Voxel-based map of the inter-arterial watershed zones in children

Abstract

Purpose

To create a voxel-based map of the inter-arterial watershed derived from children who have sustained a hypoxic-ischemic injury involving this region at term.

Materials and methods

Patients 0–18 years of age diagnosed with a hypoxic-ischemic injury of the watershed on magnetic resonance imaging (MRI) were included. Two pediatric neuroradiologists segmented the lesions as visualized on the T2-weighted sequence. All lesion maps were normalized to a brain template and overlapped to create a frequency map in order to highlight the frequency of involvement of portions of the cortical watershed.

Results

A total of 47 patients (35 boys) were included in the final sample. Their mean age was 7.6 ± 3.6 years. The cortical watershed was successfully mapped. Three watershed regions were defined: the anterior, peri-Sylvian, and posterior watershed zones. The anterior and peri-Sylvian watershed zones are connected through the involvement of the middle frontal gyrus. The peri-Sylvian and the posterior watershed zones are connected through the involvement of the inferior parietal lobule, the posterior aspect of the superior temporal gyrus, and the angular gyrus with the occipital lobe. The temporal lobe and orbital part of the frontal lobe are largely spared in all patients.

Conclusion

A voxel-based lesion map of children with watershed hypoxic ischemic injury at term was created and three inter-arterial watershed zones defined: anterior, peri-Sylvian, and posterior watersheds.

Introduction

Hypoxic-ischemic injury (HII) is a devastating cause of disability and death in children around the world. It is estimated that HII affects around two infants per 1000 live births in western Europe and North America, and as many as 25% of the survivors will have a sequela.^1,2 These figures may be even higher in developing countries. ¹ Neurological outcome and imaging findings depend on the maturity of the brain at the time of the injury and on the timing, duration, and severity of the insult. ³ When partial hypoxia with a prolonged duration occurs in an infant delivered at term, the brain shunts the blood flow from the anterior and posterior circulation to perfuse adjacent vital highly metabolic structures such as the brainstem, basal ganglia, and cerebellum.^3,4 As a result, the inter-arterial watershed (or boundary) zones (the end zones of the vessels) are commonly affected given their inferior supply from the cerebral arteries.^5,6

Identifying partial-prolonged HII involving the watershed and distinguishing it from differential diagnoses represents a medical challenge. Clinical, electroencephalographic, and biochemical findings are not accurate as their sensitivity and specificity depend on multiple factors associated with the anoxic injury.^7,8 The diagnosis of partial-prolonged HII has not only clinical but also medico-legal implications. ⁹ In clinical practice, distinguishing bilateral infarctions in arterial territories from watershed injury has significant management implications. In malpractice cases, information regarding the timing, severity, and duration of the injury, which can be derived from magnetic resonance imaging (MRI) scans, are relevant to the judge and jury as a partial-prolonged HII is assumed to be preventable.^3,10 MRI remains as the most accurate tool to diagnose and assess HII in children.^8,10 Different patterns of HII detected on MRI suggest the severity and duration of the insult that caused them.

The inter-arterial watershed zones of the brain have been mapped between the major cerebral arterial territories in adults who have experienced arterial territory stroke. ¹¹ However, these maps may not accurately represent the watershed zones of children with HII, because these represent the susceptibility and metabolic demand of different brain regions in the perinatal period. ¹² The watershed zones in children who have sustained partial-prolonged HII have been described;^12–14 however, the literature lacks a precise topographic demonstration and characterization of the location of the watershed zone in the pediatric population. The purposes of this study were to create and describe a voxel-based lesion map of the watershed regions with color coding to denote the frequency of the region’s involvement in the injury in children with HII.

Materials and methods

This was a retrospective study approved by the University of Bristol Institutional Review Board (Faculty of Science Human Research Ethics Committee, approval code 26111527741) and has been performed in accordance with the ethical standards described in the 1964 Declaration of Helsinki. Informed consent was obtained from the parent or legal guardian by one of the investigators.

Study population

A computer search of an MRI database of children with cerebral palsy spanning the years 2014–2016 was performed to identify patients diagnosed with a HII watershed pattern or combined watershed basal ganglia-thalamus on MRI. This radiological database consists of medico-legal cases of patients with a confirmed diagnosis of cerebral palsy but no further clinical information was available. Our database also excluded cases with a sentinel event, given that these cases are usually not considered to be preventable and are not associated with medical negligence. ¹⁵ MRI scans were acquired post-natally at different ages as part of the initial work-up or follow-up once the diagnosis was already confirmed. Excluded from the study were patients with severe ventriculomegaly, extensive multicystic encephalomalacia, extensive involvement of HII (beyond the watershed regions and involving large portions of the hemispheres), patients with only basal ganglia-thalamus pattern on MRI, images of suboptimal quality, localized injury, or no definitive HII on second look. Patients with extensive injury involving large portions of the cerebral hemispheres, beyond the watershed, were excluded because this study aimed to map the topography of the inter-arterial watershed rather than the extent of disease from severe HII. Similarly excluded was global severe hypoxic injury associated with significant brain atrophy which can distort the anatomical landmarks and impair the normalization of the brain maps. Also excluded were patients with massive and severe ventriculomegaly, because large anatomical alterations affect the accuracy of voxel-based lesion mapping.

MRI protocol

Brain MRI scans were performed using a variety of 1.5 T scanners at different institutions using a standardized protocol of T1, T2, fluid-attenuated inversion recovery (FLAIR), diffusion-weighted imaging (DWI), and gradient-echo/susceptibility-weighted imaging (SWI) sequences performed on sagittal, axial, and coronal planes. Thirty-eight MRI scans included in the final sample (38/47, 80%) were performed using one type of scanner (Intera; Philips Healthcare, Best, The Netherlands). In general, the T1-weighted three-dimensional (3D) turbo spin-echo sequences were performed with the following imaging parameters: 4.5/25 TE/TR, 30° flip angle, 265 × 265 field of view, 232 × 232 matrix, 1 mm slice thickness, 1 mm slice spacing. In general, the T2-weighted sequences were performed with the following imaging parameters: 100/4800 (TE/TR), 90° flip angle, 230 × 230 field of view, 384 × 305 matrix, 5 mm slice thickness, 6 mm slice spacing. No intravenous contrast was used.

Lesion segmentation

The process of segmentation and mapping pipeline is summarized in Supplementary Figure E1. All digital imaging and communications in medicine (DICOM) images were converted to neuroimaging informatics technology initiative (NIfTI) format using dcm2nii software (https://people.cas.sc.edu/rorden/mricron/dcm2nii.html). Two pediatric neuroradiologists (FGG with 11 years of experience and SA with more than 20 years of experience) manually delineated the boundaries of the peripheral cortical and subcortical injury as visualized on the T2-weighted sequence in consensus, using the open-source software ITK-SNAP (http://www.itksnap.org). ¹⁶ The peripheral watershed was delineated bilaterally in the cortical and subcortical areas of abnormal T2 signal and/or signs of atrophy, especially when associated with ulegyria. Ulegyria was defined as preferential atrophy at the depth of the sulci and sparing of the crest of the gyri with the resultant ‘mushroom shape’. ¹⁴ Lesions in the basal ganglia were not considered part of the peripheral watershed pattern of injury and were excluded from the manual delineation process, as were any signal or atrophy of the posterior fossa structures – the goal being to map the inter-arterial watershed rather than areas affected by HII.

MRI post-processing and lesion mapping

The MRI co-registration and normalization were performed using open-source software SPM12 (Statistical Parametric Mapping 12, 2014; Functional Imaging Laboratory, Wellcome Trust Centre for Neuroimaging–Institute of Neurology, University College London, UK) running on MATLAB (R2019a; MathWorks, Natick, MA, USA). The lesion maps were co-registered first from the T2-weighted (pathological scan) to the T1-weighted (anatomical scan) sequence and then to a 1.0-mm isotropic, high resolution, T1-weighted age-appropriate brain atlas provided by the Montreal Neurological Institute (MNI; 152 spaces) using the open-source clinical toolbox tool (2017, Department of Psychology, University of South Carolina, Columbia, SC, USA) ¹⁷ with an enantiomorphic normalization. ¹⁸ After the post-processing, each normalized lesion map was reviewed after the post-processing alongside the original MRI by two of the investigators. Minor adjustments to the lesion maps were performed if deemed appropriate, using SPM12 and the open-source MRIcroGL software (2019, University of South Carolina, Columbia, SC, USA; https://www.mccauslandcenter.sc.edu/mricrogl/). All normalized lesion maps were superimposed and then color-coded according to the frequency of involvement, resulting in a frequency map for visualization using MRIcroGL. Voxel-based lesion frequency maps were created as mosaic figures (i.e. over several cross-sectional slices) in multiple planes for demonstration of both the deep watershed and the cortical watershed. A variety of color maps was obtained using different frequency thresholds (windowing) to display both the full extent and the most frequently affected portions of the watershed in the patient population. These frequency thresholds can be appreciated at the lower end of each figure. 3D-rendered models were also created to highlight watershed as seen over the surface of the brain. Objective anatomical interpretation of the frequency maps was carried out using MNI coordinates and by translating these into an automated Talairach atlas. ¹⁹ An adult atlas was used in order to homogenize the different lesions into one size and given that the algorithm used to normalize the lesions was developed in adults. Continuous variables were presented as mean ± standard deviation. No statistical analysis was performed as the goal of this study is purely descriptive.

Results

Study population

A total of 114 patients from studies performed from the 2-year period met the criteria of watershed or combined watershed basal ganglia-thalamus HII diagnosed on MRI. Of these, 57 were excluded according to the predetermined criteria: moderate or severe ventriculomegaly (n = 25, 44%), multicystic encephalomalacia (n = 12, 21%), extensive injury (n = 7, 12%), imaging findings consistent with basal-ganglia-thalamus pattern on second look (n = 7, 12%), suboptimal image quality (n = 3, 5%), no definitive imaging findings of watershed injury on the second look (n = 2, 4%), localized injury (n = 1, 2%). Out of the 57 remaining cases, 10 patients failed the normalization process due to significant anatomical distortion, secondary to the brain atrophy and head deformity, and were subsequently excluded, Figure 1. Forty-seven patients (12 girls, 35 boys) made up the final study population. Their mean age was 7.6 ± 3.6 (range 1–14) years.

Figure 1.

Flowchart of inclusion and exclusion criteria used to select our final study sample.

Only nine of the 47 (19%) MRI scans were classified as watershed pattern alone. The remaining 38 patients had combination patterns of injury as follows: 72% (34/47) had the pattern of watershed HII combined with basal ganglia-thalamus injury, and the other 9% (4/47) had the watershed HII pattern combined with occipital and thalamic pulvinar injury suggesting superadded hypoglycemia.

Voxel-based lesion map and 3D model

Two voxel-based lesion maps were created using different frequency windowing. Figure 2(a) was set with a window frequency range of 5–25 (the least frequently involved areas required at least five patients to be designated the color blue; the most frequently involved area required involvement of 25 patients to be designated the color red). The upper limit was set as 25, given that no brain region showed an overlapping involvement in more than 30 patients (Figure 3). Figure 2(b) is the result of using a narrower window set at a higher threshold, representing only the most commonly affected brain regions, with a color range from blue to red. Figure 4 displays the lesion maps in the sagittal (Figure 4(a)) and coronal planes on wide windows (Figure 4(b)).

Figure 2.

Voxel-based lesion map of the watershed zones in the axial plane in children (n = 47) with watershed injury, using a wide window with a low-setting upper limit (a) and a narrow window with a higher-setting upper limit (b) to highlight the least (dark blue) and most (yellow–red) frequently involved areas.

Figure 3.

Voxel-based lesion map of the watershed zones in the axial plane in children (n = 47) with watershed injury, using a window that includes the entire sample. The absence of red areas implies that there is no brain region present in all 47 cases with watershed injury.

Figure 4.

Voxel-based lesion map of the watershed zones involved in children (n = 47) with watershed injury using a wide window with a low-setting upper limit in the sagittal plane (a) and in the coronal plane (b) to demonstrate anatomical landmarks of the least (dark blue) and most (yellow–red) frequently involved areas.

Three watershed regions emerged from the frequency mapping: the anterior, peri-Sylvian, and posterior watershed zones and these are defined anatomically below. Frequently involved portions of the anterior watershed (yellow–red areas) extend anteriorly to the superior frontal gyrus and posteriorly through the paracentral lobule, pre and postcentral gyri up to the marginal sulcus. Superiorly there is involvement of the superior frontal gyrus and inferiorly involvement up to and including the cingulate sulcus – also affecting predominantly the inferior part of the paracentral lobule. The anterior watershed also extends laterally to involve the superior frontal sulci and, medially, up to the interhemispheric fissure (the medial frontal gyrus). Other less affected anterior watershed regions (green areas) include, laterally, the middle frontal gyrus and, inferiorly, the cingulate gyrus. There were two distinctive patterns within the anterior watershed: along the superior frontal sulcus (parasagittal) and along the interhemispheric fissure (superomedial) including the medial frontal gyrus and cingulate sulcus.

The frequently involved portions of the peri-Sylvian watershed (yellow–red areas) extend anteriorly from the inferior frontal gyrus to the supramarginal gyrus. There is involvement of the inferior aspect of the Rolandic fissure superiorly, and both pre and postcentral gyri, inferiorly. There is involvement up to the Sylvian fissure, laterally, and the insula, medially. Other less frequently involved regions (green areas) include the superior aspect of the Rolandic fissure up to the vertex (perirolandic), the inferior parietal lobule, inferiorly, and the middle and inferior frontal gyri, anteriorly.

The most frequently involved portions of the posterior watershed (yellow–red areas) extend anteriorly from the cuneus to the occipital cortex, posteriorly. There is involvement from the parieto-occipital sulcus, superiorly, to the calcarine fissure, inferiorly, and from the middle occipital gyrus, laterally, to the interhemispheric fissure, medially. Other less affected regions (green areas) include the precuneus, superiorly, and the inferior occipital gyrus, inferiorly, with sparing of the cortex just superior to the tentorium.

The three watershed zones described above demonstrate continuity with each other in limited frequency, Figure 5. The anterior and peri-Sylvian watershed zones are connected through the involvement of the middle frontal gyrus (green area). The peri-Sylvian and the posterior watershed zones are connected through the involvement of the inferior parietal lobule, the posterior aspect of the superior temporal gyrus, and the angular gyrus with the occipital lobe. The superior parietal lobe is included partly by the peri-Sylvian zone (anteriorly) and posterior watershed (posteriorly) in a minimal degree. The temporal lobe and orbital part of the frontal lobe are largely spared in all patients (Figure 2), which is well demonstrated on the 3D models (Figure 5). Visually, the left hemisphere is more frequently affected than the right. Supplementary Figure E2 shows a voxel-based lesion map including only the nine patients with only a watershed pattern, which displays involvement of the anterior, peri-Sylvian, and posterior watershed zones.

Figure 5.

Three-dimensional (3D) rendering of the voxel-based lesion frequency maps (n = 47) in wide window with a low-setting upper limit of the watershed zones in children with cerebral palsy according to the frequency color scale provided, with blue indicating the least frequently and red the most frequently injured regions. The orientation of the models is described by a cube in the lower right corner of each image. A: anterior; R: right; L: left; S: superior; P: posterior.

Discussion

When there is a HII to a term infant that is mild to moderate and prolonged, the pattern of brain injury is described as peripheral – rather than central or deep nuclear – and involves the inter-arterial watershed to different degrees. We successfully created a voxel-based lesion frequency map of the watershed regions in children with cerebral palsy who sustained HII at term, and three distinctive regions of involvement emerged: anterior, peri-Sylvian, and posterior watersheds. There was sparing of the temporal lobe, the orbital part of the frontal lobe, and the portion of the occipital lobe just superior to the tentorium cerebelli. The ability of a radiologist to recognize the pattern of distribution that represents a watershed injury, especially when only portions of this are involved, is of importance in the context of perinatal imaging and other causes of HII for diagnosis, management, and prognosis of both acute and chronic stages. Demonstration of the watershed injury is of paramount importance for diagnosing HII in term neonates and may also become important during litigation as partial-prolonged HII may be deemed preventable. ²⁰

The important finding from mapping the watershed in our patients is that no brain injury pattern was identified consistently in every patient. Patients may present with injury in any of the three watershed zones, or the injury may extend beyond the limits described in this paper, into the deep arterial territories. The sparing of the orbital part of the frontal lobe and the temporal lobe parenchyma around the Sylvian fissure is an important clue to differentiate watershed injury from bilateral arterial territory infarcts and other causes of bilateral cerebral necrosis. Similarly, the posterior watershed territory can be distinguished from other types of lesions by the sparing of the calcarine cortex and the inferior occipital cortex superior to the tentorium, although this distinction may be blurred when there is superadded neonatal hypoglycemic injury. A further variation of lesions according to the severity of the injury is also evident in the parasagittal and superomedial patterns involving the anterior watershed territory. Parasagittal cerebral injury is reported as being characteristic of a milder form of HII, whereas more severe insults extend to the interhemispheric fissure (superomedial) and can manifest as bi-frontal atrophy.^12,14,21,22 Although the involvement of the perirolandic cortex is classically associated with acute-profound/deep nuclei HII, repetitive short episodes of hypoxia over a longer period can also injure this region, sometimes alongside lesions of the other watershed regions. ²³ The involvement of the perirolandic is minimal but evident in our maps, which could be explained by either the inclusion of patient with basal ganglia-thalamus pattern or as part of the watershed zone in a small proportion of patients.

We have demonstrated, through the creation of a patient-based atlas, that the inter-arterial watershed is a continuous zone that extends from the anterior portions of the hemispheres through the parasagittal and peri-Sylvian regions to the posterior portion of the hemispheres. This may not be evident from a single case or from a small series of cases that involve one area more than another. Previous publications, using different visual techniques, have also displayed the watershed as a continuum without indicating frequency of involvement of any of the three isolated zones making up the watershed.^22,24 According to our lesion maps and 3D models, the anterior and peri-Sylvian watershed are sometimes connected by the middle frontal gyrus, and the peri-Sylvian and the posterior watershed are sometimes connected by the parieto-occipital region. However, according to our maps, the watershed is rarely involved as one continuous zone, except in those severe injuries which were excluded from this research that showed extensive injury involving the large portions of the arterial territories themselves. Although the parieto-occipital and posterior regions have been described as more frequently affected in children with partial-prolonged HII, ²⁵ our maps demonstrated all three watershed regions being affected with equal frequency.

Limitations include the fact that this was a retrospective study with a small sample of subjects. Given the purpose of this study, to map the watershed in children (rather than demonstrate patterns and extent of HII), many patients with extensive cerebral injury involving large portions of the arterial zones were excluded, limiting the sample size further. The inclusion of combined watershed and basal ganglia-thalamic injury was necessary because of the relative rarity of isolated watershed injury in the overall sample. More research with a more isolated (i.e. exclusively of patients with prolonged HII) and larger sample is needed for further characterization of the watershed zones. In addition, co-registration and normalization were performed using an algorithm and atlas initially created for adults, which limits the accuracy of our map in children and the reproducibility of our results. Image distortion due to brain size differences and atrophy between patients should be considered as a factor affecting image co-registration and normalization. We mitigated this by excluding most patients who were likely to yield a failed normalization, by reviewing each normalized lesion map against the original, and by making minor adjustments where necessary to account for these drawbacks. More patients are needed to perform a more accurate map and ideally should be made using different atlases based on the patient’s age. Our goal was to bring to light some of the potential lesion patterns seen in children with prolonged HII for future studies to come. Second, this was a purely descriptive study with the goal of displaying the different patterns seen in partial prolonged HII. Our database did not have clinical data that was readily available to use for analysis. Finally, we used MRI scans performed long after the perinatal injury, with the assumption that the imaging findings would be equally or more conspicuously demonstrated when relying on recognition of both signal abnormality and atrophy (with ulegyria) on conventional sequences. However, the retrospective nature of the study confines us to a wide age range (1–14 years) at the time of scanning, i.e. after the time of injury. A planned study in a larger patient group would provide some standardization of patient age at the time of the MRI scan, and would also assist with the technical aspects of mapping.

Conclusion

A voxel-based lesion frequency map of children with watershed HII on MRI was created to demonstrate the full spectrum of involvement of each watershed zone (anterior, peri-Sylvian, and posterior watershed) and assist radiologists in making the diagnosis.