How reliable is perfusion MR in acute stroke? Validation and determination of the penumbra threshold against quantitative PET

Background and Purpose - Perfusion magnetic resonance imaging ( pMR) is increasingly used in acute stroke, but its physiologic significance is still debated. A reasonably good correlation between pMR and positron emission tomography ( PET) has been reported in normal subjects and chronic cerebrovascular disease, but corresponding validation in acute stroke is still lacking. Methods - We compared the cerebral blood flow ( CBF), cerebral blood volume, and mean transit time ( MTT) maps generated by pMR ( deconvolution method) and PET ( O-15 steady-state method) in 5 patients studied back-to-back with the 2 modalities at a mean of 16 hours ( range, 7 to 21 hours) after stroke onset. We also determined the penumbra thresholds for pMR-derived MTT, time to peak ( TTP), and Tmax against the previously validated probabilistic PET penumbra thresholds. Results - In all patients, the PET and pMR relative distribution images were remarkably similar, especially for CBF and MTT. Within-patient correlations between pMR and PET were strong for absolute CBF ( average r(2) = 0.45) and good for MTT ( r(2) = 0.35) but less robust for cerebral blood volume ( r(2) = 0.24). However, pMR overestimated absolute CBF and underestimated MTT, with substantial variability in individual slopes. Removing individual differences by normalization to the mean resulted in much stronger between-patient correlations. Penumbra thresholds of approximate to 6, 4.8, and 5.5 seconds were obtained for MTT delay, TTP delay, and Tmax, respectively. Conclusions - Although derived from a small sample studied relatively late after stroke onset, our data show that pMR tends to overestimate absolute CBF and underestimate MTT, but the relative distribution of the perfusion variables was remarkably similar between pMR and PET. pMR appears sufficiently reliable for clinical purposes and affords reliable detection of the penumbra from normalized time-based thresholds.