Cranial sonography (CS) is the most commonly used tool for imaging the neonatal brain. Yet, little is known about inter-rater reliability and the effect of image quality on final interpretation. As part of a multi-center study, 1524 CS on 590 infants were interpreted for presence and severity(Papile grade) of intracranial hemorrhage (ICH) by 3 pediatric radiologists experienced in cranial sonography. Discrepancies among the 3 central readers were adjudicated and final interpretation reached by consensus. The quality of a subset of sonograms (1004) was assessed as good, poor but readable, and unreadable. Final interpretation by central readers was compared to that of local readers and congruence assessed by the Kappa statistic.
Observed overall agreement was high (86%-88%, Kappa.507-.587). However, central and local readers disagreed with respect to grade of ICH in 230/1524 sonograms (15%). The most common discrepancies were normal vrs Grade I ICH (n= 84/152, 55%), Grade I vrs. Grade II or higher (n=19, 13%), and Grade II vrs Grade III or higher (n=9, 6%). In discrepant cases, local readers interpreted the study as abnormal when consensus reading was normal in 53%, and assigned a higher grade of ICH than central readers in 64% of sonograms. Of 1004 sonograms, 655 (35%) were judged as poor quality by central readers. Frequency of discrepant readings was twice as high (23%) in studies judged to be of poor quality, compared to 11% in good quality studies.
