Interobserver agreement in magnetic resonance brain and neck imaging

Aim. To investigate the interobserver reliability of brain and neck magnetic resonance imaging (MRI) in older community subjects previously recruited to a case-control study of dizziness. Methods. Subjects aged over 65 years were identified through a local survey and advertising. Several investigations were performed, including MRI of the head and neck using a Siemens 1.5 Tesla machine. Images were coded independently by two consultant neuroradiologists. Interobserver agreement was calculated using kappa statistics. Results. Of the 246 participants, scans from 84 dizzy subjects and 79 non-dizzy subjects were available for coding by two neuroradiologists. Agreement was “fair” for cerebral atrophy, “good” for white matter lesions (WML) in the cerebral hemispheres, and moderate for WML in the posterior fossa. Agreement for cervical spine disease ranged from “moderate” (severity of cord compression) to “poor” (vertebral artery occlusion). Conclusion. Interobserver variability in the evaluation of brain and cervical cord MRI was substantial. The development of standardized scores, particularly for disease of the cervical cord, may improve interobserver reliability.