NIMG-15EMPLOYING PRE-CLINICAL GLIOBLASTOMA MODEL AND MRI DERIVED TEXTURE FEATURES FOR FUNCTIONAL VALIDATION OF RADIOGENOMICS

Abstract

A precise clinically relevant characterization of glioblastoma (GBM) heterogeneity and underlying genomic variability is urgently needed. To evaluate the true potential of radiogenomics, an emerging field for GBM research, we have tested its scope in a pre-clinical setting. First, glioblastoma stem cells (GSCs) are altered to express decreased levels of Periostin (POSTN), previously identified in a radiogenomic screen as a key pro-invasive gene. Second, orthotopic tumors arising from these GSCs, were scanned and analyzed to identify gene specific MRI texture feature (MRTF) signatures, which can simultaneously predict POSTN status in GBM patients. Doxycycline inducible short hairpin RNA mediated knockdown of POSTN in GSC lines were established. 5X10^6 cells were stereotactically implanted in nude mice frontal cortex and randomized into POSTN knockdown and control groups. Monitored tumor growth longitudinally by MRI (n = 3-4 per group). Region of interest (ROI) methods were applied to the rigid registration images based on axial T1-weighted imaging and corresponding fluid-attenuated inversion recovery sequences. After pre-processing, gray level co-occurrence matrices (GLCMs) were generated in 4 angular directions and 3 invariant measures (mean, range and angular variance) of each texture feature is calculated. A matrix of 3x20 invariant features is generated for each gray level. Mean values of normalized MRTFs from edema T1 post and T2, subgroup tumors arising from independent GSCs (p <0.009- < 0.039) whereas, a distinct set of MRTFs correctly subgroup tumors arising from GSCs with POSTN knockdown and Control (p < 0.01- < 0.047). Additionally, these features are highly conserved in their respective groups between mouse and human MRI scans. In conclusion, our results provide one of the first pre-clinical validation for the field of radiogenomics, where a specific gene status can be recognized by a specific set of MRI texture features, which in turn can predict gene status in an individual GBM patient.