Delta radiomics from multiparametric MRI predicts survival and treatment response in post-treatment glioma patients

Accurate prognostication in glioma and brain cancer patients remains a critical challenge, particularly in the posttreatment setting. Radiomics is a field of study that extracts high dimensional features from medical imaging, and it offers a non-invasive approach to characterize treatment response. In this study, we evaluated whether delta radiomic features, defined as the changes in imaging features between two time points, from magnetic resonance imaging (MRI) scans could predict overall survival in patients with diffuse glioma. We retrospectively analyzed 298 patients at a single institution who underwent two consecutive post-treatment MRI scans. Radiomic features were extracted from T1-weighted, contrast-enhanced T1, T2-weighted, and FLAIR sequences using the PyRadiomics library. Then, delta features were computed as the difference in radiomic values between the two scans for each patient. After filtering for complete cases and standardizing features, we trained a Cox proportional hazards model with L2 regularization using the lifelines Python package. Model performance was evaluated using the concordance index (C-index). The model achieved a C-index of 0.80, demonstrating that delta radiomics can be used to capture prognostically relevant imaging changes. We also used a Random Forrest Classifier to predict response to treatment, which received a mean cross-validation accuracy of 0.731. This study highlights the potential of multiparametric, longitudinal MRI-based radiomics as a noninvasive biomarker of survival in glioma and supports further integration of temporal imaging analysis into post-treatment risk stratification.