Imaging Perinatal Brain Injury in Premature Infants
Jeffrey J. Neil, MD, PhD*, and Terrie E. Inder, MD, PhD†
The primary methods currently in use for imaging the infant brain are cranial ultrasound
(CUS), computed tomography (CT) and magnetic resonance imaging (MRI). This review
outlines the relative strengths and weaknesses of these modalities in relation to the
premature infant, with specific focus on the correlations between imaging findings and
neurodevelopmental outcome. Since MRI is undergoing rapid development at this time, the
newer MRI methods of brain volume measurement and diffusion tensor imaging are
reviewed in more detail. Current guidelines regarding the application of these neuroimaging methods to the premature infant are discussed.
Semin Perinatol 28:433-443 © 2004 Elsevier Inc. All rights reserved.
T
he three major neuroimaging modalities used in the evaluation of the premature infant brain are cranial ultrasound (CUS), computed tomography (CT) and magnetic resonance imaging (MRI)— each of which has relative strengths
and weaknesses. In this review, we will compare and contrast
these methods with specific regard to their clinical utility in
the premature infant. A major focus of this review will be on
MRI, since there have been many recent advances in the
application of this methodology to the evaluation of the premature infant.
Due to the high incidence of neurodevelopmental disability in the premature infant, this patient population has a clear
need for accurate and reliable neuroimaging. Imaging of the
brain of the premature infant has at least two major aims:
firstly, to accurately define the nature of cerebral injury in this
population, and, secondly, to determine the imaging features
that most accurately predict adverse neurodevelopmental
outcome. In addition to defining cerebral injury, recent advances in neuroimaging techniques are now providing insight into alterations in structural brain development associated with premature birth.
Five to 15% of prematurely born children develop cerebral
palsy, with an additional 30 to 50% displaying impaired academic achievement and/or behavioral disorders requiring
additional educational resources.1-6 To understand and apply
potentially beneficial neuroprotective strategies in the neonatal intensive care unit, it is necessary to first establish the
*Departments of Neurology, Pediatrics, and Radiology, Washington University School of Medicine, St. Louis, MO.
†Neonatal Neurology, Royal Women’s and Royal Children’s Hospitals, Murdoch Children’s Research Institute, Melbourne, Australia.
Address reprint requests to: Jeff Neil, MD, PhD, Pediatric Neurology, St.
Louis Children’s Hospital, One Children’s Place, St. Louis, MO 63110.
E-mail: neil@wustl.edu
0146-0005/04/$-see front matter © 2004 Elsevier Inc. All rights reserved.
doi:10.1053/j.semperi.2004.10.004
nature, timing, and location of the cerebral lesions predisposing premature infants to long-term neurodevelopmental difficulties. In addition, accurate prognostic information before
discharge from the hospital would be of high clinical utility
by assisting in defining the need for continuing intervention
strategies, such as physical therapy. With current conventional neuroimaging techniques, severely abnormal findings
are moderately accurate for predicting an adverse outcome,
particularly in relation to motor outcomes such as cerebral
palsy (vide infra). However, there is little relationship between the findings of these conventional techniques and cognitive outcome in the premature infant, even when images
are obtained later in life.7 Newer MRI methods, including
quantitative volumetric measurements and diffusion tensor
imaging, may offer higher sensitivity to more subtle anatomic
alterations. Thus, they have the potential to provide a better
understanding of the anatomic substrate for the increased
susceptibility for cognitive difficulties in this high-risk population.
With current neuroimaging technologies there are several
well-defined forms of cerebral injury that are routinely detected in premature infants and have proven prognostic significance. The most common form of cerebral injury currently detected with conventional neuroimaging, particularly
CUS, is intraventricular hemorrhage (IVH), with an incidence of approximately 20% for all grades of IVH. The complications associated with IVH include posthemorrhagic hydrocephalus, periventricular hemorrhagic infarction,8 and an
increased incidence of periventricular leukomalacia
(PVL).9-12 Another common neuropathology is that of PVL
consisting of necrosis of white matter in a characteristic distribution, ie, dorsal and lateral to the external angles of the
lateral ventricles, involving particularly the centrum semiovale (frontal horn and body), optic radiation (trigone and
433
J.J. Neil and T.E. Inder
434
Figure 1 Cranial ultrasound studies demonstrating PVL. The image on the left is a saggital view. A, anterior; P, posterior.
The lateral ventricle, which appears dark, is designated by the black arrow. A bright area of increased echodensity,
corresponding to PVL, is designated by the white arrowhead. The image on the right is a coronal view of cystic PVL. L,
lateral ventricles. The black arrowhead indicates a cyst present at the dorsolateral angle of the lateral ventricle.
occipital horn), and acoustic radiation (temporal horn).9 The
incidence of PVL varies in relation to its method of detection,
which will be discussed further. In autopsy studies of premature infants it was found to differ between medical centers,
ranging from 25 to 75%. Other major forms of neuropathology in the premature infant, less common than PVL, include
selective neuronal injury and focal cerebral ischemic lesions.
Cranial Ultrasound
Background
Ultrasound relies on the reflection of ultrasound waves from
tissue to provide images. Its greatest advantage lies in its
portability, as it is possible to image infants without moving
them from the intensive care unit. CUS images show the
outline of the cerebral ventricles clearly, as the ventricles
appear echolucent or dark as compared with adjacent white
matter. CUS also has high sensitivity for detection of hemorrhage, which appears echogenic or bright. Thus, CUS is very
useful for detection of intracranial hemorrhage and associated posthemorrhagic hydrocephalus. CUS is also used for
detection of PVL, and a characteristic progression of imaging
features has been described.9 During the first week following
injury, there are echogenic foci in the periventricular white
matter due to local necrosis with congestion and/or hemorrhage (Fig. 1). This is followed, during weeks one through
three, by the appearance of echolucent cysts, which correlates with cyst formation due to tissue dissolution. By two to
three months, ventriculomegaly appears, often with disappearance of the cysts. This is associated with deficient myelin
formation and/or glioisis with collapse of the cysts.
One of the disadvantages of CUS is that it provides relatively poor contrast for lesions of the brain parenchyma.
Acute stroke, for example, can be difficult to detect with CUS
as compared with CT and MRI.13 In addition, since CUS
images are typically obtained through the anterior fontanel,
there is a limited field of view which does not “see” the cerebral convexities. Further, image detail in the posterior fossa,
which is relatively far from the transducer, is often poor. This
can be overcome by imaging through the posterior fontanel,
which should be encouraged as a standard of practice. Acquisition and interpretation of CUS images is also more operator-dependent than CT or MRI, and quality can vary
markedly from medical center to medical center. Finally,
assessment of increased echodensity is subjective, though
this could potentially be overcome by comparing the
echodensity of the region of possible abnormality with that of
a standard structure, such as the choroid plexus.
Correlation of Image Findings with Outcome
CUS studies have been available for evaluation of premature
infants for nearly 30 years. There is a relatively large literature
describing associations between CUS findings and outcomes.
Comparison of these studies is complicated by a variety of
factors, some of the more prominent of which are differences
in: (a) the timing and frequency of the HUS studies obtained,
(b) the technical quality of the ultrasound equipment used,
(c) the technical skills of the ultrasonographers, and (d) the
duration of time for which subjects were in clinical follow up
(more subtle neurodevelopmental abnormalities are not
readily detected at younger ages). Nevertheless, the majority
of studies suggest that 30 to 60% of premature infants who
develop cerebral palsy (CP) had lesions on CUS during the
perinatal period.14-17 A recently published study by deVries
and coworkers suggests that CUS may be more sensitive than
previously believed.18 In this study, preterm children underwent CUS studies weekly up to and including a scan at 40
weeks postmenstrual age (the latter scan was typically done
Imaging perinatal brain injury
as an outpatient). Children were followed for two years and
evaluated for CP and cognitive difficulties. Ninety-two percent of children with CP had abnormalities found on CUS
studies. These abnormalities were major (IVH with ventricular dilation, hemorrhagic periventricular infarction, cystic
PVL, subcortical leukomalacia, and basal ganglia lesions) in
83% and minor in 17%. Notably, CUS abnormalities were
first detected after 28 days of life in 29% of infants with major
abnormalities. Their data give a specificity of 95%, sensitivity
of 76%, and positive predictive value of 48% for CUS detection of CP in infants ⱕ32 weeks gestational age (GA). Similar
values were obtained for infants between 33 and 36 weeks
GA. The high sensitivity for HUS in this study as compared
with the rest of the literature may be related to the fact that
infants were scanned more regularly and for a longer period
of time than in typical CUS studies. It is also possible that a
high level of technical competence contributed to the high
sensitivity, as Wheater and coworkers, performing weekly
CUS studies and assessing children for CP at age 18 months,
found CUS abnormalities in only 56% of children with CP.16
While the studies described above address the detection of
motor disability (CP), studies have also been done to evaluate
the sensitivity of CUS to cognitive disabilities. In general,
CUS seems relatively insensitive for predicting children who
will go on to have cognitive difficulties, particularly subtle
ones.18,19 It is also worth noting that recent studies indicate
that MRI is more sensitive than CUS for the detection of
diffuse PVL, though cystic PVL is readily detectable by both
modalities.20,21
Guidelines for the Use of CUS
The frequency and duration of time for which CUS studies
should be obtained in premature infants remains an area of
controversy. It has been suggested that clinically-stable premature infants do not need repeat screening HUS if they have
two normal studies ⱖ7 days apart, though infants ⬍25
weeks GA require an additional HUS study at the time of
discharge.22 The quality standards committee of the American Academy of Neurology and the practice committee of the
Child Neurology Society recommend that infants be scanned
at 7 to 14 days of age and again at 36 to 40 weeks postmenstrual age.23 In light of the findings of deVries and coworkers18 described above, which were not available at the time
the practice parameter was developed, it seems likely that
CUS is more sensitive for detecting injury if studies are obtained more frequently. Further, the window for the appearance of injury is longer than previously believed. As a result,
it would be reasonable to obtain weekly HUS studies from
birth to 32 weeks postmenstrual age, and finally at 36 to 40
weeks postmenstrual age. This guideline applies, of course,
to clinically stable infants. Infants with clinical deterioration
or alteration in neurological state may require more frequent
imaging.
As noted above, CUS is also very useful for evaluating ventricular size. Posthemorrhagic ventricular dilation (PHVD) is a
relatively common complication of IVH and should be carefully monitored for following the detection of Grade II or
435
higher IVH. The incidence of adverse neurodevelopmental
outcome following PHVD is high, and there is an increasing
trend to earlier and more aggressive therapy for this condition to improve outcome.24,25 At present, CUS images are
typically evaluated qualitatively for ventricular size. While
this strategy has proven useful, it would appear more informative to evaluate ventricular size in a quantitative fashion,
based on established norms.26 Further, the occipital horn is
the first and the frontal horn is the last to enlarge following
IVH.27 Thus, measurement of the thalamo-occipital dimension of the ventricle,26 via the posterior fontanel, is the most
sensitive indicator of ventriculomegaly. Frequent CUS monitoring of the ventricular dimensions, particularly in the
acute period, may assist in assessing the infant’s course and
determining the need for intervention with ventricular drainage. It may also guide the frequency and volume of CSF
drainage necessary to decrease ventricular size.
Computed Tomography
Background
Computed tomography has been available for approximately
30 years. The method is based on passing an x-ray beam
(ionizing radiation) through the sample at a series of different
angles. Based on the attenuation of these beams, an image can
be constructed using a method known as filtered back projection. Contrast in CT images is dependent on differing degrees of attenuation of x-rays by various structures. Bone, for
example, causes a high degree of attenuation, and appears
bright on CT images. CSF, on the other hand, causes a low
degree of attenuation, appearing dark. CT provides excellent
views of bone and is also very sensitive for the detection of
hemorrhage, which appears bright. It allows differentiation
of white and gray matter, though the contrast between these
two types of tissue is relatively low in comparison with MRI.
While it does not suffer the field of view difficulties associated
with CUS, structures in the posterior fossa are relatively
poorly demonstrated on CT because of nonuniform attenuation of x-ray energy by the bone at the base of the skull (beam
hardening artifact). CT scans usually require that the infant
be removed from the ICU, which is a disadvantage as compared with CUS. On the other hand, the scan time is shorter
than that of a typical MRI study. Further, the infant is more
readily accessible while in the scanner, in the event of an
emergency, than for an MR scan, though there is a trend in
MR magnet design toward more open magnet configurations
which provide better patient access.
A further issue for CT is the exposure of the infant brain to
ionizing radiation. There are two main areas of concern related to this exposure—firstly the risk of future malignancy
and, secondly, cognitive impairment. The adverse impact of
cranial irradiation on brain and cognitive development following radiation treatment of brain tumors has been clearly
shown and is dose-related. However, the threshold value for
the effect is not known. Recently, Hall and coworkers28 suggested that even low doses of ionizing radiation, similar to
those delivered by CT scans, may adversely affect brain and
J.J. Neil and T.E. Inder
436
cognitive development. They found significant reductions in
scores on tests for learning ability and cognitive reasoning for
individuals who were exposed to ⬎100 mGy ionizing radiation for the treatment of cutaneous hemangioma before age
18 months. In a study of 20,000 Israeli children treated with
cranial irradiation for tinea capitis, Ron and coworkers found
poorer cognitive performance in irradiated children as compared with controls.29 Currently, it is unclear what long-term
effects, if any, low doses of cranial irradiation, such as those
delivered during a cranial CT scan, may have when administered during infancy, a phase of rapid brain development.
In relation to the risk of subsequent malignancy, a 0.1% risk
of life long fatal malignancy from a head CT at age 12 months
has been estimated.30 However, the risk of tumor may be
significantly greater with exposure at a younger age. The risk
of development of brain tumor was found to be 10-fold
higher in infants exposed to cranial irradiation under 5
months of age in comparison to over seven months of age.31
While extrapolation of findings from cranial irradiation of
children to head CT studies of premature infants must be
regarded cautiously, the studies do raise concerns about exposure to ionizing radiation during infancy. Further research
would be necessary to accurately define the nature of any risk
associated with CT scanning of the infant brain. Until such
data are available, it is reasonable to restrict the use of this
neuro-imaging technique to selected settings in which the
information obtained from the imaging study is clearly of
benefit to the patient.
CT scanning of the premature infant has been employed
for detection of both hemorrhage and PVL. Although it is
likely no more sensitive than CUS for detecting cystic PVL
during the perinatal period, findings of reduced volume of
periventricular white matter and ventriculomegaly with irregular outline of the lateral ventricles have been described in
older children with clinical findings consistent with PVL (ie,
spastic diplegia).32 CT can also be used to detect stroke,
though image contrast for nonhomorrhagic stroke is weak,33
particularly in comparison with MR.
Correlation of Image Findings with Outcome
There are relatively few studies correlating CT findings with
outcome, and the majority of them are from the mid 1980s.
In general, the correlations have not been especially strong.
For example, there was no relationship found between neonatal CT and clinical outcome at age 18 months in a study of
145 children.34 However, in a series of 45 infants imaged at
40 weeks postmenstrual age, the finding of ventriculomegaly
correlated with the presence of learning disabilities at school
age.35
Guidelines for the Use of CT
CT scanning has not been demonstrated to contribute significantly to the evaluation of the brain of the premature infant.
Correlations between CT findings and outcome are weak,
and similar information can be gleaned from CUS studies. In
addition, the concerns regarding exposure to ionizing radiation must be considered. CT has a role in the rapid evaluation
of the newborn infant with traumatic head injury following
delivery in consideration of urgent neurosurgical intervention, but this situation is rarely encountered in preterm infants.
Magnetic Resonance Imaging
Background
The signal from which magnetic resonance images are derived arises from the 1H atoms of water (1H2O). The concentration of 1H in water is on the order of 100 mol/L. This high
concentration provides a strong signal from which to obtain
images. Other nuclei, such as 23Na and 31P, also provide MR
signal, but their concentrations in the brain are on the order
of tens of mM—roughly 10,000 times less than that of 1H in
water. While signals from these nuclei are detectable by MR
spectroscopy, for which signal is obtained from a relatively
large volume of brain, they are not strong enough to be used
to obtain images of high spatial resolution. Though MR spectroscopy has been applied to evaluation of the neonate brain
and has proven to have prognostic value,36,37 this review is
focused on MR imaging.
Conventional MR Imaging
Image contrast obtained in MR images is primarily the result
of the different MR properties for the 1H atoms of water
residing in different parts of the brain. Depending on the
image acquisition conditions (or acquisition parameters), images with a variety of contrasts can be obtained. Two common forms of contrast are based on the T1 and T2 MR imaging characteristics of water. (T1 and T2 represent different
MR relaxation time constants, which will simply be referred
to as “T1” and “T2” for this discussion.) The T1 and T2
characteristics of water differ between the tissue types of CSF,
white matter, and gray matter providing contrast for the images. For example, water in CSF has a relatively long T1, and
thus appears dark on T1-weighted images while in T2 images, water in CSF looks bright relative to brain tissue. It can
sometimes be difficult to distinguish injured brain from adjacent CSF-containing spaces on T2-weighted images. This is
because both injured tissue and CSF appear bright. One
means of compensating for this is the use of fluid-attenuated
inversion recovery (FLAIR) imaging. In this case, an additional RF pulse is applied during image acquisition, serving
to eliminate signal from CSF, making it appear dark on an
otherwise T2-weighted image. In current clinical practice,
T1-weighted, T2-weighted, and FLAIR images are usually
obtained. In addition, Gd-based contrast agents are often
injected into the intravascular space. These agents, when in
contact with water, shorten the T1 so that the water appears
bright on T1-weighted images. T1-weighted images following injection of intravenous contrast appear bright in areas
which are highly vascular or for which contrast agent has
extravasated into the extravascular space due to disruption of
the blood brain barrier.
Imaging perinatal brain injury
Diffusion MR Imaging
MR images can also be obtained in which contrast is dependent on water displacement (diffusion-weighted imaging or
DWI). With this method, water displacements on the order
of 10 ␮m can be detected. The physical constant characterizing this water motion is called the “apparent” diffusion
coefficient (ADC) in recognition of the fact that water displacements in tissue are influenced by factors other than
simple Brownian motion, such as restrictions due to cell
membranes. While one might not expect tissue water displacements to be especially useful for image contrast, diffusion-based MR imaging provides remarkably rich information about cerebral structure and its disruption. For example,
water ADC values decrease within minutes after a variety of
forms of tissue injury, including stroke.38 As a result, diffusion-based MR images provide one of the earliest indicators
of tissue injury, showing changes many hours to days before
abnormalities are detectable with other forms of imaging.
There are a variety of ways of obtaining and displaying
diffusion images, and this has lead to confusion in the interpretation of diffusion-based MR studies. It is possible to obtain a conventional, diffusion-weighted MR image as a single
scan. This permits fast image acquisition and processing.
However, image contrast with this single-scan method of
diffusion imaging is influenced by both water ADC and T2.
As a result, a region may appear bright either because of a low
water ADC or a long T2. During the subacute phase of stroke,
water ADC values are decreased and T2 constants increased.
Both changes make the region of abnormality bright, providing reinforcing contrast. Unfortunately, these two forms of
contrast may interfere with, or even cancel, one another under other circumstances. The finding of an area of increased
signal intensity on a diffusion-weighted image in which the
ADC is normal but the T2 is long is referred to as “T2 shine
through.”39 One means of avoiding this admixture of contrasts is to obtain images both with and without diffusion
weighting. These images can be combined to eliminate the
effect of T2 weighting, providing quantitative values for water ADC (in units of ␮m2/ms). The resulting data are shown
as parametric maps in which image intensity is directly related to quantitative ADC values. Image acquisition time for
parametric maps is slightly longer, as it requires at least two,
and typically four, acquisitions, but parametric maps provide
more easily interpreted, quantitative diffusion information.
Diffusion sequences that correct for T2 relaxation are available on most commercial MR scanners, and their use should
be encouraged as a standard of clinical practice. It is worth
noting that parametric maps and diffusion-weighted images
have opposite image intensity conventions—areas of low
ADC values appear dark on parametric maps, but bright on
diffusion-weighted images.
Another form of contrast available with diffusion imaging
is based on diffusion anisotropy. Anisotropy refers to the
condition in which water ADC values differ depending on the
direction along which they are measured. In myelinated
white matter, for example, water molecular displacements
are smaller perpendicular to fibers than parallel to them be-
437
cause motion perpendicular to fibers requires passing
through or around layers of myelin membrane, whereas motion parallel to them does not. Thus, water apparent diffusion
in mature white matter is highly anisotropic. In this manner,
diffusion imaging characterizes tissue microstructure. Two
parameters related to diffusion anisotropy are particularly
useful for microstructural characterization: the degree of anisotropy of water motion (eg, relative anisotropy or RA) and
the direction along which water apparent diffusion is greatest
(the direction of the major eigenvector of the diffusion tensor). RA values approach zero for CSF, for which diffusion is
isotropic and ADC values are equal in all directions. RA values are highly positive for myelinated white matter, for which
ADC values are on the order of three times smaller for diffusion perpendicular to fibers as compared with parallel to
them. For white matter, the direction along which water ADC
values are greatest represents the primary orientation of myelinated fibers within an image vfibers withoxel (remember
that water displacements are greatest parallel to fibers). This
information can be used for a form of “tract tracing” in which
neural connections can be inferred.40 For gray matter, the
direction along which water ADC values are greatest reflects
the radial organization of developing cortex41 (Fig. 2).
Volumetric Measurements through MR
MR images can be used to measure cerebral volumes. Images
of different contrasts, such as T1, T2, and proton-density
weighted, are analyzed to classify tissue as CSF, gray matter
or white matter in a process referred to as “segmentation”
(Fig. 3). The number of image elements, or voxels, corresponding to each class can be summed to compute volumes
in units of cm3. There is not yet a consensus as to the optimal
means of assessing and comparing these volumes. While it
has proven useful to measure total intracranial volumes for
tissue types, it is also desirable to evaluate volumes on a
regional basis because injuries affecting cerebral volumes
may preferentially affect some brain areas. One approach is to
apply arbitrary boundaries for regions, similar to lines for
latitude and longitude on a map.42 Though this approach is
fairly simple and relatively straightforward to implement, it
tends to group brain regions of disparate function and physiology, such as brainstem and temporal lobe, in the same
volume. It also may miss volume abnormalities that involve
small brain regions that cross the border of adjacent areas
chosen for analysis. A more sophisticated approach consists
of applying statistical methods to compare the volume of the
brain of interest to that of a “standard” brain on a voxel-byvoxel basis.43 This method provides parametric maps of the
regions for which volumes are statistically different, similar to
the maps generated for functional MRI studies. There is also
no agreement on how these volumes should be normalized.
For example, should gray matter be expressed as an absolute
volume (ml), as a percentage of total brain volume, or as a
percentage of intracranial volume? It might also be useful to
express brain volumes in relation to postconceptional age.
Though the uncritical application of the volumetric approach
has been unfavorably likened to phrenology,44 identification
of areas of volume abnormality is an important step in iden-
J.J. Neil and T.E. Inder
438
Figure 2 An example of diffusion anisotropy in the cerebral cortex of an infant at 26 weeks GA. The magnetic resonance
image on the left is an axial view of the brain. The background of the image is an ADC map. Regions of white matter
appear brighter than gray matter because water in white matter has a higher apparent diffusion coefficient at this stage
of development. The yellow lines superimposed on the image show the direction in which water diffusion is highest for
each brain region. In the cortex, diffusion is highest in a radial direction. This is particularly conspicuous in the
magnified view of the frontal cortex (inset). The image on the right is a silver stain by Cajal from the precentral gyrus
of a 1-month-old child. Note the radial orientation of the apical dendrites of the pyramidal cells. This organization,
along with that of the radial glia, tends to restrict longitudinal water motion relative to radial motion. This is the likely
explanation for the preferred radial direction of water diffusion in developing cortex (reproduced with permission,
Cajal Institute, CSIC, Madrid, Spain, copyright inheritors of Santiago Ramón y Cajal; and Oxford University Press41).
tifying the anatomic substrate of brain injury in premature
infants.
Overall, MR has the disadvantage that it typically requires
that the infant be transported from the NICU to the scanner
for study. The risk to the infant associated with transport out
of the ICU has been ameliorated by the availability of MRcompatible, purpose-built transport isolettes. These devices
include support equipment for the infant as well as the RF
coil necessary for MR studies. The isolette holding the infant
can be placed in the MR scanner for imaging. Once placed in
such an isolette in the ICU, an infant can be taken to the
scanner and left undisturbed until he/she returns to the ICU
after the scan. These isolettes are now commercially available. MRI is also relatively expensive, requiring equipment
costing millions of dollars and involving careful and sometimes costly preparation of the site at which the scanner is
housed. MRI has several advantages. Unlike CT, it does not
involve ionizing radiation, making it relatively safer. It also
offers a rich variety of tissue contrasts, providing tissue characterization superior to both CT and CUS. Preliminary stud-
ies suggest that MR is more sensitive to subtle tissue injury
than either CT or CUS.
Correlation of Image Findings with Outcome
MR imaging of premature infant brain has been available for
approximately 20 years and remains an area of active research. The earliest studies used conventional MR imaging
and focused on studies obtained after discharge from the
hospital on patients for whom PVL had been diagnosed by
CUS during the perinatal period. Areas of abnormal signal
intensity in periventricular white matter, ventriculomegaly,
varying degrees of cerebral atrophy, thinning of the corpus
callosum, and delayed myelination have been described in
children as old as five years.45-47 These findings are common
in prematurely born children48,49 and suggest that MR can be
used to detect the sequelae of PVL in older children. However, MR abnormalities consistent with PVL detected in prematurely born children at ages 850 and 15 to 17 years7 show
only moderate correlation with cognitive and motor deficits.
In contrast, abnormalities of visual pathways detected on MR
Imaging perinatal brain injury
439
Figure 3 An example of image segmentation for quantitative volumetric analysis. The image on the left is a coronal
T1-weighted image. The image in the center is the corresponding T2-weighted image. The image on the right is the
segmentation map derived from these MR images. Blue represents CSF, gray is gray matter, red is unmyelinated white
matter, white is basal ganglia, and yellow is myelinated white matter. This map can be used to determine relative
volumes (in cm3) of these different tissue types.
images obtained in late infancy show strong correlation with
visual impairment.51
Conventional MR studies obtained at term equivalent,
near the time of discharge of the premature infant from the
hospital, have prognostic significance. The presence of parenchymal lesions gave a sensitivity of 100% and specificity
of 79% for motor abnormality in a study of 51 infants, 11 of
whom had neurologic deficit at follow up.52 The parenchymal lesions consisted of hemorrhage, changes consistent with
PVL, infarction, and reduction in white matter volume. In a
study of 15 premature infants, 13 of whom had neurologic
deficits at follow up, findings of changes consistent with PVL
and widespread infarction were associated with adverse neurologic outcome at age 1 year.53
There has been focus on abnormalities of the posterior limb of
the internal capsule (PLIC) and outcome. Abnormal signal intensity in the PLIC in term infants following hypoxic–ischemic
injury correlates with adverse neurodevelopmental outcome.54
In preterm infants with intraventricular hemorrhage and unilateral hemorrhagic parenchymal involvement, signal abnormalities in the PLIC on MR studies obtained at term equivalent
strongly predicted future hemiplegia.55 Abnormalities of diffusion parameters of the PLIC also correlate with outcome. In term
infants with hypoxic ischemic encephalopathy, low values for
ADC56 predict poor outcome. For preterm infants imaged at
term equivalent, low values for diffusion anisotropy57 correlate
with adverse outcome. It is likely that changes in the MR properties of the PLIC reflect the effects of Wallerian degeneration as
a consequence of gray and/or white matter injury.58,59
Broader applications of diffusion-weighted MR imaging
have also proven useful for evaluating premature infant
brain. As noted above, ADC values decrease rapidly following
injury, and provide an early means of detecting injury.60,61
When applying this method, it is important to bear in mind
that the changes in ADC are dynamic.62 As shown in Fig. 4,
ADC values may be normal during the first day after injury,
are maximally decreased between 2 and 4 days after injury,
and return to normal (“pseudonormalization”) at approximately 7 days after injury.63 Following pseudonormalization,
ADC values are higher than normal and remain so for many
months. A number of studies have been done to evaluate the
sensitivity of changes in ADC for detection of injury in term
infants with hypoxic ischemic encephalopathy. In general,
those for which images were obtained near the time of
pseudonormalization64 suggest that this method is less sensitive than those for which images were obtained sooner after
injury.65,66 T2-weighted images, on the other hand, do not
show injury for up to 2 days, but show injury well by 7 days.
Thus, the imaging method most sensitive to injury depends
on the time after injury. During the first days after injury,
diffusion-weighted imaging is most sensitive, particularly at
2 to 4 days. At 1 week after injury, conventional T1- and
T2-weighted images are most sensitive.
Abnormalities in ADC are associated with MR evidence of
white matter injury. Abnormally high ADC values have been
found for white matter in subjects with diffuse signal abnormalities on T2-weighted images67 and focal white matter signal abnormalities on T1- and T2-weighted images.68 In a
study of twelve term infants with hypoxic ischemic encephalopathy, the presence of white matter abnormalities on diffusion imaging correlated with adverse outcome in infants
followed for up to 11 months.69
While evaluation of ADC values is proving useful for detecting injury, measures of diffusion anisotropy have the potential to provide additional information regarding tissue microstructure. As noted above, diffusion anisotropy is present
in both white and gray matter. For white matter, anisotropy
values are relatively low until myelination takes place.70,71
For gray matter, diffusion anisotropy values are highest during early development, due to the radial organization of developing cortex, and decrease steadily up to term.41 Thus,
anisotropy values change with development for both tissues.
To date, there have been relatively few studies of anisotropy
in injured premature brain. However, alterations in white
J.J. Neil and T.E. Inder
440
Figure 4 A plot of ADC values versus time after injury for term infants with perinatal brain injury. The ADC values are
normalized, with 1.0 representing a normal value, values less than 1.0 a decreased value, and values greater than 1.0
an increased value. Note that the changes in ADC are dynamic, varying with time after injury. Based on these data, one
would expect diffusion MR imaging to be most sensitive for detection of injury at two to four days after injury, a time
when conventional MR images often do not yet show injury. At approximately one week after injury, ADC values are
near normal (pseudonormalization). As a result, diffusion MR imaging is relatively insensitive to injury at this time,
though conventional MR images typically show injury well by this time (reproduced with permission, Lippincott,
Williams & Wilkins.62
matter anisotropy values and organization have been described in infants with PVL.72 Further, using diffusion anisotropy for white matter “tract tracing,” abnormalities in sensory
cortex pathways have been demonstrated in two children
with spastic cerebral palsy.73 These methods show tremendous potential for evaluation of premature infant brain, but
more studies are necessary to fully define their clinical role.
MR images can also be evaluated for tissue volumes. In
general, prematurity and white matter injury are associated
with reduced brain tissue and increased cererbrospinal fluid
volumes. Studies of former premature infants when they
reach adolescence show reductions in overall brain volume
and gray matter volume with an increase in lateral ventricular
volume.74 The presence of signal abnormalities in white matter of former premature infants evaluated at the age of 15
years is associated with reduced white matter volume.75 Fur-
Figure 5 MRI of three grades of abnormality in gray matter gyral maturation in premature infants at term. The image on
the left shows normal gyral maturation. The center image demonstrates a reduction in complex gyral folding with
preservation of secondary gyri in the transverse sulci and gyri consistent with 36 to 37 weeks GA. The image on the right
shows severe impairment of gyral development in all regions, consistent with 34 weeks GA (reproduced with permission, Mosby Publishing).78
Imaging perinatal brain injury
ther, increased lateral ventricular size in this population is
associated with cognitive and motor deficits.76 Volumes of
sensorimotor and midtemporal cortices are associated positively with full-scale, verbal, and performance IQ scores.42
Thus, alterations in cerebral volumes have a significant clinical correlate.
Fewer studies have been obtained in which volumes are
reported for premature infants imaged near the time of discharge at term equivalent, and there are not yet any published
studies that include clinical follow up. Nevertheless, a variety
of MR abnormalities have been described. In infants with
prior evidence of PVL by either CUS or MR, there is a marked
reduction in cerebral cortical gray matter as compared with
either preterm infants without PVL or term infants. This
change is associated with a significant decrease in total brain
myelinated white matter and an apparent compensatory increase in total cerebrospinal fluid volume.77 Further, studies
indicate that gyral development is markedly immature in
infants with evidence of white matter injury78 (Fig. 5). Thus,
while there has been a strong focus on white matter injury in
premature infants, these studies provide evidence for cortical
gray matter injury as well. It is not yet known whether these
gray matter changes are a consequence of white matter injury, or occur independently.
Guidelines for the Use of MRI
The use of MRI to evaluate the brain of premature infants
represents a moving target, as the more advanced MR diffusion and volumetric methods are still under active investigation. However, MRI clearly has a role to play in evaluating
premature infants. It is important to note that brain water
content is higher for infant brain than for older children, and
T1 and T2 relaxation time constants are consequently
greater. This, in association with other changes related to
brain maturation, causes the gray/white contrast on T1- and
T2-weighted images of infants to be exactly opposite that of
children greater than one year of age. Thus, it is critically
important that the image acquisition parameters be optimized for newborn infants. In general, it would be useful to
obtain T1-weighted, T2-weighted, T2*-weighted (sensitive
to hemorrhage) and diffusion images around term equivalent
to evaluate myelination through the posterior limb of the
internal capsule. In addition, if there is any evidence of acute
cerebral ischemia, then diffusion MR imaging may define the
full extent of the cerebral injury early. If undertaken as an
acute or subacute evaluation, the diffusion images should be
obtained in a way that allows compensation for T2 shine
through and permits calculation of ADC maps. Such sequences are widely available on commercial MR imaging systems.
References
1. Hack M, Flannery DJ, Schluchter M, et al: Outcomes in young adulthood for very-low-birth-weight infants. N Engl J Med 346:149-157,
2002
2. Hack M, Taylor HG, Klein N, et al: Functional limitations and special
health care needs of 10- to 14-year- old children weighing less than 750
grams at birth. Pediatrics 106:554-560, 2000
3. Hack M, Fanaroff AA: Outcomes of children of extremely low birth-
441
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
weight and gestational age in the 1990’s. Early Hum Dev 53:193-218,
1999
Saigal S, Szatmari P, Rosenbaum P, et al: Cognitive abilities and school
performance of extremely low birth weight children and matched term
control children at age 8 years: A regional study. J Pediatr 118:751-760,
1991
Whitaker AH, Van Rossem R, Feldman JF, et al: Psychiatric outcomes in
low-birth-weight children at age 6 years: Relation to neonatal cranial
ultrasound abnormalities. Arch Gen Psychiatry 54:847-856, 1997
Whitfield MF, Grunau RV, Holsti L: Extremely premature (ⱕ800 g)
schoolchildren: Multiple areas of hidden disability. Arch Dis Child
Fetal Neonatal Ed 77:F85-F90, 1997
Cooke RW, Abernethy LJ: Cranial magnetic resonance imaging and
school performance in very low birth weight infants in adolescence.
Arch Dis Child Fetal Neonatal Ed 81:F116-F121, 1999
Guzzetta F, Shackelford GD, Volpe S, et al: Periventricular intraparenchymal echodensities in the premature newborn: Critical determinant
of neurological outcome. Pediatrics 78:995-1006, 1986
Volpe JJ: Neurology of the Newborn (ed 4). Philadelphia, PA, W.B.
Saunders Co, 2001
Armstrong DL, Sauls CD, Goddard-Finegold J: Neuropathologic findings in short-term survivors of intraventricular hemorrhage. Am J Dis
Child 141:617-621, 1987
Takashima S, Mito T, Houdou S, et al: Relationship between periventricular hemorrhage, leukomalacia and brainstem lesions in prematurely born infants. Brain Dev 11:121-124, 1989
Leviton A, Gilles F: Ventriculomegaly, delayed myelination, white matter hypoplasia, and “periventricular” leukomalacia: How are they related? Pediatr Neurol 15:127-136, 1996
Golomb MR, Dick PT, MacGregor DL, et al: Cranial ultrasonography
has a low sensitivity for detecting arterial ischemic stroke in term neonates. J Child Neurol 18:98-103, 2003
Stewart A, Kirkbride V: Very preterm infants at fourteen years: Relationship with neonatal ultrasound brain scans and neurodevelopmental status at one year. Acta Paediatr Suppl 416:44-47, 1996
O’Shea TM, Klinepeter KL, Dillard RG: Prenatal events and the risk of
cerebral palsy in very low birth weight infants. Am J Epidemiol 147:
362-369, 1998
Wheater M, Rennie JM: Perinatal infection is an important risk factor
for cerebral palsy in very-low-birthweight infants. Dev Med Child Neurol 42:364-367, 2000
Nelson KB, Grether JK, Dambrosia JM, et al: Neonatal cytokines and
cerebral palsy in very preterm infants. Pediatr Res 53:600-607, 2003
De Vries LS, Van Haastert IL, Rademaker KJ, et al: Ultrasound abnormalities preceding cerebral palsy in high-risk preterm infants. J Pediatr
144:815-820, 2004
Pinto-Martin JA, Whitaker AH, Feldman JF, et al: Relation of cranial
ultrasound abnormalities in low-birthweight infants to motor or cognitive performance at ages 2, 6, and 9 years. Dev Med Child Neurol
41:826-833, 1999
Inder T, Anderson N, Spencer C, et al: White matter injury in the
premature infant: a comparison between serial cranial ultrasound and
MRI at term. AJNR Am J Neuroradiol 24:805-809, 2003
Miller SP, Cozzio CC, Goldstein RB, et al: Comparing the diagnosis of
white matter injury in premature newborns with serial MR imaging and
transfontanel ultrasonography findings. AJNR Am J Neuroradiol 24:
1661-1669, 2003
Nwafor-Anene VN, DeCristofaro JD, Baumgart S: Serial head ultrasound studies in preterm infants: How many normal studies does one
infant need to exclude significant abnormalities? J Perinatol 23:104110, 2003
Ment LR, Bada HS, Barnes P, et al: Practice parameter: Neuroimaging of
the neonate: Report of the Quality Standards Subcommittee of the
American Academy of Neurology and the Practice Committee of the
Child Neurology Society. Neurology 58:1726-1738, 2002
Whitelaw A, Pople I, Cherian S, et al: Phase 1 trial of prevention of
hydrocephalus after intraventricular hemorrhage in newborn infants by
drainage, irrigation, and fibrinolytic therapy. Pediatrics 111:759-765,
2003
J.J. Neil and T.E. Inder
442
25. Whitelaw A, Cherian S, Thoresen M, et al: Posthaemorrhagic ventricular dilatation: New mechanisms and new treatment. Acta Paediatr
Suppl 93:11-14, 2004
26. Davies MW, Swaminathan M, Chuang SL, et al: Reference ranges for the
linear dimensions of the intracranial ventricles in preterm neonates.
Arch Dis Child Fetal Neonatal Ed 82:F218-F223, 2000
27. Allan WC, Holt PJ, Sawyer LR, et al: Ventricular dilation after neonatal
periventricular–intraventricular hemorrhage. Natural history and therapeutic implications. Am J Dis Child 136:589-593, 1982
28. Hall P, Adami HO, Trichopoulos D, et al: Effect of low doses of ionising
radiation in infancy on cognitive function in adulthood: Swedish population based cohort study. Br Med J 328:19, 2004
29. Ron E, Modan B, Floro S, et al: Mental function following scalp irradiation during childhood. Am J Epidemiol 116:149-160, 1982
30. Brenner DJ, Doll R, Goodhead DT, et al: Cancer risks attributable to low
doses of ionizing radiation: Assessing what we really know. Proc Natl
Acad Sci U S A 100:13761-13766, 2003
31. Karlsson P, Holmberg E, Lundell M, et al: Intracranial tumors after
exposure to ionizing radiation during infancy: A pooled analysis of two
Swedish cohorts of 28,008 infants with skin hemangioma. Radiat Res
150:357-364, 1998
32. Flodmark O, Roland E, Hill A, et al: Radiologic diagnosis of periventricular leukomalacia. Acta Radiol Suppl 369:664-666, 1986
33. De Vries LS, Regev R, Connell JA, et al: Localized cerebral infarction in
the premature infant: An ultrasound diagnosis correlated with computed tomography and magnetic resonance imaging. Pediatrics 81:3640, 1988
34. Fitzhardinge PM, Flodmark O, Fitz CR, et al: The prognostic value of
computed tomography of the brain in asphyxiated premature infants.
J Pediatr 100:476-481, 1982
35. Ishida A, Nakajima W, Arai H, et al: Cranial computed tomography
scans of premature babies predict their eventual learning disabilities.
Pediatr Neurol 16:319-322, 1997
36. Roth SC, Edwards AD, Cady EB, et al: Relation between cerebral oxidative metabolism following birth asphyxia, and neurodevelopmental
outcome and brain growth at one year. Dev Med Child Neurol 34:285295, 1992
37. Penrice J, Cady EB, Lorek A, et al: Proton magnetic resonance spectroscopy of the brain in normal preterm and term infants, and early changes
after perinatal hypoxia-ischemia. Pediatr Res 40:6-14, 1996
38. Moseley ME, Cohen Y, Mintorovitch J, et al: Early detection of regional
cerebral ischemia in cats: Comparison of diffusion- and T2-weighted
MRI and spectroscopy. Magn Reson Med 14:330-346, 1990
39. Burdette JH, Elster AD, Ricci PE: Acute cerebral infarction: Quantification of spin-density and T2 shine-through phenomena on diffusionweighted MR images. Radiology 212:333-339, 1999
40. Mori S, van Zijl PC: Fiber tracking: Principles and strategies: A technical
review. NMR Biomed 15:468-480, 2002
41. McKinstry RC, Mathur A, Miller JP, et al: Radial organization of developing human cerebral cortex revealed by non-invasive water diffusion
anisotropy MRI. Cereb Cortex 12:1237-1243, 2002
42. Peterson BS, Vohr B, Staib LH, et al: Regional brain volume abnormalities and long-term cognitive outcome in preterm infants. J Am Med
Assoc 284:1939-1947, 2000
43. Sowell ER, Thompson PM, Holmes CJ, et al: Localizing age-related
changes in brain structure between childhood and adolescence using
statistical parametric mapping. Neuroimage 9:587-597, 1999
44. Mink JW, McKinstry RC: Volumetric MRI in autism: Can high-tech
craniometry provide neurobiological insights? Neurology 59:158-159,
2002
45. Wilson DA, Steiner RE: Periventricular leukomalacia: Evaluation with
MR imaging. Radiology 160:507-511, 1986
46. De Vries LS, Connell JA, Dubowitz LM, et al: Neurological, electrophysiological and MRI abnormalities in infants with extensive cystic leukomalacia. Neuropediatrics 18:61-66, 1987
47. Flodmark O, Lupton B, Li D, et al: MR imaging of periventricular
leukomalacia in childhood. AJNR Am J Neuroradiol 10:111-118, 1989
48. Barkovich AJ, Truwit CL: Brain damage from perinatal asphyxia: Cor-
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
relation of MR findings with gestational age. AJNR Am J Neuroradiol
11:1087-1096, 1990
Truwit CL, Barkovich AJ, Koch TK, et al: Cerebral palsy: MR findings in
40 patients. AJNR Am J Neuroradiol 13:67-78, 1992
Olsen P, Paakko E, Vainionpaa L, et al: Magnetic resonance imaging of
periventricular leukomalacia and its clinical correlation in children.
Ann Neurol 41:754-761, 1997
Lanzi G, Fazzi E, Uggetti C, et al: Cerebral visual impairment in
periventricular leukomalacia. Neuropediatrics 29:145-150, 1998
Valkama AM, Paakko EL, Vainionpaa LK, et al: Magnetic resonance
imaging at term and neuromotor outcome in preterm infants. Acta
Paediatr 89:348-355, 2000
Aida N, Nishimura G, Hachiya Y, et al: MR imaging of perinatal brain
damage: comparison of clinical outcome with initial and follow-up MR
findings. AJNR Am J Neuroradiol 19:1909-1921, 1998
Rutherford MA, Pennock JM, Counsell SJ, et al: Abnormal magnetic
resonance signal in the internal capsule predicts poor neurodevelopmental outcome in infants with hypoxic-ischemic encephalopathy. Pediatrics 102:323-328, 1998
De Vries LS, Groenendaal F, van Haastert IC, et al: Asymmetrical myelination of the posterior limb of the internal capsule in infants with
periventricular haemorrhagic infarction: An early predictor of hemiplegia. Neuropediatrics 30:314-319, 1999
Hunt RW, Neil JJ, Coleman LT, et al: Apparent diffusion coefficient in
the posterior limb of the internal capsule predicts outcome following
perinatal asphyxia. Pediatrics 114:999-1003, 2004
Arzoumanian Y, Mirmiran M, Barnes PD, et al: Diffusion tensor brain
imaging findings at term-equivalent age may predict neurologic abnormalities in low birth weight preterm infants. AJNR Am J Neuroradiol
24:1646-1653, 2003
Mazumdar A, Mukherjee P, Miller JH, et al: Diffusion-weighted imaging of acute corticospinal tract injury preceding Wallerian degeneration
in the maturing human brain. AJNR Am J Neuroradiol 24:1057-1066,
2003
Neil JJ, Inder TE: Detection of Wallerian degeneration in a newborn by
diffusion MR imaging. J Child Neurol 2004 (in press)
Inder T, Huppi PS, Zientara GP, et al: Early detection of periventricular
leukomalacia by diffusion-weighted magnetic resonance imaging techniques. J Pediatr 134:631-634, 1999
Soul JS, Robertson RL, Tzika AA, et al: Time course of changes in
diffusion-weighted magnetic resonance imaging in a case of neonatal
encephalopathy with defined onset and duration of hypoxic-ischemic
insult. Pediatrics 108:1211-1214, 2001
McKinstry RC, Miller JH, Snyder AZ, et al: A prospective, longitudinal
diffusion tensor imaging study of brain injury in newborns. Neurology
59:824-833, 2002
Mader I, Schoning M, Klose U, et al: Neonatal cerebral infarction diagnosed by diffusion-weighted MRI: Pseudonormalization occurs early.
Stroke 33:1142-1145, 2002
Forbes KP, Pipe JG, Bird R: Neonatal hypoxic-ischemic encephalopathy: Detection with diffusion-weighted MR imaging. AJNR Am J Neuroradiol 21:1490-1496, 2000
Cowan FM, Pennock JM, Hanrahan JD, et al: Early detection of cerebral
infarction and hypoxic ischemic encephalopathy in neonates using
diffusion-weighted magnetic resonance imaging. Neuropediatrics 25:
172-175, 1994
Wolf RL, Zimmerman RA, Clancy R, et al: Quantitative apparent diffusion coefficient measurements in term neonates for early detection of
hypoxic-ischemic brain injury: Initial experience. Radiology 218:825833, 2001
Counsell SJ, Allsop JM, Harrison MC, et al: Diffusion-weighted imaging
of the brain in preterm infants with focal and diffuse white matter
abnormality. Pediatrics 112:1-7, 2003
Miller SP, Vigneron DB, Henry RG, et al: Serial quantitative diffusion
tensor MRI of the premature brain: Development in newborns with and
without injury. J Magn Reson Imaging 16:621-632, 2002
Johnson AJ, Lee BC, Lin W: Echoplanar diffusion-weighted imaging in
neonates and infants with suspected hypoxic-ischemic injury: Correlation with patient outcome. AJR Am J Roentgenol 172:219-226, 1999
Imaging perinatal brain injury
70. Huppi PS, Maier SE, Peled S, et al: Microstructural development of
human newborn cerebral white matter assessed in vivo by diffusion
tensor magnetic resonance imaging. Pediatr Res 44:584-590, 1998
71. Neil JJ, Shiran SI, McKinstry RC, et al: Normal brain in human newborns: Apparent diffusion coefficient and diffusion anisotropy measured using diffusion tensor imaging. Radiology 209:57-66, 1998
72. Huppi PS, Murphy B, Maier SE, et al: Microstructural brain development after perinatal cerebral white matter injury assessed by diffusion
tensor magnetic resonance imaging. Pediatrics 107:455-460, 2001
73. Hoon AH Jr., Lawrie WT Jr., Melhem ER, et al: Diffusion tensor imaging
of periventricular leukomalacia shows affected sensory cortex white
matter pathways. Neurology 59:752-756, 2002
74. Nosarti C, Al-Asady MH, Frangou S, et al: Adolescents who were born
very preterm have decreased brain volumes. Brain 125:1616-1623,
2002
443
75. Panigrahy A, Barnes PD, Robertson RL, et al: Volumetric brain differences in children with periventricular T2-signal hyperintensities: A
grouping by gestational age at birth. AJR Am J Roentgenol 177:695702, 2001
76. Melhem ER, Hoon AH Jr., Ferrucci JT Jr., et al: Periventricular leukomalacia: relationship between lateral ventricular volume on brain
MR images and severity of cognitive and motor impairment. Radiology
214:199-204, 2000
77. Inder TE, Huppi PS, Warfield S, et al: Periventricular white matter
injury in the premature infant is followed by reduced cerebral cortical
gray matter volume at term. Ann Neurol 46:755-760, 1999
78. Inder TE, Wells SJ, Mogridge NB, et al: Defining the nature of the
cerebral abnormalities in the premature infant: A qualitative magnetic
resonance imaging study. J Pediatr 143:171-179, 2003