAA/Creatine (≤0.67), and myo-inositol (mln)/creatine (≤0.76) ratios in the BG,²⁹ NAA concentration (≤5.6 mmol/kg wet weight),³⁴ and mln/NAA ratios (≥0.6798) in the thalamus.³⁸ In a prospective multicentre study of 223 infants, the NAA concentration in the thalamus(≤5.6 mmol/kg wet weight) alone had the highest predictive value of adverse neurocognitive outcome at a median age of 2 years [AUC, 0.99 (0.94-1.0)] compared to other metabolite ratios.³⁴ While Barta et al, in another study of 51 subjects, compared various parameters of ¹H-MRS at three echo-times and showed that mIns/NAA (≥0.6798) has the highest predictive value of adverse neurocognitive outcome (AUC, 0.9084; sensitivity: 84.62%, specificity: 95.24%),³⁸ other metabolite ratios such as NAA/Cr at TE = 144 ms (<0.627³⁸, ≤0.67²⁹) NAA/Cho (≤ 1.82²⁹), lactate/NAA,⁴³ (lactate+lipids)/NAA (≥3.33²⁹), mlns/creatine (>0.780³⁸, ≤0.76²⁹), and mIns/choline ratio (≤2.04²⁹) have also shown predictive values for adverse outcomes.

Additional techniques

Fluid-attenuated inversion recovery imaging is useful at identifying hyperintensities within the periventricular and deep white matter and is capable of detecting cystic leukomalacia associated with advanced HIE damage. However, it may not offer additional prognostic value over other imaging sequences.¹² Susceptibility-weighted imaging improves the detection of haemorrhages and calcifications. However, current research suggests that intracranial haemorrhages, despite being present in one-third of neonatal HIE cases, do not impact neurodevelopmental outcomes.⁴⁴ Arterial Spin Labelling (ASL) offers a non-invasive approach to assess cerebral perfusion using magnetically labelled water protons as intrinsic tracers. Hyperperfusion in affected areas is commonly seen in the onset of neonatal HIE, regardless of whether hypothermia treatment is administered.^45,46 Research with a small cohort (n = 28) indicates that increased perfusion in the BGT correlates with poor neurodevelopmental outcomes.⁴⁷

MRI injury scoring systems

There are a multitude of MRI-based scoring systems which serve as valuable tools to quantify the degree of brain injury and forecast future outcomes in neonates with HIE. Some of the widely recognized scoring systems, including those proposed by Barkovich,⁴⁸ National Institute of Child Health and Human Development (NICHD),¹⁷ and Rutherford,⁴⁹ leverage T1-weighted and T2-weighted imaging to scrutinize cerebral injuries. These scoring systems pay particular attention to specific regions, namely, the BGT, posterior limb of the internal capsule (PLIC), white matter (WM), and cortical areas. Supplementing these conventional imaging techniques, Bednarek et al proposed a scoring system that underscores the significance of DWI and assigns a heavier weightage to injuries present in the deep grey matter nuclei and the posterior limb of the internal capsule.^21,50 Weeke et al introduced a more comprehensive and intricate scoring system. This innovative approach combines the capabilities of traditional MRI techniques (including T1W, T2W, and DWI) with ¹H-MRS, leading to a nuanced and robust grading system.⁵¹ An expanded NICHD NRN score has been proposed to separate WM/WS patterns and BGT/PLIC injuries and also incorporate the severity and laterality of lesions in the NICHD Optimizing Cooling Strategies trial. Unpublished preliminary results indicate promising results in predicting LTNO.^52,53  Table 2 provides a detailed and comparative summary of different MR scoring systems.

Table 2.

Comparative Summary of MRI scoring systems.

Scores	Barkovich score	Rutherford score	NICHD NRN Score	Injury Score	Weeke Score
Area assessed	BGT, watershed (WM and cortex)	BGT, PLIC, WM, cortex	BGT, ALIC, PLIC, watershed	BG, WM, Cortex, cerebellum, brainstem	GM, WM, cerebellum, 1H-MRS of BG, IVH, SDH, CVST
Descriptions and characteristics	Categorizes injuries into BGT, watershed areas, a combination of both, and the total extent of injury	Includes PLIC injury as a distinct category, separate from BGT injury. No total score	Integrates ALIC and PLIC injuries under the category of BGT injury. BGT injury is classified as more severe than WM injury	Encompasses infratentorial structures, including the brainstem and cerebellum. Quantify T1W, T2W, and DWI independently for each hemisphere	Includes infratentorial structures. More detailed subscores Integrates findings from structural MRI, MRS, intracranial haemorrhages, and CVST
Levels	5 (0-4)	BGT (0-3), PLIC (0-2), WM (0-3), GM (0-3)	6 (0,1A&B, 2A&B, 3)	139 (48-186)	56 (0-55)
Reference	Barkovich et al48	Rutherford et al11	Shankaran et al17	Bednarek et al50	Weeke et al51

Abbreviations: ALIC = anterior limb of internal capsule, BGT = basal ganglia and thalamus, CVST = cerebral venous sinus thrombosis, DWI = Diffusion-weighted imaging, IVH = intraventricular haemorrhage, MRS = Magnetic Resonance Spectroscopy, NAA = N-Acetylaspartate, PLIC = posterior limb of internal capsule, SDH = subdural haemorrhage, T1W = T1-weighted imaging, T2W = T2-weighted imaging, WM = white matter.

Predicting LTNO based on injury scores

Multiple research studies have affirmed the close association between MRI injury scores, such as NICHD, Barkovich, Rutherford, Weeke, and Trivedi scores, and subsequent unfavourable neurodevelopmental outcomes after TH treatment.^{19,21,43,51,54,55} Notably, a comparative study scrutinizing the Barkovich, NICHD, and Weeke scoring systems found that all three systems provided reliable predictive values for cognitive and motor outcomes in HIE-affected infants at the age of two years. However, only the Weeke scoring system was significantly linked with language development scores at the same age.⁵⁶ A novel metric, WMxBGT, proposed by Thoresen et al. was proposed to be more predictive of adverse neurocognitive development than the total injury score.¹⁹

Long-term epilepsy outcomes in HIE

Perinatal Hypoxic-Ischaemic Encephalopathy (HIE) is a major precursor to neurological repercussions such as epilepsy. In newborns affected by HIE, neonatal seizures emerge as a common manifestation, with prevalence approximating 75%.^57,58 Notably, More than two-thirds of neonates with HIE have self-limiting seizures in the neonatal period and do not develop postneonatal epilepsy.^59–64 These seizures manifest in diverse forms, including infantile spasms, focal seizures, and generalized seizures.⁶⁵ TH has been purported to reduce the incidence of epilepsy subsequent to HIE.^66–68 However, a study contested this notion, indicating no discernible difference in epilepsy incidence between infants undergoing TH and those not undergoing this treatment.⁵⁸

MRI predictors of epilepsy outcome in HIE

The incidence of postneonatal epilepsy was low (<3%) in cases with either normal MRI or isolated cortical injuries.^58,62 Both the watershed and BGT injury patterns were associated with increased odds of epilepsy in later childhood, with the latter exhibiting a significantly higher risk.^58,62,65,69 In fact, over a third of newborns with the BGT injury pattern developed postneonatal epilepsy. Infants who survived brainstem and total injury were at the highest risk of developing infantile spasms, although a significant percentage of newborns did not survive these severe injuries.^62,69 These associations between injury patterns and the incidences of postneonatal epilepsy were not influenced by hypothermia treatment.⁵⁸ Injuries to specific brain regions, including the hippocampi, motor cortices, and occipital lobes, are also associated with an increased epilepsy risk. However, when adjusting for the prevailing watershed and basal ganglia injury patterns, these associations were no longer significant.⁵⁸ Coexisting MRI abnormalities indicative of hypoglycemic injury, such as abnormal signals involving the posterior white matter, optic radiations and pulvinars, were associated with a heightened risk of neonatal seizures, although they did not predict epilepsy in later childhood.^58,67

Effect of TH on imaging appearance

TH has demonstrated a positive trend in increasing the number of normal scans in infants following HIE. Moreover, it was shown to reduce the extent of infarction in infants with abnormal MRIs when compared to infants in the control group.^11,17 TH can impact the evolution of diffusion abnormalities. Infants treated with hypothermia demonstrate a phenomenon known as pseudonormalization of DWI at a later stage, typically around the 10th day of life, in contrast to untreated infants who exhibit this phenomenon between days 6 and 8.⁵⁰ Additionally, hypothermia has been observed to modify diffusion characteristics.⁷⁰ Furthermore, TH has been found to significantly influence metabolite concentrations across different brain regions. This includes increased energy reserves, such as phosphocreatine and free creatine concentrations, as well as decreased levels of excitatory neurotransmitters and cellular energy demand, such as glutamate, glutamine, GABA, and aspartate concentrations, when compared to metabolite levels after rewarming.⁷¹ Despite these alterations, recent analyses suggest that hypothermia treatment does not impact the prognostic value of MRI.⁹

Optimal timing of MRI

Recent evidence highlights the effectiveness of MRI within the first two weeks after birth as a reliable prognostic tool, potentially surpassing other neurophysiological evaluations like continuous video electroencephalography (EEG) and amplitude-integrated EEG in predicting long-term neurological outcomes.⁷ However, the optimal timing for MRI examination remains a subject of debate. The draft framework proposed by the British Association of Perinatal Medicine (BAPM) suggests that MRI should be conducted within the first 4-14 days of life and that early imaging (<6 days) has demonstrated the ability to accurately detect injuries.¹² In the 2nd edition of guidelines of Neonatal Encephalopathy and Neurologic Outcome (reaffirmed in 2019) published by the American College of Obstetrics and Gynecology and the American Academy of Pediatrics, a two-phase MRI approach is proposed for neonates undergoing TH. An early MRI (24-96 h) helps determine the injury onset, while a subsequent MRI (7-21 days) reveals the full extent of the injury.⁷² However, obtaining an early MRI within 24-96 h may not be feasible for patients undergoing TH during the initial 72 h. Studies comparing early (≤6 days) and late MRI (≥7 days) scans indicate that both have high sensitivity (100% in identifying unfavourable outcomes, but early MRI demonstrates higher specificity (96.3% vs. 89.3%)) due to its ability to detect mild abnormalities missed by later scans.³⁰ In the Total Body Hypothermia for Neonatal Encephalopathy (TOBY) trial, early MRIs (<8 days) were more effective at identifying major abnormalities compared to later scans (63% vs. 47% of cases).¹ A variety of MRI injury scores show significant agreement between early and late MRI (75%-88%), with early MRI exhibiting superior predictive value for adverse outcomes.^16,21,73 Two studies reported the MRI injury scores increased in 57%-66% of late MRI, and one reported decrease in 58% of late MRIs.^16,21,73 Meta-analyses indicate that early MRI (the first week) scores have better sensitivity, specificity, and diagnostic odds ratio values compared to later MRI (the second week) scans.⁹ Early MRI also prevents pseudonormalization of mean diffusivity observed on day 9 in cooled neonates.⁵⁰ Overall, these findings support the use of early MRI for prognosticating neonatal HIE.

HIE mimics

In addition to HIE, a variety of clinical conditions, such as vascular anomalies, perinatal stroke, metabolic diseases, infections, or congenital abnormalities, can lead to neonatal encephalopathy. It is crucial to recognize distinct patterns and severity in neuroimaging to determine the underlying cause. Neurometabolic disorders are rare but significant contributors to paediatric brain conditions and can mimic HIE in imaging. It should be considered in the differential diagnosis of HIE when the clinical presentation is ambiguous. The most common inherited metabolic disorders during the neonatal stage are mitochondrial diseases, urea cycle disorders, and organic acidemias.⁷⁴ Leigh syndrome, a most common neonatal-onset mitochondrial disease,⁷⁵ typically presents with bilateral symmetrical T2 hyperintensities in the basal ganglia, especially in the putamen and caudate nucleus.⁷⁶ Other susceptible areas include the thalamus, brainstem, medulla oblongata, cerebellum, and spinal cord.⁷⁶ Neonatal-onset organic acidemias are varied in imaging presentations and usually display symmetrical brain disease, impacting the bilateral cortex, white matter, or basal ganglia.⁷⁷

In neonatal-onset urea cycle disorders, two main neuroimaging patterns are recognized: diffuse and central. The diffuse pattern is the more severe form, widely involving the cerebral cortex and basal ganglia, and may occasionally affect the thalami and brainstem, which can make it challenging to distinguish from HIE.⁷⁸ Conversely, the central pattern is more specific to urea cycle disorders, characterized by symmetric edema—whether vasogenic, cytotoxic, or intramyelinic—typically located in the insular/periinsular, sylvian/perisylvian, perirolandic, and basal ganglia regions. A notable characteristic that differentiates it from HIE is the thalami typically remaining unaffected.⁷⁸

Conclusions and future directions

Brain MRI, which incorporates DWI, is considered the benchmark for evaluating brain injury in infants suffering from HIE. The patterns and scope of injury provide vital information about the timing and intensity of the hypoxic-ischaemic event, aiding in forecasting possible neurodevelopmental outcomes. BGT predominant and Near-total injury patterns are linked with the unfavourable LTNO and an increased likelihood of postneonatal epilepsy. An accurate characterization of these patterns is essential for providing valuable prognostic information. Several MRI scoring systems have been instituted to quantify brain injury using neonatal MRI. These methodologies have proven effective in predicting the neurodevelopmental outcomes of infants affected by HIE. ¹H-MRS has been shown to offer high predictive values in prognosticating newborns with HIE by providing crucial metabolic data.^33,34 However, the techniques and metabolite ratios involved in ¹H-MRS are not yet standardized. The methods for predicting LTNO using MRI are summarized in Table 3.

Table 3.

Overview of approaches to predict long-term neurodevelopmental outcomes using MRI.

Approaches	Predictors associated with unfavourable long-term neurodevelopmental outcome
Brain Injury Pattern	BGT pattern, near total injury pattern
Quantitative diffusion Imaging	Lower ADC values of several anatomic regions, such as centrum semiovale, BGT, PLIC The ADC values of PLIC have been reported to have the highest predictive value
Proton MRS	Various metabolite ratios, such as lactate/NAA, NAA/Cr, and NAA/Cho. The NAA concentration of thalamus and mlns/NAA ratio have been reported to have the highest predictive values
ASL	Increased perfusion in the BGT region
injury Score	The details of the most commonly used injury scoring systems are presented in Table 2. Higher scores correlate with less favourable outcomes.

Abbreviations: ADC = apparent diffusion coefficient, ASL = arterial spin labelling, BGT = basal ganglia and thalamus, Cho = choline, Cr = creatine, mln = myoinositol, MRS = magnetic resonance spectroscopy, NAA = N-Acetylaspartate, PLIC = posterior limb of internal capsule.

In terms of timing of imaging, it is considered optimal to conduct the MRI, including DWI and ¹H-MRS, immediately following TH, typically within the first week after birth. This is due to the peak of diffusion abnormalities occurring between 3 and 5 days post birth, with pseudonormalization of both DWI and ¹H-MRS occurring by the end of the first week.

Despite these advances, using MRI for outcome prediction in HIE is not without challenges. MRI interpretation can be subjective, and the application of previous research is often limited by the relatively small subject numbers and the heterogenenity of severity of the disease among subjects. This is compounded by differences in the percentages of various grades of encephalopathy and adverse outcomes in research.

Future studies could potentially address these limitations through multi-site studies and the application of advanced AI algorithms, which could offer a more objective and accurate interpretation of MR images.⁷⁹ The integration of MRI findings with other biomarkers and clinical data could also further enhance the accuracy of outcome prediction.

Funding

None declared.

Conflict of interest

The authors have no conflict of interest.