Cerebral magnetic resonance imaging within 6 hours of stroke onset: inter- and intra-observer reproducibility

Background: Magnetic resonance imaging (MRI) provides valuable pathophysiological information during the very first hours of cerebral ischemia. However, the reliability of prime-time MRI in the setting of emergency care remains unknown.

Aim: To evaluate the reproducibility between and within observers of the assessment of MRI scans in stroke patients.

Method: We performed a MRI scan within 6 h of stroke onset, with time-of-flight (TOF), T2* gradient echo, FLAIR, diffusion- (DWI) and perfusion- (PWI) weighted images, in 17 consecutive patients. Four observers, blinded to the clinical history, separately performed a visual assessment of all scans, and repeated the assessment 2-8 days later. Two neuroradiologists made volumetric measures of diffusion and perfusion abnormalities using a semi-automatic technique 2 weeks after the 2nd visual assessment. We evaluated: (i) in the whole set of MRI scans, the quality of scans and their ability to identify primary hemorrhages on T2* gradient echo sequences; (ii) in patients with acute cerebral ischemia only, the inter- and intra-observer agreement for the presence of arterial occlusion and cerebral abnormalities on TOF sequences, and (iii) on DWI and PWI sequences, the relationship between visual and automatic assessments for the presence of a mismatch (defined as the difference between the perfusion and diffusion abnormalities) of >20%. Statistics used the kappa (kappa) method.

Results: The median delay between clinical onset and MRI was 285 min. Two patients had primary cerebral hemorrhages, 1 a post-ictal deficit, and 14 cerebral ischemia. The quality of the scans was judged as appropriate for all scans in all sequences except for FLAIR. All observers identified the 2 patients with hemorrhages. The inter- and intra-observer reliability was substantial to excellent (kappa values ranging from 0.63 to 1.00) for all sequences. The agreement between visual and automatic assessments for the presence of a mismatch of >20% was excellent in all observers.

Conclusion: The visual assessment of T2* gradient echo, TOF, diffusion and perfusion sequences at the acute stage of stroke is reproducible between and within observers. The visual assessment is as good as the volumetric assessment to detect a mismatch of >20%.