Automatic vessel removal in gliomas from dynamic susceptibility contrast imaging.

The presence of macroscopic vessels within the tumor region is a potential confounding factor in MR-based dynamic susceptibility contrast (DSC)-enhanced glioma grading. In order to distinguish between such vessels and the elevated cerebral blood volume (CBV) of brain tumors, we propose a vessel segmentation technique based on clustering of multiple parameters derived from the dynamic contrast-enhanced first-pass curve. A total of 77 adult patients with histologically-confirmed gliomas were imaged at 1.5T and glioma regions-of-interest (ROIs) were derived from the conventional MR images by a neuroradiologist. The diagnostic accuracy of applying vessel exclusion by segmentation of glioma ROIs with vessels included was assessed using a histogram analysis method and compared to glioma ROIs with vessels included. For all measures of diagnostic efficacy investigated, the highest values were observed when the glioma diagnosis was based on vessel segmentation in combination with an initial mean transit time (MTT) mask. Our results suggest that vessel segmentation based on DSC parameters may improve the diagnostic efficacy of glioma grading. The proposed vessel segmentation is attractive because it provides a mask that covers all pixels affected by the intravascular susceptibility effect.