NIMG-10. A MRI-BASED RADIOMICS MODEL PREDICTING DNA DAMAGE REPAIR GENES EXPRESSION SUBTYPES AND PROGNOSIS IN GLIOBLASTOMA

Abstract

Abstract BACKGROUND Glioblastoma (GBM) is the most common primary brain tumor, with a dismal prognosis. Some DNA Damage Repair (DDR) pathways are related to progression and worse prognosis in GBM patients. However, the association between overall expression levels of DDR genes and prognosis for GBM has not been reported. Radiomics has become an essential tool for assessment of clinical information in GBM patients, expected to be a noninvasive tool for predicting DDR genes expression and prognosis. METHODS The TCGA database was used to obtain the GBM patients imaging, clinical and RNAseq data. We screened DDR genes and clustered them into subgroups based on their expression status, and determined differences between subgroups in prognosis and molecular mechanisms. Radiomics model based on MRI CE-T1 image was investigated to predict DDR subgroups and the output was defined as the radiomics score (RS). Prognosis analysis of RS and traditional clinical indicators was performed in TCGA cohorts, and external validated in 2 medical centers. RESULTS A total of 156 individuals with GBM were identified from the TCGA cohorts. 140 genes were screened in DDR gene libraries, which were classified into 2 clusters. Survival analysis showed that overall survival (OS) was significantly higher in DDR suppression subgroup than activation subgroup (21.1 vs 14.9, p=0.001), and differences in immune microenvironment, biologically active pathways were observed in the 2 groups. A model composed of 4 radiomics features were built, which can distinguish 2 DDR subgroups. RS was an independent protective factor for OS (HR=0.630, 95%CI=0.400−0.992, P=0.046). The result was externally validated by two cohorts.