Key words
brain - areas - structures & systems - ultrasound - methods & techniques - hypoxia - themes
Introduction
Neurosonography is the initial imaging modality of choice in the evaluation of
neonatal hypoxic-ischemic encephalopathy (HIE). This evaluation, which has been
mainly performed qualitatively by comparing intracranial structures with different
echogenicities, requires visual recognition of patterns by trained radiologists
[1]. Qualitative neurosonography has proven
reliable in diagnosing focal intracerebral lesions such as intraventricular
hemorrhage, periventricular hemorrhagic infarction, and cystic periventricular
leukomalacia [2].
However, diffuse processes resulting from neonatal HIE, such as white matter (WM)
edema, are more challenging to recognize and seem to be more dependent on the
expertise of the radiologist. It is well known that WM edema is depicted in
neurosonography images as areas of increased echogenicity as well as enhanced or
diminished gray-WM differentiation [3]
[4]
[5]. The echogenicity
of WM is typically higher than that of the less echogenic and more compact cortical
gray matter [6], but the accentuation or reversal of
this difference in echogenicity can be seen in evolving HIE. Nevertheless, since
cerebral edema takes time to evolve, qualitative neurosonography can be
inconspicuous and negative during the first 24 to 48 hours following HIE.
Increased echogenicity commonly found in the deep brain structures such as the basal
ganglia and thalami may not be evident until approximately 2 to 3 days after HIE.
Moreover, increased cortical and subcortical echogenicity changes may even take 5 to
7 days to become conspicuous [2].
Quantitative neurosonography is not performed routinely in clinical practice,
particularly in cases of neonatal HIE. Quantitative neurosonography has been
previously attempted in infants by Barr et al. [7],
Simaeys et al. [8], Vansteenkiste et al. [9], and Padilla et al. [10], although these infants were rarely term.
We hypothesized that quantitative neurosonography could demonstrate differences in
echogenicity in term HIE patients compared to controls during the first days
following injury. If this is correct, quantitative neurosonography may increase the
sensitivity of neurosonography in the early detection of HIE. Therefore, our study
aimed to compare the echogenicity measurements of several brain regions in patients
with HIE with those of the same regions in healthy controls.
Methods
Study design
Retrospective study performed in an institutional review board-approved and
HIPAA-compliant manner.
Participants
Two groups of term neonates were retrospectively identified from the period Jan
2018 to Dec 2019 in a major tertiary pediatric hospital. The first group (HIE
group) was composed of 20 neonates (15 males and 5 females). The number
of individuals in both the HIE and control groups was based on similar prior
studies. The diagnosis of HIE was made by combining clinical history, blood gas
measurement, and magnetic resonance imaging (MRI) findings consistent with HIE.
Numbers are expressed as mean±standard deviation (SD). Only patients
with MRI reports consistent with HIE were included. Neurosonography was
performed at 0.40±0.60 days of life. Images correlating neurosonography
and MRI findings from two patients (
[Figs.
1]
[2]
) and from the group HIE
(
[Figs. 3]
[4]
[5]) are shown. All patients in the
HIE group underwent hypothermia treatment at 0.4±0.49 days of life. MRI
examinations were performed at 5.1±1.48 days of life. APGAR (1, 5, and
10 minutes), pH and basis excess (arterial blood gas), age at
neurosonography, age at hypothermia treatment, age at MRI, and short-term
survival outcomes (by the end of the study period) of all patients in the HIE
group can be found in [Table 1].
Fig. 1 Diffusely increased echogenicity involving the white matter
(white arrows) and bilateral basal ganglia/thalami (white arrowheads) in
a patient with HIE.
Fig. 2 Signal changes involving the bilateral basal ganglia (white
arrows) and thalami (white arrowheads) in a neonate with
hypoxic-ischemic encephalopathy, with increased T1-weighted signal in a,
increase T2-weighted signal in b, and increased DWI signal in
c.
Fig. 3 Diffusely increased echogenicity involving the white matter
(white arrows) in a, b, c and also in the bilateral
basal ganglia and thalami (white arrowheads) in a.
Fig. 4 Diffuse and bilateral signal changes in the basal ganglia,
corpus callosum, cortex, and posterior limbs of the internal capsules
(PLIC). In a, increased T1-weighted signal in the bilateral PLIC.
In b, increased T2-weighted signal in the bilateral basal
ganglia. In c and d, restricted diffusion in the
perisylvian cortex and basal ganglia (white arrows) and splenium of the
corpus callosum (white arrowhead). Corresponding ADC map demonstrates
dark signal in these regions (black arrows).
Fig. 5 Bilateral signal changes in the perirolandic cortex. In
a, increased T1-weighted signal, no evident signal changes in
the T2-weighted images in b, with evident DWI signal changes in
c, and restricted diffusion in d.
Table 1 APGAR at 1, 5, and 10 minutes, pH, basis
excess, age at neurosonography, age at hypothermia treatment, age at
magnetic resonance imaging, and short-term survival outcomes of all
patients in the hypoxic ischemic encephalopathy group. Ages are
shown in days.
|
CASE
|
APGAR 1, 5, 10
|
pH
|
BASIS EXCESS
|
AGE AT US
|
AGE AT HYPOTHERMIA
|
AGE AT MRI
|
ALIVE
|
|
1
|
2/4/7
|
6.95
|
−17
|
1
|
1
|
5
|
YES
|
|
2
|
2/3/2
|
7
|
−7
|
0
|
1
|
3
|
NO
|
|
3
|
1/7/8
|
7
|
−21
|
1
|
1
|
7
|
YES
|
|
4
|
0/0/0
|
7
|
−23
|
0
|
0
|
5
|
NO
|
|
5
|
1/3/7
|
6.9
|
−18
|
1
|
0
|
7
|
YES
|
|
6
|
1/1/7
|
7
|
−15
|
0
|
0
|
6
|
YES
|
|
7
|
4/5/6
|
6.4
|
−15
|
0
|
1
|
5
|
YES
|
|
8
|
1/5/6
|
7
|
−16
|
1
|
1
|
7
|
YES
|
|
9
|
1/2/6
|
7
|
−20
|
0
|
0
|
4
|
YES
|
|
10
|
2/3/4
|
6.97
|
−19
|
0
|
1
|
6
|
NO
|
|
11
|
6/7/9
|
7
|
−14
|
1
|
0
|
7
|
YES
|
|
12
|
1/5/8
|
6.89
|
−23
|
0
|
0
|
5
|
YES
|
|
13
|
1/2/6
|
6.9
|
−23
|
0
|
1
|
3
|
YES
|
|
14
|
2/3/5
|
6.97
|
−17
|
0
|
0
|
6
|
YES
|
|
15
|
2/3/2
|
6.97
|
−18
|
0
|
0
|
7
|
YES
|
|
16
|
2/5/6
|
6.95
|
−18
|
0
|
0
|
5
|
YES
|
|
17
|
0/1/1
|
6.97
|
−20
|
0
|
0
|
4
|
YES
|
|
18
|
NA/4/9
|
6.9
|
−15
|
0
|
0
|
2
|
YES
|
|
19
|
2/6/7
|
7
|
−16
|
2
|
1
|
4
|
YES
|
|
20
|
0/2/4
|
7
|
−13
|
1
|
0
|
4
|
YES
|
NA - not available.
The second (control group) was composed of 20 healthy term neonates (11
males and 9 females). None of the neonates had any clinical evidence of HIE.
Blood gas measurement during labor and delivery showed no signs of hypoxia.
Neurosonography examinations were performed at 3.50±2.72 days of life.
Indications for ultrasound in the 20 healthy newborns were increased head
circumference (50%), retrocerebellar cyst on previous gestational
ultrasound (15%), large anterior fontanelle (10%), hypoplastic
vermis on gestational MRI (10%), hyperreflexia (5%), ventricular
asymmetry (5%), and ventricular cyst (5%).
The exclusion criteria included congenital heart diseases, chromosomal
abnormalities, absence of MRI, (or, in the HIE group, no evidence of injury on
their MRI), congenital infections, multiple pregnancies, lack of an
anteroposterior scan from the neurosonography study, anteroposterior scan not
involving all the brain from the frontal lobes to the occipital lobes,
hydrocephalus, intracranial hemorrhage, and cephalohematoma. Control infants
meeting these criteria with similar ages to the HIE neonates were challenging to
enroll, and therefore scans up to 10 days of life were included for the control
neonates.
Neurosonography Protocol
Neurosonography examinations were performed using one of two scanners, one
Philips and one GE, each with a 5–8 MHz micro convex transducer
using the anterior fontanelle as the acoustic window. Examinations were
performed by four ultrasonographers with at least ten years of experience each.
The examinations included sagittal and coronal standard still images and
continuous anteroposterior scan in the coronal plane with the same settings of
the ultrasound machine during the scan. This continuous coronal neurosonographic
acquisition included all brain structures from the frontal to the occipital
poles in the coronal plane.
Post-Processing and Data Collection
Neurosonography images were post-processed by Parametric MRI (pMRI) software
available for common use and downloaded from
https://www.parametricmri.com. Images were imported
from an online PACS server or offline DICOM files and arranged by study and
series in searchable tables. Selected series were loaded for analysis. Regions
of interest were placed over grayscale images. In the case of ultrasound images,
mean grayscale values were computed and stored automatically. Results were
exported in DICOM or tabulated text format per case or in batch. Video tutorials
on the use of the software, including loading data and image segmentation, are
available on the website listed above.
Brain regions of interest were segmented (frontal, parietal, parieto-occipital,
and perirolandic WM, caudate nucleus head, lentiform nucleus, and thalamus), and
their internal mean grayscale values were determined with segmented regions of
bone and the choroid plexus (CP) of the lateral ventricle atria as internal
controls (
[Fig. 6a–g]
).
Fig. 6 Segmentation of brain regions of interest that were
post-processed by the parametric software: a) frontal white
matter, b) caudate nucleus head, c) lentiform nucleus,
d) thalamus, e) parietal white matter, f)
parieto-occipital white matter, and g) perirolandic white matter.
For the internal controls, the echogenicity of the calvarial bones was
measured at the level of the thalamus h), and the echogenicity of
the choroid plexus was measured at the level of the glomus
i).
The mean grayscale value, which will be referred to as echogenicity, of the
different brain regions was measured as follows:
-
Frontal WM, two slices before the rostral portion of the frontal horn to
avoid volume averaging (
[Fig.
6a])
-
Caudate nucleus head, immediately before the level of the caudothalamic
groove (
[Fig. 6b]
)
-
Lentiform nucleus at the level of the middle cerebral artery
(
[Fig. 6c])
-
Thalamus, two slices before the quadrigeminal plate (
[Fig. 6d])
-
Parietal WM, at the level of the thalamus (
[Fig. 6e]
)
-
Parieto-occipital WM, two slices after the caudal portion of the
occipital horn, to avoid volume averaging (
[Fig. 6f])
-
Perirolandic WM, at the level of the pars marginalis
(
[Fig. 6g]
)
-
Bone, at the level of the thalamus (
[Fig.
6h])
-
CP, the level of the glomus (
[Fig.
6i]
).
Since the frontal and the perirolandic WM are the farest anterior and posterior
planes of a neurosonography examination, they may be challenging to measure,
especially in infants with a limited acoustic window due to a narrow or
partially closed fontanelle.
The echogenicity ratios of the brain regions of interest to the bone and CP were
calculated. The rationale for using two different ratios was exploratory and
based on the understanding that these two structures represent the most
echogenic regions in the field of view. HIE and intraventricular hemorrhage can
coexist in up to 10% of cases, being more prevalent in those treated
with hypothermia [11]. In the event of
intraventricular hemorrhage, the CP ratio calculation may be artefactual. For
intraobserver reliability evaluation, each segmentation was performed twice,
approximately a month apart, by the same user (FGG), a fellowship-trained
neuroradiologist with ten years of experience.
Analysis
Statistical testing was performed using R: A language and environment for
statistical computing version 3.6.1. The alpha error was set to 0.05.
Demographics were compared using proportion tests, including the proportion of
the control and HIE groups with Cesarean and vaginal deliveries and the
proportion of males and females.
A permutational multivariate analysis of variance using distance matrices
(permANOVA) was used to analyze the effect of several factors on the
echogenicity ratios for each brain region of interest to CP and bone. These
factors were: patient group (control versus HIE), age at neurosonography, sex,
length of pregnancy, and delivery type (vaginal versus Cesarean). Multiple
post-hoc two-sample permutation tests were performed with the Hommel adjustment
for multiple comparisons to control for type I error. Pearson correlation
coefficient was calculated for the two replicates of each ratio to evaluate
intra-observer reliability.
The empirical receiver operating characteristic curves were calculated for each
ratio. The area under the curve was calculated for each receiver operating
characteristic curve.
Results
Participants
The neonates in the HIE group were born between 37 weeks, 2 days and 44 weeks of
gestation (39.7 weeks±11.5 days) with a birth weight of
3495±555 g. Cesarean section was performed in 14 and vaginal
delivery in 6 pregnancies. The neonates in the control group were born between
37 weeks, 5 days and 41 weeks of gestation (39.3 weeks±6.5 days) with a
birth weight of 3336.7 g±403. In the control group, cesarean
section was performed in 8 and vaginal delivery in 12 pregnancies. The
echogenicity in the frontal and perirolandic WM was not measured in 4 patients
due to technical challenges, such as motion artifacts and lack of acoustic
window.
Test results
There was no significant difference between the proportions of vaginal and
cesarean deliveries in the control and HIE groups (p=0.057) or
between the proportions of males and females in the control and HIE groups
(p=0.185).
Patient group (control versus HIE) and delivery type (vaginal versus cesarean)
significantly affected the brain region echogenicity ratios
(p<0.001 and p=0.046, respectively; [Table 2]). Echogenicity ratios were significantly
lower for control neonates than for HIE neonates in each brain region
(
[Table 3]
;
[Fig. 2]
). The differences in ratios between
the control and HIE subject ratios were smallest for the caudate and putamen
ratios. Caudate and putamen to CP ratios were both 0.08 greater in HIE neonates,
and caudate and putamen to bone ratios were both 0.10 greater in the HIE
subjects. The differences were greatest for the perirolandic WM region, with the
ratios to CP and bone being 0.23 and 0.22 greater, respectively, for the HIE
patients compared to the controls ([Fig. 7]).
Although delivery type had a significant p-value in our overall multivariate
analysis, when pairwise permutation tests were performed comparing vaginal and
cesarean deliveries within the HIE and control groups, there were no significant
differences.
Fig. 7 Mean echogenicity ratios for each brain region to the
echogenicity of two internal controls, the choroid plexus (CP), and
calvarial bone (B), for both healthy control infants and infants with a
diagnosis of hypoxic-ischemic encephalopathy. HIE – Hypoxic
ischemic encephalopathy LENTNUC – Lentiform nucleus OCCWM
– Occipital white matter PARWM – Parietal white matter
FRTWM – Frontal white matter
Table 2 Results of permutational multifactorial analysis
of variance evaluating the effects of multiple factors group
(control versus hypoxic-ischemic injury), subject age, length of
pregnancy, and delivery type (vaginal versus Cesarean) on
echogenicity ratios of brain regions of interest to the choroid
plexus and calvarial bone.
|
F
|
p-value
|
|
Group
|
40.594
|
0.001***
|
|
Age
|
1.177
|
0.290
|
|
Gender
|
2.537
|
0.104
|
|
Length
|
0.737
|
0.460
|
|
Delivery
|
3.453
|
0.046*
|
|
Group:Age
|
0.852
|
0.421
|
|
Group:Sex
|
0.900
|
0.404
|
|
Age:Sex
|
1.402
|
0.270
|
|
Group:Length
|
1.910
|
0.153
|
|
Age:Length
|
2.786
|
0.087
|
|
Sex:Length
|
1.696
|
0.192
|
|
Group:Delivery
|
1.018
|
0.323
|
|
Age:Delivery
|
3.351
|
0.046*
|
|
Sex:Delivery
|
1.949
|
0.150
|
|
Length:Delivery
|
0.807
|
0.444
|
|
Group:Age:Sex
|
1.169
|
0.283
|
|
Group:Age:Length
|
0.425
|
0.703
|
|
Group:Sex:Length
|
0.962
|
0.375
|
|
Age:Sex:Length
|
1.410
|
0.227
|
|
Group:Age:Delivery
|
2.353
|
0.091
|
|
Group:Sex:Delivery
|
3.064
|
0.078
|
|
Age:Sex:Delivery
|
0.652
|
0.527
|
|
Group:Length:Delivery
|
1.289
|
0.263
|
|
Age:Birth:Delivery
|
0.954
|
0.413
|
|
Sex:Length:Delivery
|
2.612
|
0.085
|
|
Group:Sex:Length:Delivery
|
2.075
|
0.114
|
Table 3 Results of multiple post-hoc two-sample
permutation tests comparing echogenicity ratios of the control and
hypoxic-ischemic injury groups. CP – choroid plexus ratio; B
– calvarial bone ratio.
|
Echogenicity ratio
|
p-value
|
|
Caudate nucleus head_CP
|
0.017
|
|
Caudate_B
|
0.008
|
|
Lentiform nucleus_CP
|
0.017
|
|
Lentiform nucleus_B
|
0.011
|
|
Thalamus_CP
|
0.013
|
|
Thalamus_B
|
0.011
|
|
Occipital white matter_CP
|
<0.001
|
|
Occipital white matter_B
|
<0.001
|
|
Parietal white matter_CP
|
<0.001
|
|
Parietal white matter_B
|
<0.001
|
|
Frontal white matter_CP
|
<0.001
|
|
Frontal white matter_B
|
0.012
|
|
Perirolandic white matter_CP
|
<0.001
|
|
Perirolandic white matter_B
|
<0.001
|
The generated receiver operating characteristic curves and area under the curve
calculations revealed a similar trend (
[Fig.
8]; [Table 4]
). The lowest areas
under the curve, which were still in the acceptable range, belonged to the
caudate and putamen to CP ratios at 0.733 and 0.745, respectively. The
perirolandic WM had a high area under the curve, at 0.980 for both the CP and
bone ratios.
Fig. 8 Empirical receiver operating characteristic curves for the
echogenicity ratios of the brain region of interest, with either the
choroid plexus or calvarial bone as the internal control.
Table 4 Areas under the curve for each brain region
echogenicity ratio (CP – choroid plexus; B –
calvarial bone) calculated from the empirical receiver operating
characteristic curves.
|
Echogenicity Ratio
|
Area Under the Curve
|
|
Caudate nucleus head_CP
|
0.733
|
|
Caudate_B
|
0.810
|
|
Lentiform nucleus_CP
|
0.745
|
|
Lentiform nucleus_B
|
0.818
|
|
Thalamus_CP
|
0.753
|
|
Thalamus_B
|
0.820
|
|
Occipital white matter_CP
|
0.880
|
|
Occipital white matter_B
|
0.923
|
|
Parietal white matter_CP
|
0.863
|
|
Parietal white matter_B
|
0.938
|
|
Frontal white matter_CP
|
0.863
|
|
Frontal white matter_B
|
0.873
|
|
Perirolandic white matter_CP
|
0.980
|
|
Perirolandic white matter_B
|
0.980
|
The intra-observer reliability for all ratios was high, with the caudate to bone
ratio being the lowest at 0.832 and the anterior WM to CP ratio being the
highest at 0.992 (
[Table 5]
).
Table 5 Pearson correlation coefficients demonstrating
intra-observer reliability between the two measurements of each
echogenicity ratio.
|
Echogenicity Ratio
|
Pearson Correlation Coefficient
|
|
Caudate nucleus head_CP
|
0.963
|
|
Caudate_B
|
0.832
|
|
Lentiform nucleus_CP
|
0.954
|
|
Lentiform nucleus_B
|
0.834
|
|
Thalamus_CP
|
0.973
|
|
Thalamus_B
|
0.878
|
|
Occipital white matter_CP
|
0.981
|
|
Occipital white matter_B
|
0.937
|
|
Parietal white matter_CP
|
0.988
|
|
Parietal white matter_B
|
0.915
|
|
Frontal white matter_CP
|
0.992
|
|
Frontal white matter_B
|
0.924
|
|
Perirolandic white matter_CP
|
0.988
|
|
Perirolandic white matter_B
|
0.914
|
Discussion
HIE is a major cause of morbidity and mortality in neonates and can result in
long-term, persistent motor, sensory, and cognitive impairment [12]. Early detection of neonates with HIE is of
paramount importance for prompt treatment in order to reduce neurodevelopmental
damage and improve outcomes [2]. In the early phases
of HIE, dysfunction of membrane ion transport from hypoxia leads to cellular edema,
which later activates cellular pathways that ultimately lead to cell death and
tissue necrosis [13].The most sensitive imaging
modality to detect cytotoxic edema is diffusion-weighted imaging [14]. Diffusion-weighted imaging is highly recommended
in the evaluation of cases of HIE, as it can be employed to assess the extent of the
injury and for outcome prediction [15].
Despite advances in other imaging techniques, neurosonography, which is
radiation-free, cost-effective, and portable, remains a fundamental screening method
in the evaluation of focal or diffuse changes in the brain, particularly in cases of
HIE. These changes alter the echogenicity of brain structures, which translates into
either increasing or decreasing echogenicity relative to the unaffected areas of the
brain (qualitative neurosonography) [2].
Though neurosonographic evaluation of HIE is the most convenient imaging method for
this type of injury, it has historically been qualitative and subjective, which may
give rise to ambiguity and inconsistency [8]
[9]. Accurate and reliable interpretation of HIE by
neurosonography is heavily dependent on the experience of trained radiologists and
visual recognition of patterns [1]. Moreover,
depending upon the acquisition settings, transducer, size of the fontanelle, and the
selected window/level settings of the neurosonography images, the degree of
brain injury may be under- or overestimated if the ultrasound images are reviewed
only qualitatively. Even experienced pediatric radiologists may misinterpret the
degree of changes of WM echogenicity [6]. In addition
to these technical variations, the severity of HIE can lower radiologists’
sensitivity for subtle findings of HIE, and mild HIE may appear normal or
near-normal on neurosonography, especially if the neurosonography examination is
performed during the first two or three days of life [2]. The presence of coexistent complications, such as leukomalacia or
intracranial hemorrhage, could also present challenges for radiologists.
These limitations of the typical application of neurosonography may be overcome by
quantitative neurosonography. Quantitative data may also serve as a biomarker for
outcome prediction in neonates suffering from HIE. These biomarkers may play a major
role, especially in settings in which MRI is less available or not feasible due to
critical conditions. Previous studies [8]
[16] have proposed different quantitative techniques to
quantify neonatal WM lesions. Simaeys et al. [8] used
a software-based method to compensate for the variable ultrasound acquisition
factors by comparing the echogenicity of intracerebral lesions with that of the CP,
whose echogenicity is considered to be relatively constant. Padilla et al. [10] showed that calculation of the relative
echogenicity (ratio of the mean pixel brightness measured in brain regions and the
bone at the same depth of sampling) might offer a semiquantitative method to
evaluate various anatomical regions of the neurosonography examination. However,
there is a lack of data showing the diagnostic accuracy of the quantitative
evaluation of brain echogenicity in neonatal HIE.
Technical factors may also limit the validity of quantitative measures used to
identify HIE. Studies that have attempted to quantify ultrasound signals from the
neonatal brain have often used raw gray levels instead of ratios. Barr et al. [7] found higher gray levels in the brain parenchyma of
infants with brain injury. These authors used an 8-bit image with optimized scanning
parameters for each patient to obtain the best images. However, no normalization was
made, and only the raw pixel intensity was evaluated. This limits the
generalizability of the protocol of these authors secondary to bias introduced by
technical differences. In our study, we employed internal controls to help determine
the effects of technical factors. Similarly, Simaeys et al. [8] compared the echogenicity of the periventricular WM with that of the
CP and found elevated ratios in neonates with HIE.
Here, we describe multiple brain region echogenicity (mean grayscale value) ratios
that could help identify early sonographic evidence of HIE. These echogenicity
ratios of multiple brain regions that are known to be involved in HIE compared to
internal controls (CP and calvarial bone) were able to discriminate between healthy
neonates and neonates with HIE. We focused our measurements on those areas of the
brain most vulnerable to damage from HIE, specifically the deep gray nuclei [2]. The frontal, parietal, parieto-occipital, and
perirolandic WM, while more challenging to identify with consistency, also represent
vulnerable areas in neonates with hypoxia-ischemia [17]
[18]. Our results show that while the
deep gray matter demonstrated an acceptable to a good degree of discriminatory
capacity, WM, mainly in the perirolandic region, had a greater ability to identify
individuals with HIE.
The echogenicity of the neonatal CP may be potentially affected in cases of
infections (ventriculitis, ependymitis, and choroid plexitis), hemorrhage, or
congenital CP tumors. CP hemorrhage is more commonly found in premature infants but
may occur in term neonates with HIE [11] and is rarely
described in healthy neonates. Heibel et al. found eleven cases of CP hemorrhage
(two isolated and nine associated with intraventricular extravasation) among 1000
consecutive neurosonography examinations of clinically normal, full-term neonates
[19]. CP infarcts (commonly associated with
thalamic infarct) are very uncommon in the pediatric population and are more
commonly reported in adults [20]
[21]. In animal studies, the CP has been shown to be
highly vulnerable to HIE [22]. However, there is no
clear evidence that this occurs in humans. There were no signs of CP hemorrhage in
neurosonography examinations in the HIE and control groups and no mention of CP
hemorrhage in the magnetic resonance reports of the HIE group. Both CP and bone were
chosen as internal controls. Calvarial bones may be a potentially better internal
control since they are seen in all images and are unaffected by HIE [10].
Sagittal scans were not used for analysis since these were only available in a few
neonates. Sagittal scans would be beneficial to better delineate deep and
periventricular WM, the majority of the length of the caudate nucleus, and larger
sections of the thalamus. In sagittal scans, the immediate comparison between the
hemispheres would be lost in paramedian and periventricular sections. However, since
they are standard sections, reliability should be high. Alternative acoustic windows
such as the posterior fontanelle or mastoid would help measure the echogenicity of
the posterior fossa structures.
Beyond sensitivity and specificity, it is essential to have a test that is reliable
and reproducible. Limited intra-observer reliability testing with our ratios
demonstrated an excellent correlation between two independent segmentation events
spaced approximately one month apart. The individual performing the measurements in
our study was a fellowship-trained neuroradiologist, and as such, likely has greater
knowledge of anatomy on neonatal neurosonography than a radiologist with less
specialized training. However, it is possible to teach radiologists how to identify
these regions, and our research could also form the basis for automated,
computer-aided segmentation in the future.
Our study has limitations. The study was retrospective, and only a limited number of
healthy control neonates were available. Also related to the retrospective nature of
our study, the ultrasound studies were performed on multiple different machines with
different technologists. Future studies addressing interobserver variability should
be performed to validate widespread clinical use of the method. Our protocol, which
utilized internal controls, was designed to help mitigate this limitation. Also, it
was not possible to completely blind the individual (FGG) analyzing the studies
because the reader is a trained neuroradiologist who can recognize the sonographic
signs of HIE. Lastly, since neurosonography in controls occurred three days (mean)
later than in HIE patients, perhaps a transient increase in echogenicity ratios
could be found after stressful but not asphyxiating spontaneous vaginal
delivery.
In conclusion, quantification of brain parenchyma echogenicity, when coupled with
internal controls, demonstrated differences in echogenicity in term HIE patients
compared to controls during the first days following the injury. Quantitative
neurosonography represents a potential method for early identification of HIE and
increases diagnostic accuracy, which warrants further validation. Our ratios were
elevated in cases of HIE and produced promising receiver operating characteristics.
Further work with a larger cohort could reveal whether a quantitative approach can
discern between degrees of severity of HIE. In the future, neurosonography protocols
should be tailored to evaluate this region, which requires dedicated posterior
coronal scanning. Furthermore, our method could, in principle, especially
prospectively, offer an opportunity to employ machine learning/artificial
intelligence under standardized conditions to improve the assessment of HIE.
