Data processing and analysis
All PCT images and parameters were analysed using a standard workstation (MMWP, Siemens, Erlangen, Germany) with commercially available software (Volume Perfusion CT, Siemens, Erlangen, Germany). Automatic segmentation was applied to exclude non-parenchymal pixels such as bone, cerebrospinal fluid (CSF) or vessels. In order to obtain peak vascular enhancement in blood, the superior sagittal sinus was selected as the venous reference vessel for the PCT process, as with 10-mm-thick axial sections a reliable absolute density evaluation of cerebral arteries can be restricted because of partial volume effects [1]. By applying the perfusion software, quantitative parameter images were generated from the time-attenuation curves. For each patient, four types of parameter maps were calculated for each section: temporal maximum intensity projection (MIP) in Hounsfield units (HU), CBV (ml/100 ml), CBF (ml/100 ml/min) and the volume transfer constant K Trans as a measure of permeability (ml/100 ml/min). CBF was calculated using the maximum slope model [9], CBV and K Trans were calculated using Patlak analysis (Appendix 1). The shape of the arterial input function necessary for the Patlak analysis was automatically determined from branches of the MCA or ACA, the peak of the input function was normalized to the peak of the superior sagittal sinus.
The raters then independently determined and manually drew regions of interest (ROIs) on the maps. Initially, ROIs were drawn on the MIP images, on the solid part of tumour, trying to exclude areas with necrosis or vessels. The ROIs were then automatically copied onto the perfusion maps and corresponding CBV, CBF and K Trans values were acquired. For every patient, reference ROIs were also drawn on the healthy contralateral hemisphere and perfusion parameters were obtained as control values.