Artificial Intelligence and Radiomics in Glioblastoma Surgery: Current Applications, Clinical Challenges, and Future Directions - A Narrative Review

Abstract

Glioblastoma (GBM) remains the most aggressive primary malignant brain tumor in adults despite advances in neurosurgical techniques, radiotherapy, chemotherapy, and molecular diagnostics. The marked heterogeneity of GBM, coupled with its infiltrative growth pattern and poor prognosis, continues to pose substantial challenges in diagnosis, surgical planning, treatment selection, and postoperative surveillance. In recent years, artificial intelligence (AI) and radiomics have emerged as promising technologies capable of extracting quantitative information from medical imaging that extends beyond conventional visual interpretation. Machine learning and deep learning algorithms have demonstrated considerable potential in tumor segmentation, molecular characterization, surgical planning, intraoperative guidance, prognostic modeling, and prediction of therapeutic response. This narrative review summarizes current evidence regarding the integration of AI and radiomics into the management of glioblastoma from a neurosurgical perspective. The review discusses recent developments in preoperative imaging analysis, automated tumor segmentation, prediction of molecular biomarkers, intraoperative applications, survival prediction, and postoperative monitoring. Current limitations including data heterogeneity, limited external validation, algorithm interpretability, ethical concerns, and regulatory challenges are also examined. Finally, future directions involving multimodal data integration, explainable AI, federated learning, and prospective clinical validation are highlighted. Although AI has not yet replaced clinical judgment, growing evidence suggests that AI-assisted decision-making may substantially improve diagnostic accuracy, surgical precision, personalized treatment planning, and clinical outcomes in patients with glioblastoma. Continued multidisciplinary collaboration between neurosurgeons, radiologists, computer scientists, and data engineers will be essential for translating these technologies into routine clinical practice.