Purpose: To perform a large-scale systematic comparison of the accuracy of all commonly used perfusion computed tomography (CT) data postprocessing methods in the definition of infarct core and penumbra in acute stroke.
Materials and methods: The collection of data for this study was approved by the institutional ethics committee, and all patients gave informed consent. Three hundred fourteen patients with hemispheric ischemia underwent perfusion CT within 6 hours of stroke symptom onset and magnetic resonance (MR) imaging at 24 hours. CT perfusion maps were generated by using six different postprocessing methods. Pixel-based analysis was used to calculate sensitivity and specificity of different perfusion CT thresholds for the penumbra and infarct core with each postprocessing method, and receiver operator characteristic (ROC) curves were plotted. Area under the ROC curve (AUC) analysis was used to define the optimum threshold.
Results: Delay-corrected singular value deconvolution (SVD) with a delay time of more than 2 seconds most accurately defined the penumbra (AUC = 0.86, P = .046, mean volume difference between acute perfusion CT and 24-hour diffusion-weighted MR imaging = 1.7 mL). A double core threshold with a delay time of more than 2 seconds and cerebral blood flow less than 40% provided the most accurate definition of the infarct core (AUC = 0.86, P = .038). The other SVD measures (block circulant, nondelay corrected) were more accurate than non-SVD methods.
Conclusion: This study has shown that there is marked variability in penumbra and infarct prediction among various deconvolution techniques and highlights the need for standardization of perfusion CT in stroke.
Supplemental material: http://radiology.rsna.org/lookup/suppl/doi:10.1148/radiol.12120971/-/DC1.
© RSNA, 2013.