Radiomic and Genomic Approaches to Survival Stratification in Adult-type Diffuse Glioma

Abstract

Adult-type diffuse gliomas are the most common primary adult malignant brain tumors and are generally considered incurable. Alongside tumor molecular heterogeneity and barriers to drug delivery, inadequate survival stratification is a critical challenge for patient clinical management, drug development, and clinical trial enrollment. In this work, we will discuss several radiomic and genomic approaches to risk stratification of types of adult-type diffuse glioma using magnetic resonance imaging (MRI), somatic copy number alteration (SCNA), gene expression, DNA methylation, and clinical data. From MRI data, we train an efficient adult-type diffuse glioma tumor segmentation model that automates a critical piece of standard radiomic pipelines, present a novel radiomic feature selection method that allows for accurate prediction of survival-associated subgroups of IDH-wildtype glioblastoma, and showcase the potential utility of multitask learning and unlabeled MRI data on adult-type diffuse glioma classification and survival prediction tasks. From genomic data, we assemble a large SCNA dataset and train models to predict the mutational status of genes IDH1 and IDH2 mutation and the codeletion status of chromosome arms 1p and 19q in order to assign molecular diagnoses to patients with antiquated SCNA data and unknown diagnoses. We also nominate an oncogene that may characterize an aggressive subgroup of IDH-mutant and 1p/19q-codeleted oligodendroglioma associated with, but independently prognostic of, patient age. Altogether, our work in risk-stratifying adult-type diffuse glioma has implications for upfront clinical trial recommendations for patients diagnosed with IDH-wildtype glioblastoma, treatment toxicity calibration for risk-assessed oligodendroglioma, and clinical trial arm sorting, which may accelerate the approval of new effective therapies.