Background: For MR perfusion-diffusion mismatch to be clinically useful as a means of selecting patients for thrombolysis, it needs to occur in real time at the MRI console. Visual mismatch assessment has been used clinically and in trials but has not been systematically validated. We compared the accuracy of visually rating console-generated images with offline volumetric measurements using data from the Echoplanar Imaging Thrombolytic Evaluation Trial (EPITHET).
Methods: Perfusion time-to-peak (TTP) and diffusion-weighted images (DWI) (as generated by commercial MRI console software) and T(max) perfusion maps (which required offline calculation) were visually rated. Perfusion-diffusion mismatch, defined as a ratio of perfusion:diffusion lesion volume of >1.2, was independently scored by 1 expert and 2 inexperienced raters blinded to calculated volumes and clinical information. Visual mismatch was compared with region-of-interest-based volumetric calculation, which was used as the gold standard.
Results: Volumetric calculation demonstrated perfusion-diffusion mismatch in 85/99 patients. Visual TTP-DWI mismatch was correctly classified by the experienced rater in 82% of the cases (sensitivity: 0.86; specificity: 0.54) compared to 73% for the inexperienced raters (sensitivity: 0.75; specificity: 0.57). The interrater reliability for TTP-DWI mismatch was moderate (kappa = 0.50). Visual T(max)-DWI mismatch performed better (agreement - 93 and 87%, sensitivity - 95 and 88%, specificity - 77 and 82% for the experienced and inexperienced raters, respectively).
Conclusions: The assessment of visual TTP-DWI mismatch at the MRI console is insufficiently reliable for use in clinical trials. Differences in perfusion analysis technique and visual inaccuracies combine to make visual TTP-DWI mismatch substantially different to volumetric T(max)-DWI mismatch. Automated software that applies perfusion thresholds may improve the reproducibility of real-time mismatch assessment.
Copyright 2010 S. Karger AG, Basel.