Objective. To compare findings on hard copies of cranial ultrasound (US) and magnetic resonance imaging (MRI) obtained between birth and term in a group of preterm infants.
Participants and Methods. Infants born at or below a gestational age of 30 weeks who underwent cranial US scan and MRI on the same day were eligible for this study. Infants underwent, whenever possible, 3 scans between birth and term. We calculated the predictive probability (PP) of US findings as a predictor of findings on MRI.
Results. Sixty-two paired MRI and US studies were performed between birth and term in 32 infants born at a median gestational age of 27 (range: 23–30) weeks and a median birth weight of 918 (530–1710) grams. US predicted some MRI findings accurately: germinal layer hemorrhage (GLH) on US had a PP of 0.8 with a 95% confidence interval of (0.70–0.90) for the presence of GLH on MRI, intraventricular hemorrhage (IVH) on US had a PP of 0.85 (0.76–0.94) for the presence of IVH on MRI, and severe white matter (WM) echogenicity on US had a PP of 0.96 (0.92–1.0) for the presence of WM hemorrhagic parenchymal infarction on MRI. Other MRI changes were less well-predicted: mild or no WM echogenicity on US had a PP of 0.54 (0.41–0.66) for the presence of normal WM signal intensity on MRI, and moderate or severe WM echogenicity on US had a PP of 0.54 (0.42–0.66) for the presence of small petechial WM hemorrhage and/or diffuse excessive high-signal intensity (DEHSI) in the WM on T2-weighted images on MRI. However, mild/moderate or severe WM echogenicity on US scans performed at ≥7 days after birth had a PP of 0.72 (0.58–0.87) for the presence of WM hemorrhage and/or DEHSI on MRI. There were no cases of cystic periventricular leukomalacia.
Conclusion. US accurately predicted the presence of GLH, IVH, and hemorrhagic parenchymal infarction on MRI. However, its ability to predict the presence of DEHSI and small petechial hemorrhages in the WM on T2 weighted images is not as good, but improves on scans performed at ≥7 days after birth. In addition, normal WM echogenicity on US is not a good predictor of normal WM signal intensity on MRI.
- GLH =
- germinal layer hemorrhage •
- IVH =
- intraventricular hemorrhage •
- HPI =
- hemorrhagic parenchymal infarction •
- WM =
- white matter •
- US =
- ultrasound •
- MRI; magnetic resonance imaging; GA =
- gestational age •
- FSE =
- fast spin echo •
- DEHSI =
- diffuse excessive high-signal intensity
The developing brain is susceptible to injury from a variety of ischemic, infective, inflammatory, and neurotoxic factors.1 Preterm infants are at high risk of developing germinal layer hemorrhage (GLH), intraventricular hemorrhage (IVH), hemorrhagic parenchymal infarction (HPI),2 cystic periventricular leukomalacia, and perhaps most commonly and most importantly, diffuse noncystic white matter (WM) injury.1Infants with these abnormalities, excluding possibly GLH and small IVH, are at increased risk of developing motor, cognitive, and other impairments.3–9
Although cranial ultrasound (US) detects IVH, GLH, HPI, and cystic changes, its sensitivity and specificity in detecting diffuse or subtle brain injury is poor based on pathologic correlation studies.2,10,11 WM cysts on cranial US are a strong predictor for the development of cerebral palsy,5 but there is accumulating evidence that very preterm infants without definite cranial US abnormalities are at risk of deficits in the cognitive, visuospatial, and visuomotor domains, although they may not have cerebral palsy.12 Although US is a very safe and accessible bedside tool, the image acquisition is machine-, probe-, and operator-dependent and is limited by the size of the fontanelle, the angulated view, and the signal attenuation with distance. In addition, abnormal echogenicity is not lesion-specific. All these issues may account for the difficulty in detecting subtle WM abnormalities using cranial US scans.
Magnetic resonance imaging (MRI), although not very accessible and very sensitive to movement artifact, provides high-resolution images of the brain without the use of ionizing radiation. It allows better definition of lesions in terms of site, extent, and type of pathology13 than US and computed tomography and provides more detailed characterization of WM both in infants13,14and in fetuses.15,16 Brain MRI findings correlate well with postmortem findings in the fetus17 and in preterm infants.18 A recent study correlating single MRI studies with serial US findings in preterm infants found a significant correlation between echogenicity on US and changes in signal intensity on MRI.19
The aim of this study was to compare findings on hard-copy cranial US scan with findings on contemporaneous MRI to determine if normality and abnormality co-occur between cranial US and MRI obtained from birth and term-age equivalent in a group of extremely preterm infants.
Ethical approval for studying the preterm infants' brain using MRI was given by the Hammersmith Hospitals Research Ethics Committee and parental consent was obtained in each case. Preterm infants born at or below a gestational age (GA) of 30 weeks who underwent a cranial US and MRI on the same day were eligible for this study. Infants were imaged either as part of a research cohort study enrolling all infants born below 30 weeks GA at the Hammersmith Hospital20 or for clinical indications. Therefore, not all infants were thought to have definite pathology on their US scan, and those from the research cohort were selected based on place of birth, GA at birth, and parental permission. GA was calculated from the date of the last menstrual period and confirmed using data from early antenatal US scan.
MRI and US scans were divided into 3 groups: Group A scans were obtained within the first 4 days after birth; Group B scans were obtained at a postnatal age of 7 to 95 days and were all done at a GA ≤36 weeks; Group C scans were all done between GA 38 and 44 weeks. US scans were performed on the same day as MRI.
MRI was performed using a 1 Tesla neonatal magnetic resonance (MR) system (Oxford Magnet Technology/Marconi Medical Systems, Cleveland, OH) located in the neonatal intensive care unit. The magnet has a 380-mm length bore, which allows good access to the infant during scanning. Full intensive care, including mechanical ventilation and monitoring, was continued during scanning when necessary, as previously described.21
The sequence parameters for T1-weighted conventional spin echo, T2-weighted fast spin echo (FSE), and inversion recovery FSE sequences are summarized in Table 1.
Magnetic resonance images were analyzed by 2 of the authors (E.F.M. and M.A.R.) by consensus. Features documented were the degree and distribution of WM signal intensity; petechial WM hemorrhages; HPI; IVH; GLH; basal ganglia; choroid plexus; and posterior fossa lesions.
Unmyelinated WM is characterized by a long T1 and a long T2, and was seen as high-signal intensity with the T2-weighted FSE sequence. Different degrees of high-signal intensity were identified within the WM.21 The WM was classified as having “diffuse and excessive high signal intensity' (DEHSI) if there were widespread areas within the WM with signal intensity approaching that of cerebrospinal fluid. This WM finding was considered to be abnormal.19,21
Myelin and hemorrhage were characterized by a short T1 and short T2, and appeared as low-signal intensity on T2-weighted FSE sequence. Hemorrhage was distinguished from myelin by its site and shape.13.
Germinal Layer, Basal Ganglia, Choroid Plexus, and Posterior Fossa
The germinal layer was characterized by very short T1 and T2 giving a low-signal intensity on T2-weighted images. GLH was distinguished from residual germinal layer by its irregular shape and asymmetry as previously described.13
Basal ganglia, choroid plexus, and posterior fossa lesions consistent with hemorrhage appeared as high-signal intensity on T1-weighted images and low-signal intensity on T2-weighted images.
Cranial US scans were obtained using an Ultra-Mark-4 mechanical sector US scanner with a multifrequency transducer (5–7.5–10 MHz crystals)(Advanced Technology Laboratories, Letchworth, UK). Operating at 7.5 MHz, paper copies using high quality printing paper (Sony Type I, Normal, UPP-1105, Sony Corporation, Tokyo, Japan) were made of US images for later analysis. US images were acquired before MRI in all cases. Five coronal and 5 sagittal and parasagittal views were taken in each case.
Hard copies of cranial US scans were analyzed by 1 of the authors (F.C.) who was blinded to infants' clinical details, except their GA and postnatal age at scan, to previous and subsequent US findings and to findings on MRI. Features documented were the degree of WM echogenicity; IVH; GLH; basal ganglia abnormality; choroid plexus; echogenicity; and posterior fossa echogenicity.
The degree of WM echogenicity was assessed and graded from 0 to 3 on a linear analog scale with the help of 4 sample hard-copy US scans representing grades 0,1,2, and 3 WM echogenicity. The WM was classified as having “normal echogenicity” when all regions within the WM scored 0; “mild increase in echogenicity” when the maximum WM echogenicity score was >0 and ≤1; “moderate increase in echogenicity” when the maximum WM echogenicity score was >1 and ≤2; and “severe increase in echogenicity” when the maximum WM echogenicity score was >2 and ≤3. WM echogenicity was recorded with respect to degree of echodensity but without an attempt at diagnosis or inference of underlying pathology. Examples of the degrees of echogenicity are seen in Figs 2–7. The presence or absence of echolucencies (cysts) within the WM was documented.
Germinal Layer, Ventricles, Basal Ganglia, Choroid Plexus, and Posterior Fossa
Infants were classified as having a GLH if they had increased echogenicity within the germinal layer in the region of the caudothalamic notch and as having an IVH if they had increased echogenicity within the ventricular system. The degree of echogenicity within the basal ganglia was assessed using the scale described above for WM echogenicity. Presence or absence of echolucencies (cysts) within the basal ganglia was documented. The choroid plexus and posterior fossa were described as echogenic if they contained areas of moderate and/or severe echogenicity using the scale described above.
The value of US as a predictor of MRI signal intensity was assessed by calculation of sensitivity and specificity as well as positive and negative predictive values. A Bayesian approach was used to analyze the value of the test further. β-density curves were calculated according to the method of Berry22 and the probability that a single examination would yield a correct prediction (the predictive probability) determined, together with 95% confidence limits for that value. The predictive probability is thus a useful summary value of the probability of a correct prediction.
Because US scans were analyzed with the observer blinded to previous and subsequent findings and because echogenicity on US and signal intensity on MRI in preterm infants evolve with time, images obtained from the same infant were analyzed separately. Each data pair was considered independent of any other data pair.
Based on previous knowledge, we chose the following US appearances as predictors for the presence of MRI findings: GLH as a predictor for the presence of GLH on MRI; IVH as a predictor for the presence of IVH on MRI; mild or no WM echogenicity as a predictor for the presence of normal WM signal intensity on MRI; severe WM echogenicity as a predictor for the presence of WM hemorrhage (either HPI or discrete punctate hemorrhages) on MRI; moderate or severe WM echogenicity as a predictor for the presence of WM hemorrhage and/or DEHSI on MRI; mild, moderate or severe WM echogenicity on scans performed at ≥7 days after birth, as a predictor for the presence of WM hemorrhage; and/or DEHSI on MRI.
Thirty-two infants born at a median GA of 27 (range: 23–30) weeks and a median birth weight of 918 (530–1710) grams underwent 62 US and MRI scans. Group A contained 24 scans, Group B contained 19 scans, and Group C contained 19 scans. Nine of the 32 infants had 3 scans each, 12/32 infants had 2 scans and 11/32 infants had 1 scan.
The sensitivity, specificity, positive predictive value, negative predictive value, and predictive probability (with 95% confidence interval) of cranial US findings as a predictor of MRI findings are shown in Table 2.
GLH/IVH, Choroid Plexus, and Posterior Fossa
Figure 1 summarizes the findings in Group A, B, and C scans. In Group A (n = 24), 7 infants had GLH on US and MRI and 2 had GLH on US only. Six infants had IVH on US and MRI, 1 had IVH on US only, and 2 had IVH on MRI only. In Group B (n = 19), 7 infants had GLH on US and MRI, 1 had GLH on US only, and 4 had GLH on MRI only. Three infants had IVH on US and MRI, 3 had IVH on US only, and 2 had IVH on MRI only. In Group C (n = 19), 4 infants had appearances consistent with a GLH on US only and 1 infant had IVH on US only.
The few IVHs (n = 4) that were seen on MRI but not on US were small hemorrhages within the posterior horns.
Increased choroid plexus echogenicity was documented in 27/62 (44%) US scans: 14 in Group A; 12 in Group B; and 1 in Group C. Posterior fossa echogenicity was documented in 8/62 (13%) US scans: 2 in Group A; 4 in group B; and 2 in group C. All posterior fossa echogenicity was located within the cerebellum (Fig 2). There were no lesions documented in the choroid plexus or in the posterior fossa on MRI.
Details of the WM findings on US and MRI for all groups of infants are shown in Table 3.
In Group A, 4/13 infants with severe or moderate echogenicity on US had abnormal signal intensity on MRI (Fig 3and 4). Eight of 11 infants with mild echogenicity on US had normal signal intensity on MRI.
In Group B, 7/9 infants with severe or moderate echogenicity on US had abnormal signal intensity on MRI. Five of 10 infants with mild echogenicity on US had normal signal intensity on MRI.
In Group C, 7/8 infants with moderate echogenicity on US had abnormal signal intensity on MRI (Fig 5 and 6). Three of 11 infants with mild or no echogenicity on US had normal signal intensity on MRI.
In Group A (n = 24), 2 infants had severe echogenicity on US, 1 of whom had hemorrhage seen on MRI (Fig 7). Three infants had mild echogenicity on US but no changes seen on MRI. One infant had no echogenicity on US, but a small discrete punctate head of caudate nucleus hemorrhage was seen on MRI.
In Group B (n = 19), 2 infants had severe echogenicity on US and basal ganglia hemorrhage was seen on MRI. Two infants had moderate and 3 had mild echogenicity on US, but no changes were seen on MRI. Two infants had no echogenicity on US, but small hemorrhages in the head of caudate nucleus were seen on MRI.
In Group C (n = 19), 1 infant had mild echogenicity on US, but no changes were seen on MRI.
WM and Basal Ganglia Echolucencies/Cysts
There were no cases of cystic periventricular leukomalacia. In Group A (n = 24), 1 infant had a single basal ganglia cyst on US and on MRI. Two infants had single basal ganglia cysts on US, but these were not seen on MRI. One infant had no WM cysts on US, but a single cyst was seen on MRI. In Group B (n = 19), 1 infant had a single basal ganglia cyst and another had a single WM cyst on US but none were seen on MRI. In Group C (n = 19), 1 infant had a single cyst in the WM on US but this was not seen on MRI.
Cranial US remains the main clinical tool for imaging the preterm infant brain. However, there is considerable controversy particularly about the significance of WM echogenicities in the preterm infant. Comparison of US with MRI findings may enable better understanding of their cause and prediction of clinical outcome, although the clinical significance of signal intensities, particularly DEHSI in WM on MRI in preterm infants, remains to be established.
In this study, we compared findings on reproducible hard-copy cranial US scans with findings on MRI performed on the same day. This method of obtaining and recording US images is identical to the day-to-day clinical practice in many neonatal intensive care units, although, unlike our study, information from preceding scans is usually available. The quality of hard-copies images depends on the type of printing paper used and the brightness/contrast settings of the printer. Hence, many clinicians might prefer to analyze cranial US scans on line but in practice most use hard copy. In addition, clinical records require the acquisition of reproducible copies of US scans.
We calculated the value of findings on a single US scan as a predictor of MRI signal intensity. Each US scan was analyzed, clinically and statistically, as a separate entity without reference to serial findings on previous or subsequent images.
US provides less even coverage of the brain compared with MRI making identification of abnormalities in deeper and peripheral structures such as the basal ganglia, posterior fossa, and cortex more difficult. However, cranial US is thought to be reliable at detecting lesions such as IVH, GLH, and HPI. Reported sensitivity of US in predicting the presence of IVH, GLH, and HPI at necropsy varies from 50%23 in some studies to 91% in others.24The specificity in these studies varied from 100%23 to 85%.24
There is less agreement about the clinical and pathologic significance of WM echogenicities on cranial US scans. Both the degree25,26 and the duration3,27–29 of WM periventricular echogenicities have been related to WM necrosis at necropsy and to neurodevelopmental disability. However, nonhemorrhagic WM necrosis may show no abnormality on cranial US scans.25The reported sensitivity of cranial US predicting the presence of WM lesions at necropsy ranges from 85%24 to 67%10 and 28%.2 The specificity in these studies varied from 93%24 to 86%.2 In one additional study, the specificity of US in predicting WM lesions at necropsy varied from 50% to 92% with different US interpreters.30
MRI has been used to obtain brain images during follow-up of ex-preterm infants and children31–34 but the systematic use of magnetic resonance imaging (MRI) in sick extremely low birth weight infants has been impractical. It is now possible to perform serial imaging safely to study brain development and to provide information on the timing and evolution of brain injury.35
There are very few studies that compare brain MRI findings with histopathologic findings in preterm infants18 and fetuses.16,17 However, MR is an accurate technique for diagnosing hemorrhagic brain lesions. GLH, IVH, HPI, and small petechial WM hemorrhages in preterm infants can be accurately detected using MRI.13 In addition, it has been suggested that MRI is superior to autopsy in diagnosing IVH.17 This is because the lengthy formalin preservation required for preterm brains may result in leakage of intraventricular blood from a hemorrhage,17 making the diagnosis of IVH more difficult at autopsy if the examination is delayed.
Detailed WM structure, including layers of migrating glial cells, are also clearly illustrated using brain MRI.13–16 In a recent study comparing brain MRI to histopathologic findings in preterm infants, excessively low signal intensity on FSE inversion recovery images in the periventricular WM corresponded to areas of necrosis.18 In addition, in a recent cohort of preterm infants, diffuse/excessive high signal intensity on FSE T2-weighted images and low-signal intensity on FSE inversion recovery images was closely associated with the development of features suggestive of cerebral damage, such as dilatation of the lateral ventricles, widening of the extracerebral space, and the interhemispheric fissure at term.21 Changes in signal intensity on MRI may therefore correspond to nonhemorrhagic WM lesions.
The results of this study showed that US was a good predictor for the presence of GLH, IVH, and HPI on MRI. Slightly more GLH was thought to be present on US on the early scans than was seen on MRI, indicating that it is easy to overestimate this diagnosis just after birth. The appearances may occur for the same reason as that of choroid plexus echogenicity, the cause of which is not known, which was often seen on early scans but not found to be hemorrhagic on MRI (see below). In the Group B scans, more GLH was seen on MRI. This may be because resolving GLH may seem normal on US, but irregularity and persisting low-signal intensity of the germinal matrix remains easily detectable on MRI. The late appearances of GLH on US not confirmed on MRI may be attributable to nonhemorrhagic echogenicity seen in this region that has been reported in older preterm infants.36
Not all IVH's were detected by US. MRI detected small IVH's in the depth of the posterior horns, which were not seen on US, probably because their site was relatively inaccessible to transfontenellar US. If the infants had been examined with US through the posterior fontanelle they might have been detected. It is unlikely that they were of any clinical significance.
Choroid plexus echogenicity was commonly seen on initial and repeat US scans, but was much less common at term. The cause and clinical relevance of choroid plexus echogenicity is not known. The choroid plexus is difficult to see on MRI unless there is hemorrhage within it or intravenous contrast is given. These choroid plexus echogenicities seen on US are, therefore, probably not related to hemorrhage, but may be related to relatively high blood flow in first days after birth. Similar appearances are also seen in term infants and they do not seem to carry any clinical significance.37 Posterior fossa echogenicity within the cerebellum was seen in an equal number of infants on early and term US scans but no posterior fossa abnormalities were seen; in particular, no evidence of hemorrhage was documented on MRI. The cause and clinical relevance of cerebellar echogenicity on US is not known.
Severe echogenicities within the basal ganglia were associated with basal ganglia hemorrhage on MRI, but small petechial hemorrhages in the head of the caudate nucleus that were present on MRI in 3 infants were not associated with echogenicity on US. This is not likely to be of any clinical significance, although some persisting biochemical abnormality has been detected in the caudate nuclei following GLH.38
There were only a few infants with echolucencies on US and none with frank cystic periventricular leukomalacia. This made it difficult to compare the accuracy of US in predicting the presence of cysts on MRI.
For the purpose of this study, WM echogenicity on US was assessed as a predictor of WM hemorrhage and/or DEHSI on T2-weighted images on MRI. On the early scans, WM echogenicity was common but not usually associated with the presence of hemorrhage or DEHSI on T2-weighted images on MRI. Even when echogenicity was severe, MRI scans could look normal This finding is consistent with what is already known (ie, that WM echogenicities in the first week after birth may be transient and of benign significance).28
WM echogenicities on US scans performed at a postnatal age >7 days co-occurred with MRI abnormality more often. All infants with severe echogenicity and 4 of 6 with moderate echogenicity had abnormal MRI scans, and half of the infants with mild echogenicity had abnormal MRI scans. All US scans in this interim period were thought abnormal when the MRI scan was abnormal. However, mild abnormality on US could be associated with a normal MRI scan.
In the later Group C scans, US was abnormal in most but not all infants where DEHSI was seen on MRI.
Thus severe WM echogenicity beyond the first week generally correlated with abnormality on MRI when the co-occurrence of US echogenicity and DEHSI occurred in ∼2/3 of the infants. This suggests that they may often represent the same process. Significantly, normal WM on US did not always predict normal WM on MRI. The physics of the 2 techniques is completely different, and normal developmental and pathologic processes may not be reflected with equal conspicuity. It is known that hemorrhage may remain visible on MRI for weeks, and longer than it is seen on US. The time course of nonhemorrhagic WM changes may also be different on US than on MRI.
A recent study19 compared WM findings on single MRI's performed at a mean GA of 33.4 (range: 30.6–37) weeks (mean age at MRI: 18.7 days) with findings on serial US scans including 1 performed on the same day as MRI. The infants all had normal neurologic examinations, and the WM on US was either normal or had mild echogenicities. A zone of high-signal intensity on T2-weighted images in the periventricular WM was reported to be associated with periventricular WM echogenicities on US. The authors concluded that MRI is probably more sensitive than US in early detection of mild periventricular WM lesions in preterm infants.19 No later MRI scans were performed, and therefore the evolution of the echogenicities could not be compared with the evolution of signal intensity on MRI. Our finding of DEHSI in the WM is probably similar to the high-signal intensity on T2-weighted images described in that study. The clinical significance of DEHSI is not yet known. It may partly reflect abnormal maturation rather than a pathologic process, and ongoing studies will correlate it with indices of perinatal sickness and infection, follow-up MRI scans, head growth, and neurodevelopmental outcome.
Cranial US is good at predicting the presence of GLH, IVH, and HPI on MRI. Normal WM echogenicity on US is not highly predictive of normal signal intensity on MRI. Mild, moderate, or severe WM echogenicity on US scans performed ≥7 days after birth is sensitive, but not specific for the presence of WM DEHSI or petechial hemorrhages on T2-weighted images on MRI. However, other cranial US scan findings, such as moderate echogenicities within the basal ganglia, choroid plexus, and cerebellum, are not associated with abnormal signal intensity on MRI in the preterm infant.
This study was supported by Wellbeing, Medical Research Council, Garfield Weston Foundation, Marconi Medical Systems, and Oxford Magnet Technology.
We thank Professor G. M. Bydder for his support and his help in interpreting the magnetic resonance images.
- Received December 16, 1999.
- Accepted December 8, 2000.
Reprint requests to (A.D.E.) Department of Pediatrics, Imperial College School of Medicine, Hammersmith Hospital, Ducane Rd, London W12 ONN. E-mail:
- Jongmans M,
- Henderson S,
- de Vries L,
- Dubowitz L
- Pinto-Martin JA,
- Riolo S,
- Cnaan A,
- Holzman C,
- Susser MW,
- Paneth N
- Robertson C,
- Sauve R,
- Christianson H
- The Scottish low birth weight study: 1. Survival, growth, neuromotor and sensory impairment. Arch Dis Child. 1992 67;675–681
- Battin M,
- Maalouf EF,
- Counsell S,
- et al.
- Childs AM,
- Ramenghi LA,
- Evans DJ,
- et al.
- Girard N,
- Raybaud C,
- Poncet M
- Felderhoff-Mueser U,
- Rutherford MA,
- Squier WV,
- et al.
- Berry D. Statistics, A Bayesian Approach. London, England: Duxberry Press; 1996:200–215
- Trounce JQ,
- Fagan D,
- Levene MI
- Hesser U,
- Katz-Salamon M,
- Mortensson W,
- Flodmark O,
- Forssberg H
- Appleton RE,
- Lee REJ,
- Hey EN
- Baarsma R,
- Laurini RN,
- Baerts W,
- Okken A
- van de Bor M,
- den Ouden L,
- Guit LG
- Copyright © 2001 American Academy of Pediatrics