Article Text

Download PDFPDF

Serial cranial ultrasonography or early MRI for detecting preterm brain injury?
  1. Annemarie Plaisier1,2,
  2. Marlou M A Raets1,
  3. Ginette M Ecury-Goossen1,
  4. Paul Govaert1,3,
  5. Monique Feijen-Roon1,2,
  6. Irwin K M Reiss1,
  7. Liesbeth S Smit1,4,
  8. Maarten H Lequin2,
  9. Jeroen Dudink1,2
  1. 1Division of Neonatology, Department of Pediatrics, Erasmus Medical Center—Sophia, Rotterdam, The Netherlands
  2. 2Department of Radiology, Erasmus Medical Center, Rotterdam, The Netherlands
  3. 3Department of Pediatrics, Koningin Paola Children's Hospital, Antwerp, Belgium
  4. 4Division of Pediatric Neurology, Department of Neurology, Erasmus Medical Center, Rotterdam, The Netherlands
  1. Correspondence to Dr Jeroen Dudink, Erasmus Medical Center—Sophia, Dr. Molewaterplein 60, Rotterdam 3015 GJ, The Netherlands; j.dudink{at}


Objective To investigate detection ability and feasibility of serial cranial ultrasonography (CUS) and early MRI in preterm brain injury.

Design Prospective cohort study.

Setting Level III neonatal intensive care unit.

Patients 307 infants, born below 29 weeks of gestation.

Methods Serial CUS and MRI were performed according to standard clinical protocol. In case of instability, MRI was postponed or cancelled. Brain images were assessed by independent experts and compared between modalities.

Main outcome measures Presence of preterm brain injury on either CUS or MRI and discrepant imaging findings on CUS and MRI.

Results Serial CUS was performed in all infants; early MRI was often postponed (n=59) or cancelled (n=126). Injury was found in 146 infants (47.6%). Clinical characteristics differed significantly between groups that were subdivided according to timing of MRI. 61 discrepant imaging findings were found. MRI was superior in identifying cerebellar haemorrhage; CUS in detection of acute intraventricular haemorrhage, perforator stroke and cerebral sinovenous thrombosis.

Conclusions Advanced serial CUS seems highly effective in diagnosing preterm brain injury, but may miss cerebellar abnormalities. Although MRI does identify these lesions, feasibility is limited. Improved safety, better availability and tailored procedures are essential for MRI to increase its value in clinical care.

  • Neonatology
  • Imaging
  • Neurology

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

What is already known on this topic?

  • MRI is considered the optimal imaging method to identify preterm brain injury, but clinical circumstances may preclude its use.

  • Cranial ultrasonography (CUS) allows serial scanning at bedside and technical developments are improving detection of injury.

What this study adds?

  • Combined use of advanced serial CUS and MRI improves detection of common patterns of preterm brain injury.

  • Compared to MRI, CUS seems more sensitive for recognising acute intraventricular haemorrhage, perforator stroke and sinovenous thrombosis, but less for small cerebellar haemorrhages.

  • Clinical feasibility of MRI is limited for critically ill preterm infants.


Neurodevelopmental problems are common in preterm infants.1 ,2 Early objective diagnosis of brain injury is important for prognostication and decision-making in neonatal intensive care. Current neuroimaging tools, such as MRI, are suited for quantitative assessment of injury and can provide insight into pathogenesis of preterm brain injury.3 MRI is powerful, non-ionising and has several advances to evaluate preterm brains; diffusion tensor imaging (DTI), functional MRI, volumetric MRI and proton MR spectroscopy allow quantification of disturbances in brain maturation and elucidate brain connectivity and functionality of infants born preterm. MRI is considered the best method to detect and quantify diffuse non-cystic white matter injury (WMI)4 and is increasingly performed at preterm age to provide early diagnosis of lesions.2 However, MRI is expensive, time consuming and challenging for critically ill infants.

Cranial ultrasonography (CUS) is relatively cheap, directly available and allows serial bedside scanning with limited disturbance of the infant. Traditionally, CUS is used to detect germinal matrix haemorrhage and intraventricular haemorrhage (GMH-IVH) and periventricular leukomalacia (PVL). Its value in detecting other lesions as well is increasing owing to technical developments such as high-resolution ultrasound (<200 µ), quantitative measurements and use of supplemental acoustic windows (mastoid and posterior fontanel).5–11 Limitations of CUS include observer dependency,12 the challenge of reproducible objective measurement and problems to detect posterior fossa abnormalities and cerebral cortical changes.9

Based on comparative studies between MRI and CUS regarding abilities to predict outcome, MRI is proposed as the imaging method of choice for high risk preterm infants13–15 especially when performed around term-equivalent age. However, these studies did not use additional acoustic windows, high-resolution ultrasound and Doppler imaging—as recommended by others.5 ,16–18 And, most importantly, the limitations of MRI in clinical context are often not fully considered.

Routine MRI scans at 30 weeks gestation have the advantage of providing early detection of preterm related brain injury, which can help clinical decision-making while infants are still requiring critical levels of support. Because of the critical period of brain plasticity between 30 weeks and term gestation, early MRI scans seem a logical neuroimaging starting point. Currently, more neonatal centres only perform routine term-equivalent scanning, which provides important information regarding brain injury just before discharge, but lack the advantages mentioned above. Although several hospitals perform scans at early preterm age, the clinical value of routine early MRI scans must first be established.

This study focuses therefore on feasibility of routine, clinical early MRI scanning compared with serial CUS in a vulnerable population in a prospective cohort study. Our aims were to investigate detection accuracy and clinical feasibility of serial CUS from birth until discharge compared with a routine MRI scan obtained from 30 weeks’ postmenstrual age (PMA) onwards in infants born <29 weeks’ gestational age (GA). We hypothesised that dedicated advanced serial CUS is equally effective as a single routine MRI scan at 30 weeks’ PMA to diagnose common brain lesions in preterm infants and has higher clinical availability.



Between May 2010 and January 2013, infants born below 29 weeks GA were recruited prospectively. Standard clinical neuroimaging included serial CUS from birth until discharge and MRI at 30 weeks’ PMA (29 4/7–30 4/7 weeks). MRI scans are timed at 30 weeks’ PMA to enable early detection of brain injury and early parental counselling. Furthermore, our NICU policy includes that infants are transferred to other (non-NICU) hospitals once certain clinical criteria are met. As a result, term-equivalent scanning raises important logistic issues as most preterm infants are still admitted in other hospitals when they reach term-equivalent age. Of 336 eligible infants, 29 were excluded because of congenital malformation (n=18), uncertainty regarding GA (n=5) or refusal of parental informed consent (n=6). The institutional review board approved this study and parental consent was obtained for all participants.


Serial CUS was performed by an experienced observer using Esaote MyLab 70 (Genova, Italy). According to extended clinical protocol, images were obtained in standard sections; six coronal and five sagittal/parasagittal planes through the anterior fontanel at days 0, 1, 2, 7 and, subsequently, once a week until discharge. Additional images of the cerebellum and transverse sinus were acquired through the mastoid fontanel. Serial colour Doppler imaging was performed to assess the intracranial (sino-) venous and arterial system (figure 1). Images were acquired with a convex 8.5 MHz probe. To obtain higher resolution of superficially located areas, a high frequency linear probe (13 MHz) was used at the anterior and mastoid fontanel.

Figure 1

Ultrasound images were obtained in six coronal (A–F) and five sagittal/parasagittal planes (G–I) through the anterior fontanel. Additional images were acquired through the mastoid fontanel (J) to visualise the cerebellum and colour Doppler images (K–L) were acquired to assess arterial and (sino-) venous systems.


MRI procedures were carried out according to our clinical guideline:19 MRI scanning was postponed if patients were haemodynamically and respiratorily unstable, which was evaluated by the attending neonatologist and nursing staff. All infants were accompanied by trained staff and were placed in an MRI compatible incubator, which allowed controlled temperature and humidity and MR compatible pulse oximetry and ventilation. Mouldable earplugs and neonatal earmuffs protected infants from auditory noise; sedative drugs were not administered.

Imaging data were acquired with a 1.5 T GE EchoSpeed scanner (General Electrics Healthcare Technologies, Waukesha, Wisconsin, USA) (figure 2). Clinical imaging protocol included conventional MRI sequences and DTI. Axial T2-weighted fast-spin echo was obtained with the following parameter settings: repetition time (TR) 13 100 ms; echo time (TE) 139 ms; slice thickness 1.2 mm; field of view (FOV) 190×190 mm2; and scanning time 3:51 min. Axial 3D T1-spoiled gradient recalled echo (SPGR) was acquired using: TR 9 ms; TE 3 ms; slice thickness 1.6 mm; FOV 150×150 mm2; and scanning time 2:57 min. DTI was performed using a single-shot echo-planar-imaging sequence with diffusion gradients in 25 non-collinear directions, TR 11 725 ms; TE 85.6 ms; slice thickness 3 mm; FOV 220×220 mm2; b value 750 s/mm2; number of non-diffusion-weighted images three; and scanning time 5:17 min.

Figure 2

Applied MRI sequences: (A) axial T2-weighted fast-spin echo; (B) axial 3D T1-SPGR, diffusion tensor imaging; (C) apparent diffusion coefficient-map; and (D) susceptibility weighted imaging.

For the sake of optimisation during the study, advanced sequences were added to clinical scanning protocol: susceptibility weighted imaging (SWI) was performed using: TR 75 ms; TE 48 ms; slice thickness 2.2 mm; FOV 210×210 mm2; and scanning time 4:15 min. Arterial spin labelling was executed using: TR 4200 ms; TE 10 ms; postlabel delay 1025 ms; slice thickness 4 mm; FOV 220×220 mm2; and scanning time 6.05 min.

Assessment of brain injury

CUS and MRI data were assessed for signs of preterm brain injury by experienced investigators independently (MMAR/PG for CUS and AP/MHL for MRI, with >20 years of experience in neonatal neuroimaging for PG and MHL) using a detailed classification system that covers for most common types of brain injury20 and has appropriately been described elsewhere.21 In all cases, consensus was reached between investigators. IVH was graded according to Volpe.22 WMI was classified into cystic PVL and diffuse non-cystic WMI; the latter were defined as periventricular inhomogeneous echodensities on CUS or diffuse WMI on MRI.4 ,17 Cerebellar haemorrhage was categorised into folial or lobar cerebellar haemorrhage.23 Presence of brain injury on CUS scans was based on cumulative findings of all serial assessments up until discharge from our NICU. Discrepancies between CUS and MRI were scored and we assessed whether CUS yielded additional diagnoses after the MRI was performed.

Statistical analysis

Statistical analysis was performed using SPSS V.20. Analysis was done in three steps: (1) description of clinical characteristics (table 1) and imaging findings (table 3) of all infants; (2) comparison of clinical characteristics between imaging groups (table 2) to highlight clinical feasibility of early MRI scanning in a vulnerable population compared with serial (bedside) CUS; and (3) comparison of imaging findings on CUS versus MRI to investigate their detection abilities (table 4). Descriptive statistics were applied to patient characteristics and neonatal morbidities. GA was calculated from the first date of the last menstrual period; severity of illness was assessed with the score for neonatal acute physiology perinatal extension;24 intrauterine growth restriction was defined as birth weight below 2 SDs; persistent ductus arteriosus was recorded if it required treatment; and necrotising enterocolitis was defined by pneumatosis intestinalis, hepatobiliary gas or free intraperitoneal air on conventional X-rays. Differences between imaging groups were analysed with one-way analysis of variance for continuous variables (with post hoc Bonferroni) and by χ2 tests for dichotomous/ordinal data (with post hoc analysis using standardised residuals). Combined sum of findings by CUS and MRI served to calculate abilities of both imaging techniques to detect injury patterns. A p value of <0.05 (two-sided) was considered statistically significant.

Table 1

Subject characteristics

Table 2

Subject characteristics between imaging groups

Table 3

Patterns of preterm brain injury (combination of cranial ultrasonography and MRI findings)

Table 4

Discrepant imaging findings on cranial ultrasonography (CUS) and MRI in groups I and II


Subject characteristics

A total of 307 infants (170 boys) were included in this study with mean GA of 26 weeks, 5 days and birth weight of 922 g. Additional clinical characteristics are listed in table 1. All 307 infants were serially scanned using CUS according to imaging protocol, with a maximum delay of 12 h. In contrast, MRI was not performed at all in 126 infants, as 57 died before 30 weeks’ PMA; 55 were transferred to other hospitals before MRI scanning could be performed; and MRI was not performed in 14 due to logistic difficulties. At 30 weeks’ PMA, 73 infants were considered not stable enough for MRI scanning; 59 of them were eventually scanned at a later time. Thus, three different groups with regard to MRI scanning are distinguished: group I: MRI scanning at 30 weeks’ PMA (n=122); group II: MRI scanning after 30 weeks’ PMA (n=59); and group III: no MRI scanning (n=126) (figure 3).

Figure 3

Flowchart of formation of neuroimaging groups.

Differences between imaging groups

Clinical characteristics between the three MRI imaging groups differed significantly (table 2). Post hoc analysis revealed that infants in imaging group I were born with higher GA and birth weight and seemed to have fewer complications: persistent ductus arteriosus, supplementation of postnatal steroids and death were significantly less common and score for severity of illness was significantly lower. These characteristics differed most in comparison with imaging group III. Occurrence of necrotising enterocolitis did not significantly differ between imaging groups (table 2).

Patterns of preterm brain injury

Combined imaging findings

Injury patterns found either with CUS or MRI are listed in table 3 (figure 4). GMH-IVH was detected in 100 infants, WMI in 10, cerebellar haemorrhage in 21, cerebral sinovenous thrombosis (CSVT) in 11 and in four infants perforator stroke was identified.

Figure 4

Examples of brain injury patterns on MRI (A–C) and cranial ultrasonography (D–E): (A) bilateral periventricular haemorrhagic infarction on axial T2-weighted scan, note the ventricular dilatation caused by intraventricular haemorrhage; (B) punctate white matter lesions on axial T2-weighted scan; (C) multiple folial cerebellar haemorrhages on sagittal T2-weighted scan; (D) cerebral sinovenous thrombosis in transverse sinus on cranial ultrasound, assessed through mastoid fontanel, note the large mass (arrow) between the lateral ventricle (left arrowhead) and cerebellum (right arrowhead); and (E) periventricular leukomalacia on parasagittal cranial ultrasound, assessed through the anterior fontanel.

Ultrasonographic findings

In all, 180 infants (58.6%) had normal CUS. GMH-IVH was seen in 80 infants (26.7%); in 23 this was limited to germinal matrix (7.5%), in 39 it was assigned IVH grade II (12.7%), in six IVH grade III (2.0%) and in 12 (3.9%) haemorrhage was complicated by parenchymal infarction. WMI was sonographically detected in seven infants (2.2%); in four of them (1.3%), it was diffuse non-cystic WMI and in 3 (1.0%), cystic PVL was detected. Lobar cerebellar haemorrhage was identified in 10 infants (3.3%). Folial cerebellar haemorrhages were not recognised with CUS. CSVT was present in 11 infants; in all, the transverse sagittal sinus was involved, and in one infant there was also almost complete thrombosis of the superior sagittal sinus. Four infants presented with a perforator stroke on CUS.

MRI findings

MRI was performed in 180 infants and did not show any injury in 112 infants (62.2%). GMH-IVH was present in 43 infants (23.8%): GMH in 20 (11.1%), IVH-II in 14 (7.7%) and periventricular haemorrhagic infarction in nine infants (5.0%). WMI was detected in eight infants (4.4%); in seven (3.9%), this was diffuse non-cystic WMI and one infant had cystic PVL. Cerebellar haemorrhage was detected with MRI in 15 infants (8.3%), lobar in six and folial in nine. CSVT was identified on MRI in two infants; perforator stroke was not detected at all by MRI.

Discrepant imaging findings on CUS and MRI

Table 4 compares imaging findings of CUS with MRI findings in the 180 infants who were scanned by both techniques (group I and II). Inconsistencies were predominantly found for GMH-IVH in 38 infants, cerebellar haemorrhage in 11 and CSVT in seven infants. MRI had higher abilities to detect GMH and identified all posterior fossa abnormalities, but in 27 infants, CUS excelled in the acute detection of IVH grade II–III, perforator strokes and CSVT, as these lesions were no longer clearly visible on MRI at the time of scanning. Findings for WMI were discrepant in three infants. None of the discrepancies between CUS and MRI were detected after the MRI was performed.


This study demonstrates the high number of preterm infants with detectable brain injury (47.6%) and shows how serial advanced CUS effectively detects most common lesions. CUS has higher clinical feasibility than MRI, which cannot always be performed in severely ill infants. However, despite mastoid fontanel scanning, CUS remains inferior for identifying small posterior fossa abnormalities. MRI provided important additional diagnostic information. Although other studies have predominantly investigated the value of MRI and CUS to detect WMI, the main strength of this study is that it investigated their value for several brain injuries. In addition, this study demonstrates the complementary roles of both imaging modalities to detect common patterns of preterm brain injury: MRI was better to detect GMH and posterior fossa abnormalities, whereas CUS was better at grade II–III IVH, perforator stroke and CVST.

Although MRI performed at term-equivalent age is highly predictive of neurodevelopmental outcomes, prognostication can currently only be accurately made until that moment. Performing early MRI scans may result in earlier prognostication and may be of special interest in settings where infants are often transferred to level II hospitals once certain clinical criteria are reached. Moreover, early by proxy MRI biomarkers of future neurodevelopmental outcome can also guide randomisation of future neuro-protective and -rehabilitation trials. However, before early MRI scans can be implemented as standard of care for preterm infants, clinical applicability and sensitivity must first be established. This study addressed these matters of early MRI scans in a vulnerable population.

Comprehensive application of CUS, usage of supplemental acoustic windows, colour Doppler imaging, higher transducer frequencies and careful interpretation of images by an experienced observer result in high accuracy in identifying certain lesions. Serial CUS outperformed MRI scan in diagnosing focal lesions in 27 infants, mostly because of higher sensitivity to detect perforator stroke and CSVT.

Clinical relevance of diagnosing perforator stroke and CSVT seems important as detection of these lesions may have therapeutic and diagnostic consequences. Infants with CSVT may need thrombolytic treatment and as with perforator strokes may prompt assessment of thrombophilic factors. Perforator strokes are small and more challenging to detect on MRI with lower resolution in this brain area than CUS. Therefore, we believe that timing is not the only reason for the mismatch in diagnostic tools. Thinner slices and advanced sequences (such as SWI) may improve identification of perforator strokes using MRI.

The higher ability of CUS to detect IVH grade II seems mainly attributable to consecutive application of CUS and timing of onset of IVH. Conventional MRI did not always detect low-grade IVH, possibly due to the resolving nature of IVH. Parodi et al25 recently reported a lower sensitivity (60%) of CUS to detect grade I–II GMH-IVH compared with SWI. The authors point out the unique possibilities of SWI to detect subependymal and/or intraventricular haemosiderin depositions and accentuate that dedicated timing and application of advanced MRI sequences are valuable in assessing preterm brains.

In the present study, CUS was insufficient to detect diffuse non-cystic WMI in only three infants. All other diffuse WMI on MRI were already detected on CUS by inhomogeneous echodensities on CUS. This confirms the assumption that inhomogeneous hyperechogenicities are the CUS correlates of punctate WMI and stresses the important value of advanced dedicated serial CUS to detect diffuse non-cystic WMI.7 ,9 ,16–18 However, MRI is needed to assess extent and localisation of WMI for prediction of its full impact on outcome.26 In this study, a low number of diffuse non-cystic white matter abnormalities were found. This could be explained by missing data, as MRI could not be performed in clinical unstable preterm infants, which are typically the ones at increased risk to develop WMI. In addition, the low number of diffuse WMI could also be abnormalities that are commonly seen at term-equivalent age, such as diffuse excessive high signal intensities. As we scanned at mean PMA of 31 weeks, these abnormalities may have been missed.

In correspondence with current literature,27 ,28 MRI in the present study excelled in the detection of posterior fossa abnormalities: it confirmed all cerebellar haemorrhages detected with CUS and identified all folial and two lobar cerebellar haemorrhages that had been missed with CUS.

In addition to the detection of limited cerebellar haemorrhages, neonatal care may clearly benefit from quantitative MRI sequences that could provide early objective biomarkers of outcome.3 However, MRI is a complex technique with limitations in the very young. Our study design dictated MRI scanning at 30 weeks’ PMA, but depending on clinical condition, scanning was postponed or cancelled in 185 infants (60%). Inherently, infants scanned at 30 weeks’ PMA may likely have been less troubled by complications. Postponement or cancellation seems worrying because especially preterm infants with severe illness are at risk of brain injury and may benefit most from early MRI scanning.29 ,30 It would be essential, therefore, to improve applicability of MRI. This includes: (1) safety improvement: transfers and monitoring of critically ill infants to the MR suite should be optimised and MRI sequences can be shortened to reduce procedure times.31 ,32 (2) Tailored MRI scanning: indications, usage of sequences and timing to scan should be established more individually. (3) Improved availability: a dedicated neonatal MRI scanner in the vicinity of the NICU would overcome logistic problems.

Important limitations of this study should be addressed: (1) Imaging group III was heterogeneous because it included both deceased patients and infants transferred to other hospitals before the MRI could take place. Inclusion of deceased patients may have influenced clinical characteristics. However, as severe brain injury was frequently encountered in the infants who died, MRI was preferred to enable optimal clinical decision-making, but could not take place due to clinical instability. This emphasises the value of bedside sequential CUS and difficulties to enable safe MRI scanning in a vulnerable population. (2) Due to absence of correlation between injury patterns and long term outcome in our study population, we were not able to demonstrate clinical relevance of all findings. (3) We were unable to calculate sensitivity of CUS and MRI because golden standard of preterm brain injury (eg, by histopathological findings) was not available. (4) Since strategies were being optimised over the course of this study, SWI was not performed in all MRI procedures, which may have led to lower sensitivity to detect low-grade GMH-IVH compared with other studies.25 (5) In this study, we observed imaging findings of routine, clinical MRI scans; as such, we were not able to perform postprocessing of advanced MRI sequences to obtain quantitative measurements of injury. We are aware that sophisticated use of MRI sequences, such as DTI, volumetric MRI and proton MR spectroscopy, which are usually performed in the context of research, would undoubtedly result in greater value of MRI. (6) We performed MRI scanning at 30 weeks’ PMA instead of at term-equivalent age. This could have led to loss of patients to compare. However, given the increasing demand for early quantitative biomarkers of outcome, the importance of early MRI scans will increase as well. This study therefore provides valuable information regarding practicability and clinical limitations of such early MRI scans.


Brain injury is frequently encountered in preterm infants (47.6%). Advanced CUS is adequate to detect and monitor preterm brain injury and therefore deserves more appreciation in neonatal neurology. MRI is invaluable as it allows objective quantitative assessment of microstructural brain properties and is superior to detect posterior fossa abnormalities. However, clinical use in preterm infants is currently limited because of safety and logistic issues. These issues need to be addressed in view of increasing demand for quantitative biomarkers of outcome using early MRI scans. Furthermore, dual use of sequential CUS and MRI provides high sensitivity to detect common patterns of preterm brain injury. Future research should therefore focus on improvement of their complementary applications.


The authors would like to thank Rogier de Jonge, clinical epidemiologist, for performing statistical analyses and interpreting the results of this study.



  • Contributors AP, MMAR, PG and MHL: conceptualised and designed the study, carried out the data collection and analysis, drafted the initial manuscript and approved the final manuscript as submitted. GME-G, MF-R, IKMR and LSS: carried out data collection and interpretation, critically reviewed the manuscript and approved the final manuscript as submitted. JD: conceptualised and designed the study, coordinated and supervised data collection and analysis, critically reviewed the manuscript and approved the final manuscript as submitted.

  • Funding None.

  • Competing interests None.

  • Ethics approval Medical Ethical Committee of the Erasmus Medical Center, Rotterdam, the Netherlands.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Linked Articles

  • Editorial
    Linda S De Vries Manon J N L Benders Floris Groenendaal
  • Fantoms
    Ben Stenson