Chad Ashley Holder MD, Presenter: Nothing to Disclose

James Scott Cordova BS, Abstract Co-Author: Nothing to Disclose

Eduard Schreibmann PhD, Abstract Co-Author: Research Grant, Varian Medical Systems, Inc

Constantinos G. Hadjipanayis MD, PhD, Abstract Co-Author: Nothing to Disclose

Ying Guo PhD, Abstract Co-Author: Nothing to Disclose

Hyunsuk Shim PhD, Abstract Co-Author: Nothing to Disclose

This work aims to standardize and evaluate an MR signal-based approach for tumor segmentation using an FDA 510k-approved software package (Velocity AI) that allows the rendering, fusion, and analysis of multi-modality 3D medical images.

Currently, glioblastoma (GBM) volume measurements rely on the product of orthogonal tumor diameters on post-contrast T1w MRI; however, it is difficult to measure post-resection tumor in this manner, especially when hyperintense, nonneoplastic lesions are present. Though the need for objective volumetric analysis was highlighted by the NeuroOncology Working Group (Wen, PY et al. JCO 2010; 28,11 1963-1972), a standardized image display, processing, and analysis protocol has not been developed for a clinically-utilized volume rendering software. We applied our volume determination method to compare the extent of resection (EOR) using 5-ALA-guided resection to EOR of standard resections. Datasets consisted of high-resolution pre- and post-op MR images (T1w images pre- and post-contrast) from 13 randomized patients in an Emory ALA study and 13 controls matched for tumor location. To tabulate preop tumor volume, a coarse ROI was drawn around the tumor and the software was used to segment volumes of hyper- and hypointensity on T1w MRI in the ROI in a semi-automated fashion. To estimate residual post-op tumor, image difference maps were produced by subtracting co-registered, pre- and postcontrast T1w MRI to correct for postop blood.

The average EOR without ALA-guidance–expressed as percent residual tumor–was 10.69 ± 7.45%, while that of ALA-guidance was 4.85 ± 3.98%. These values were found to be significantly different at p<0.01 using the nonparametric Wilcoxon Rank-Sum Test.

These results support the use of this semi-automated method for the unbiased and reproducible generation of contrast-enhancing tumor volumes in GBM pre- and post-resection. In addition, this technology allows the selection of voxels in discrete tumor regions on T1w MRI for the quantitative analysis of treatment-induced metabolic changes in spatially-coregistered, high-resolution MR spectroscopic images.

This method allows quantitative analysis of brain tumor response to chemo-, radiation, and surgical therapies, offering a precise tool for the longitudinal monitoring of patients in clinical trials.

Holder, C, Cordova, J, Schreibmann, E, Hadjipanayis, C, Guo, Y, Shim, H, Development of an Unbiased, Semi-automated Method of Tumor Volume Segmentation Using Image Processing Software in Glioblastoma before and after Resection.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13028535.html