Survival Rate Prediction in glioblastoma Patients Using Machine learning

Glioblastoma, an aggressive form of brain cancer, continues to pose significant challenges, with median survival rates ranging from 12 to 18 months despite ongoing advances in treatment. Traditional survival prediction models predominantly rely on non-omic data, such as MRI, PET, and CT scans, which often lack the granularity to uncover molecular biomarkers crucial for guiding personalized therapeutic strategies. In this study, we introduce a novel methodology focused solely on omic data for survival prediction in glioblastoma patients. Our approach integrates genomic features, including G-CIMP methylation, gene expression subtypes, and IDH1 mutation. Utilizing a robust dataset comprising 577 patient records and 22 features, we implement a deep learning model based on Transformer architecture. The model incorporates positional encoding to capture complex temporal relationships in survival data and leverages the Cox proportional hazards framework for survival analysis. Our results demonstrate a high concordance index (C-index) of 87% and an integrated Brier score (IBS) of 0.05 that further validates the model’s predictive accuracy. We highlight the influence of critical genomic features on survival predictions. This approach represents a significant advancement in leveraging omic data and modern machine learning techniques to enhance the accuracy and reliability of glioblastoma prognosis, offering promising implications for personalized treatment strategies and improved patient outcomes.