Glioblastoma (GBM) is the most common type of primary brain tumor, with a median survival of 15 months, even with treatment [1]. Magnetic resonance imaging (MRI) is the first choice for diagnosing brain tumors, helping radiologists identify various tumor states, the aggressiveness of GBM, and is crucial for planning tumor resection. In this study, radiomics was used to identify imaging features for more accurate prognosis prediction and a better understanding of GBM. We developed a tool that segments the tumor into different subregions based on MRI images of GBM patients and extracts features based on intensity values within the tumor, morphological characteristics, and features describing its location. The tool was used to analyze 52 patients, with the goal of combining radiomic features with general, clinical, and molecular data. The results of the study showed that tumors in patients with shorter survival are larger in surface area, volume, and axis lengths of the enclosing ellipsoid, are more irregularly shaped, and are more often located deeper in the brain. Based on Cox regression, a successful predictive model was developed (concordance=0.818), which uses five features to summarize patient survival in the study cohort. The model identified the sphericity of peritumoral edema and therapy with ionizing radiation and temozolomide as statistically significant positive survival predictors (p-value < 0.001). The flatness of the necrotic subregion also positively influenced survival, while higher expression levels of the NOTCH gene and the overlap of gadolinium-enhancing tumor, with the putamen negatively influenced patient survival.
|
