Reliability of dynamic susceptibility contrast perfusion metrics in pre- and post-treatment glioma

Abstract

Abstract Background In neuro-oncology, dynamic susceptibility contrast magnetic resonance (DSC-MR) perfusion imaging emerged as a tool to aid in the diagnostic work-up and to surveil effectiveness of treatment. However, it is believed that a significant variability exists with regard to the measured in DSC-MR perfusion parameters. The aim of this study was to assess the observer variability in measured DSC-MR perfusion parameters in patients before and after treatment. In addition, we investigated whether region-of-interest (ROI) shape impacted the observer variability. Materials and methods Twenty non-treated patients and a matched group of twenty patients post-treatment (neurosurgical resection and post-chemoradiotherapy) were included. Six ROIs were independently placed by three readers: circular ROIs and polygonal ROIs covering 1) the tumor hotspot; 2) the peritumoral region (T2/FLAIR-hyperintense region) and 3) the whole tumor region. A two-way random Intra-class coefficient (ICC) model was used to assess variability in measured DSC-MRI perfusion parameters. The perfusion metrics as assessed by the circular and the polygonal ROI were compared by use of the dependent T-test. Results In the non-treated group, circular ROIs showed good–excellent overlap (ICC-values ranging from 0.741–0.963) with the exception of those representing the tumor hotspot. Polygonal ROIs showed lower ICC-values, ranging from 0.113 till 0.856. ROI-placement in the posttreatment group showed to be highly variable with a significant deterioration of ICC-values. Furthermore, perfusion metric assessment in similar tumor regions was not impacted by ROI shape. Discussion This study shows that posttreatment quantitative interpretation of DSC-MR perfusion imaging is highly variable and should be carried out with precaution. Pretreatment assessment of DSC-MR images, however, could be carried out be a single reader in order to provide valid data for further analyses.