Hemorrhage detected using MR imaging in the setting of acute stroke: an in vivo model.

BACKGROUND AND PURPOSE The treatment algorithm for acute cerebrovascular accidents has traditionally sorted these accidents as either hemorrhagic or nonhemorrhagic, and MR imaging, with its ability to allow expeditious assessment of vascular substrates and regional blood volume, is well suited for this purpose. Our purpose was to delineate the accuracy of MR imaging in acute, hemorrhagic forms of stroke during the time frame considered beneficial for intervention in an animal model. METHODS Eighteen dogs with small, iatrogenic parenchymal, subarachnoid hemorrhage (SAH), or both were serially scanned over the initial 6-hour postictal period. Confirmatory pathologic specimens and 3-hour postictal CT scans were obtained in all animals. The MR and CT studies were then interpreted in a blinded fashion by two neuroradiologists for the presence of hemorrhage. The results were subjected to receiver operating characteristic analysis. RESULTS MR imaging depicted acute parenchymal hemorrhage and SAH with a high degree of accuracy at 1.5 T. This finding was independent of each of the time points studied during the 6-hour window. For SAH, the MR accuracy for reader 1 was 0.86 (95% CI, 0.76-0.97); for reader 2, accuracy was 0.85 (95% CI, 0.71-0.99). The CT accuracy for the two readers was 0.42 (95% CI, 0.26-0.58) and 0.66 95% CI, 0.43-0.89), respectively. Fluid-attenuated inversion-recovery images improved the conspicuity of SAH on MR images and, along with spin-density-weighted spin-echo sequences, helped to establish the hemorrhagic nature. For parenchymal hemorrhage, the MR accuracy for reader 1 was 0.90 (95% CI, 0.81-0.99); for reader 2, accuracy was 0.93 (95% CI, 0.84-1.00). With CT, the accuracy of reader 1 was 0.91 (95% CI, 0.85-0.97) whereas for reader 2 accuracy was 0.76 (95% CI, 0.69-.83). Parenchymal hemorrhage detection and diagnosis was best with T2*-weighted gradient-echo images. CONCLUSION MR imaging with appropriately selected sequences appears able to provide information regarding the presence (or absence) of hemorrhage in an acute stroke model requisite to the initiation of treatment.