UC San Diego

Glioblastoma, the most common and deadly form of primary brain cancer, is characterized by rapid progression, heterogeneity, and defiance of therapy. The relentless nature of glioblastoma emphasizes the urgency of identifying improved methods to hasten the development of tailored treatments for patients afflicted by this malignancy. Genetic profiling of clinical glioblastoma specimens has revealed that glioblastoma, like other cancers, is composed of many different subtypes that may possess unique sensitivities to therapeutics. To improve the clinical outcome of glioblastoma patients, technologies must be developed to better define and discriminate the subtypes of glioblastomas in an affordable, accurate, and noninvasive manner. The heterogeneity of glioblastoma's genomic and molecular alterations mirror the diversity of its appearance in medical imaging. The emerging field of radiogenomics integrates methods common to neuroimaging, bioinformatics, and molecular biology to identify the radiographic correlates of tumor cellular and molecular processes. Application of radiogenomics to the study of glioblastoma may facilitate its understanding, especially when considering that magnetic resonance (MR) imaging is required for the modern clinical management of this disease. Unfortunately, radiogenomic progress demands accurate and high-throughput methods to reliably segment features from vast and varied imaging archives, and the careful design of metrics which capture biological phenotypes. In this context, we developed a robust algorithm for tumor segmentation and radiophenotype parameterization termed Iterative Probabilitic Voxel Labeling (IPVL). Application of IPVL to glioblastoma tumor images from The Cancer Imaging Archive (TCIA) with associated genomic profiling available via The Cancer Genome Atlas (TCGA) led to the topographic mapping of glioblastoma spatial distributions by molecular subtype, and the discovery of two survival-associated radiographic parameters. These parameters, tumor subventricular distance (SVZd), and lateral ventricle displacement (LVd), correlate with defined physiologic mechanisms, and associated genomic profiles. Together, these results provide proof of principle that quantitative radiographic assessment of glioblastoma is a viable and effective strategy capable of augmenting the power of molecular and genomic research. With further study in the clinical setting, application of these methodologies and novel imaging parameters could impact prognostic evaluation, identify tumor therapeutic subgroups, and hopefully improve the lives of patients