Observer reliability in CT and MRI of the abdomen/pelvis

The purpose of this research was to evaluate two sources of error in the performance of computerized tomography (CT) and magnetic resonance imaging (MRI) of the abdomen/pelvis. The sources of error assessed were inter- and intra-observer reliability. Thirty abdomen/pelvis CT scans were randomly selected from each of three hospitals (university, VA, military) with different CT scanners. Two radiologists were recruited from each site to be CT observers. Forty-five abdomen/pelvis MRI scans were randomly selected from two institutions with different MRI scanners. Four observers were recruited to read the MRI scans. All scans were read blind without clinical information or patient identification. Overall inter-observer and intra-observer diagnostic agreement was significantly higher for MRI compared to CT. Inter-observer diagnostic agreement rates were also significantly higher for MRI when the etiologies of neoplastic vascular and metabolic/toxic were assigned. Observer experience in CT (range: 5-9 yr) or MRI (range: 2-4 yr) was not statistically associated with improved diagnostic agreement. This research addresses many of the criticisms of the MRI literature and compares MRI favorably to CT.