Nimg-25. Radiomic Profiling Maps Areas of Pathologically Confirmed Tumor and Predicts Overall Survival in Pre-treatment Glioblastoma
Year: 2016
Type: conference-abstract
Abstract: Quantifying tumor heterogeneity in glioblastoma (GBM) plays a critical role in diagnosis as well as therapy design. Magnetic resonance imaging (MRI) can be used to identify and differentiate heterogeneous GBM characteristics such as hypercellularity, necrosis, and treatment effects. This study involves encoding four different types of MRI contrasts to produce a single radiomic image and quantify both inter and intra tumor heterogeneity. We hypothesized that patient prognosis can be predicted by the volume and location of specific radiomic profiles (RP) prior to therapy. Initial imaging data from 82 patients pre-treatment was retrospectively analyzed for this experiment. This included T1 (pre and post Gd contrast), apparent diffusion coefficient (ADC), and fluid attenuation inversion recovery (FLAIR) images. The volume of each RP was assessed on a patient by patient basis within the T1 enhancement and FLAIR hyperintensity. These values were normalized by tumor size to correct for known relationships between tumor size and prognosis. We first performed an exploratory analysis correlating RP volume to patient survival. A log-rank test with an iterative threshold was performed to determine what RP volume cutoff best predicted prognosis. The final pre-mortem imaging was analyzed similarly in three additional autopsied patients. Tissue samples were obtained from RPs that correlated with survival in the larger group. After correction for multiple comparisons, six RPs were significantly correlated with overall survival: three within contrast enhancement and three within FLAIR hyperintensity. Regions of pseudopalisading necrosis and hypercellular tumor were found in the areas indicated by these RPs. We present a pathologically-validated method for generating MRI-based radiomic profiles correlated with overall survival in patients with glioblastoma. These predictive profiles were generated prior to treatment, indicating potential as a precision medicine tool.
Source: Neuro-Oncology
Institutions: Medical College of Wisconsin, Froedtert Hospital
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W2587938779