Abstract 2266: Image analysis techniques affect the predictive potential of MRI-based blood volume maps for assessing survival in patients treated for gliomas

Abstract

Abstract In this study we tested the hypothesis that the type of method used to analyze a physiological parameter can greatly improve the parameter's predictive value. We compared the voxel-based technique, the parametric response map (PRM) to several common methods for assessing response. In addition, we evaluated for each metric the impact of the volume of interest (VOI) and the time the mid-treatment perfusion map was acquired in a cohort of glioma patients. Patients (n=44) with Grade III/IV glioma were recruited in a retrospective imaging trial. Patients underwent MRI before RT, 1 and 3 weeks after RT. MRI scans were acquired on a 1.5T or a 3T clinical scanner. The MRI protocol included contrast-enhanced T1w imaging, fluid-attenuated inversion recovery imaging (FLAIR) and dynamic contrast-susceptibility T2*w imaging. The relative cerebral blood volume (rCBV) in the brain and tumor were computed as described previously [1]. All images were co-registered to the post T1w images acquired before RT [1]. Following co-registration, the tumor VOIs were manually contoured either on T1w or on the FLAIR images and applied to the rCBV maps. For each patient, mid-treatment time point and VOI the following parameters were analyzed: histogram based approach for monitoring percent change of the rCBV mean, median and percentiles in increments of 10% [3] (plus 25th and 95th), a histogram segmentation approach for monitoring the percent change of the tumor volume fraction with low, medium or high-rCBV values [2] and voxel-based approach which is PRM. Statistics: Receiver operator characteristic analysis (ROC), assessed for 12 month survival, was used to determine the optimal cutoff for each parameter. Only those parameters that generated significantly large area under the curves where further analyzed for predicting overall survival. Overall survival for each parameter was then determined using Kaplan-Meier curves and the log-rank test p<0.05. Among all the parameters tested, only 4 were statistically significant using the ROC analysis. Indeed, only the PRMrCBV- parameters at week 1 and 3 and using the Gd or the FLAIR VOIs appeared significant using the ROC analysis. Considering the various time points and VOIs, the optimal cutoffs had a small range of 4.6-6.8 which deviated by only 32% from the 6.8% previously reported by our group. The 4 PRMrCBV- were able to predict significantly different patient outcomes irrespective of the time the mid-treatment rCBV map was acquired or VOI. Only PRM was found to be predictive of early treatment response. In addition, this approach is shown to be very robust with negligible sensitivity to the choice of VOI or the time the mid-treatment rCBV map was acquired making it ideally suited for multi-center trials. [2] Cao, Y., et al. 2006 Int JROBP [1] Galbán, C. J., et al. 2009 Nat Med [3] Li, K. L., et al. 2005 JMRI [4] Barajas, R. F., et al. 2010 AJNR. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr 2266. doi:10.1158/1538-7445.AM2011-2266