Computed tomography (CT), which uses ionizing radiation, is employed extensively for neonatal brain imaging of term infants. Magnetic resonance (MR) ...
Pediatr Radiol (2003) 33: 442–449 DOI 10.1007/s00247-003-0933-6
Richard L. Robertson Caroline D. Robson David Zurakowski Sharon Antiles Keith Strauss Robert V. Mulkern
Received: 30 June 2002 Accepted: 26 March 2003 Published online: 13 May 2003 Ó Springer-Verlag 2003
R.L. Robertson (&) Æ C.D. Robson D. Zurakowski Æ S. Antiles Æ K. Strauss R.V. Mulkern Department of Radiology, Children’s Hospital Medical Center, Harvard Medical School, 300 Longwood Avenue, Boston, MA 02115, USA E-mail: Richard.Robertson@ tch.harvard.edu Tel.: +1-617-3556303 Fax: +1-617-2647120
ORIGINAL ARTICLE
CT versus MR in neonatal brain imaging at term
Abstract Background : Recent reports have highlighted the lifetime risk of malignancy from using ionizing radiation in pediatric imaging. Computed tomography (CT), which uses ionizing radiation, is employed extensively for neonatal brain imaging of term infants. Magnetic resonance (MR) provides an alternative that does not use ionizing radiation. Objective: The purpose of this study was to assess the cross-modality agreement and interobserver agreement of CT and MR brain imaging of the term or near-term neonate. Materials and methods: Brain CT and MR images of 48 neonates were retrospectively reviewed by two pediatric neuroradiologists. CT and MR examinations had been obtained within 72 h of one another in all patients. CT was obtained with 5 mm collimation (KV=120, mAs=340). MR consisted of T1-weighted imaging (TR/TE=300/ 14; 4-mm slice thickness/1-mm gap), T2-weighted imaging (TR/TE/etl= 3000/126/16; 4-mm slice thickness/ 1-mm gap), and line scan diffusion imaging (LSDI) (TR/TE/b factor=1258/63/750; nominal 4-mm slice thickness/3-mm gap). The brain was categorized as normal or abnormal on both CT and MR. Results: Ischemic injury was the most common brain abnormality demonstrated. McNemar’s test
indicated no significant difference between CT and MR test results for reader 1 (P=0.22) or reader 2 (P=0.45). The readers agreed on the presence or absence of abnormality on CT in 40 patients (83.3%) and on MR in 45 patients (93.8%). For CT, the kappa coefficient indicated excellent interobserver agreement (j=0.68), although the lower limit of the 95% confidence interval extends to j=0.55, which indicates only good-to-moderate agreement. For MR, the kappa coefficient indicated almost perfect interobserver agreement (j=0.88) with the 95% confidence interval extending to a lower limit of j=0.76, which represents excellent agreement. Conclusion: Because MR demonstrates findings similar to CT and has greater interobserver agreement, it appears that MR is a superior test to CT in determining brain abnormalities in the term neonate. Furthermore, since MR eliminates the use of ionizing radiation, a putative cause of malignancy, it should be the standard in neonatal brain imaging. Future efforts should be directed to improving neonatal access to MR to avoid the routine use of CT in infants. Keywords Infant Æ Newborn Æ X-ray computed magnetic resonance imaging
443
Introduction In the neonate, neurologic manifestations of intracranial abnormalities are often nonspecific and/or nonlocalizing [1, 2]. Alterations in level of consciousness, irritability, seizures, respiratory distress, poor feeding, failure to thrive and focal neurologic deficit are all possible indicators of intracranial disease. Underlying causes include intracranial hemorrhage, hypoxic-ischemic injury, focal infarction, post-hemorrhagic hydrocephalus, vascular anomalies and malformations, infections, tumors or phakomatoses. Ultrasound (US), computed tomography (CT) and magnetic resonance (MR) imaging have been used extensively to evaluate the neonatal brain. In order to help define the most effective neonatal neuroimaging approach, a practice parameter for neuroimaging of the neonate was recently issued by the Quality Standards Subcommittee of the American Academy of Neurology and the Practice Committee of the Child Neurology Society [3]. The committee’s recommendations in the term newborn focused on imaging of the encephalopathic neonate. In these infants, it was recommended that CT be obtained to exclude hemorrhage and early MR be performed to evaluate the brain further [3]. The role of cranial US in term infants was less clear, prompting the conclusion that there were few data to support the use of US in imaging of the term encephalopathic neonate [3]. Tissue contrast in CT is dependent on differential attenuation of the X-ray beam. Although capable of displaying a variety of intracranial pathologies, CT is considered particularly useful for the detection of acute hemorrhage and calcifications [4]. While the infant must be transported to the CT scanning suite for imaging, lifesupport systems are generally compatible with the CT environment and the examinations are relatively short, making CT relatively easy to obtain even in the ill neonate. Although CT has been used effectively for neuroimaging for nearly 3 decades, the theoretical lifetime risk of cancer-related mortality associated with the use of ionizing radiation in CT for pediatric imaging has recently been highlighted [5, 6, 7, 8]. The risk is greatest when exposure to ionizing radiation occurs in the neonatal period. Furthermore, since CT techniques are frequently not adjusted for age, the radiation dose to the brain from a single-head CT in the neonate is often higher than the same examination in the adult. It has been estimated that the relative lifetime cancer mortality risk attributable to the radiation exposure from a single-head CT in the neonate is 0.08%–0.10% as compared to less than 0.01% in the adult [7]. Additionally, some indications for neonatal neuroimaging such as hydrocephalus necessitate multiple examina-
tions during a patient’s lifetime, adding to the theoretical cancer risk. While arguably small, the radiation risk associated with brain CT could be avoided altogether by using other methods of imaging that do not rely on ionizing radiation. Over the last several years, MR has emerged as a viable method of imaging the newborn brain without the use of ionizing radiation. The manipulation of multiple tissue properties with MR provides greater flexibility in imaging than is available with the singletissue property evaluated in CT (X-ray attenuation). With MR, image signal intensity is related to the overall water content of the tissue, as well as the inherent tissue properties T1, T2, proton flow, proton diffusion, paramagnetism, magnetic susceptibility, and chemical shift [4]. Moreover, MR images may be obtained in any desired anatomic plane without repositioning the patient. These capabilities make MR a useful imaging tool for evaluating the central nervous system of the neonate. However, MR has historically been considered less sensitive than CT for the detection of acute hemorrhage and calcification. In recent years, the use of higher field strength and better quality MR imagers utilizing sequences sensitive to perturbations in the magnetic field have improved the delineation of hemorrhage and calcification with MR [4]. Compared to CT, though, MR imaging of the ill neonate is potentially more problematic owing to the requirement for MR compatible life-support systems and longer scan times. Because CT and MR have unique advantages and disadvantages in neonatal brain imaging, it is a common practice in many institutions, including ours, to obtain both examinations. This approach is supported by the published practice parameters already mentioned [3]. Since there is overlap in the abnormalities potentially shown by CT and MR, it is possible that MR could be routinely substituted for CT, provided lesion detection were comparable. The purpose of this study is to assess two important indicators of lesion demonstration; the cross-modality and interobserver agreement of CT and MR interpretations. The role of US in the evaluation of these infants is not addressed.
Materials and methods Patients A review of our CT and MR database revealed 51 neonates examined from 1998 through 1999 who had both brain CT and MR with diffusion imaging during the first 2 weeks of life with no more than 72 h between examinations. Three patients were excluded because of excessive motion degrading the images, leaving a total of 48 patients with diagnostic quality CT and MR
444
examinations. CT had been obtained before MR in 39 patients. The average time between the CT and MR examinations was 1.7 days. All imaging studies were retrospectively evaluated by two experienced pediatric neuroradiologists. A remote time period, 1998 to1999, was chosen for the study to decrease the likelihood that the observers would recall the individual examinations. In addition, scanning parameters and equipment for both CT and MR remained unchanged during this interval. The neonates in this study had been referred for a variety of clinical indications. Gestational age at birth ranged from 36 weeks to 43 weeks (mean=39.3 weeks). Six infants were between 36 weeks and 38 weeks gestational age. All patients were accompanied to the imaging areas by a specialized intensive care transport team. The transport team was responsible for monitoring patient vital signs during the examination and for the administration of sedation, if required. Infants were routinely tightly swaddled in both CT and MR to decrease patient motion. No sedation was required for CT. Sedation was necessary for 25 MR examinations.
Statistical methods Concordance between CT and MR for the demonstration of brain parenchymal abnormalities was evaluated using McNemar’s test [9]. The kappa (j) coefficient with a 95% confidence interval (CI) was used to determine interobserver agreement for CT and MR with the strength of agreement interpreted using the following benchmarks by Landis and Koch [10]: almost perfect agreement (j=0.81–1.00), excellent agreement (j=0.61–0.80), good or moderate agreement (j=0.41–0.60), fair to poor agreement (j= 0.00–0.40). Power analysis revealed that a total sample size of 48 patients would provide 80% statistical power (b=0.2) to detect concordance between CT and MR, as well as good-to-moderate interobserver agreement using kappa (version 4.0, nQuery Advisor, Statistical Solutions, Saugus, Mass.). Two-tailed values of P
