254. Are MRI Readings In Acute Pediatric Traumatic Brain Injury Reliable?

Authors: Myron Rolle, MD; Paul Caruso, MD; Emily Evans, PhD; Pavani Adapa, MD; Ann-Christine Duhaime, MD (Boston, MA)


Use of MRI in evaluation and management of acute traumatic brain injury (TBI) in children has become more common, because of increased sensitivity to parenchymal and physiologic abnormalities and absence of radiation.  However, the reliability of MRI interpretation in acute injury has not been studied.  We analyzed a series of consecutive TBI patients admitted to our Pediatric Intensive Care Unit to measure the inter-rater reliability of specific findings on MRI done within 4 days of injury.


Images were obtained for clinical purposes and included a variety of sequences chosen for specific clinical questions.  Two board-certified neuroradiologists and two neurosurgeons analyzed 35 consecutive MRI studies without knowledge of clinical details or other imaging results.  Images were systematically assessed for a range of specific pathoanatomic and physiologic entities including skull fracture, extra-axial and parenchymal hemorrhages and lesions, axonal injury, ischemia, midline shift, and others, following the NIH Common Data Elements Definitions Dictionary. Weighted kappa values were calculated for agreement between readers for presence, absence, or indeterminate findings of each diagnostic entity assessed on each scan.


Weighted kappa values were highest (0.62-1.0, moderate to almost perfect) for epidural hematoma, midline shift, contusion, and traumatic axonal injury, and lower for skull fracture, subarachnoid hemorrhage, and subdural hematoma. While some scans showed obvious large lesions, many scans showed small or equivocal findings. Disagreements between raters mostly involved small lesions or technically limited scans such as rapid T2 studies; larger, more clinically relevant lesions generally had higher levels of agreement.


Use of specific definitions for diagnostic entities on MRI in pediatric TBI may augment interrater reliability and promote uniform terminology, but smaller lesions on tailored clinical scans still may be interpreted differently by different readers.  This should be kept in mind when determining treatment based on MRI findings.