Brain Cancer Survival Glioblastoma Patient Based on Graph Analysis Network

Glioblastoma (GBM) remains a highly aggressive brain tumour with limited treatment options and poor patient outcomes. In this paper, we present a novel model called RAGA-Net for predicting overall survival (OS) of patients with IDH-wildtype GBM which systematically integrates radiomics, attention mechanisms and graph-based analysis. RAGA-Net employs the use of multi-parametric MRI scans (T1, T2, FLAIR, T1GD) and extracts radiomic features with graph features aiding the understanding of GBM heterogeneity. It also uses clinical data to create spatial relationships within tumours but relies on attention mechanisms to further the understanding of relevant aspects of the study. This way, a model is built with high accuracy, surpassing prior methods, and supporting personalized treatment strategies for GBM.

To support early interventions, this study examines the predictive power of regional radiomics similarity networks (R2SNs) in preoperative GBM survival predictions using routine MRI. Unlike standard MRI-based radiomics that often isolate specific regions, R2SNs account for inter-regional relationships via radiomics similarity, enhancing survival predictions. Specifically, a novel distance correlation-based R2SN (DC-R2SN) is introduced, leveraging distance correlation for complex interaction measurement among brain regions. A graph neural network (GNN) framework integrates DC-R2SNs with clinical data for OS prediction. Testing on the UPenn-GBM dataset highlights the GNN framework’s effectiveness, underscoring its utility in clinical decision-making for GBM patients. This approach aims to empower healthcare providers with precise, tailored insights to improve patient care and outcomes.