A Systematic Review and Meta-analysis of Survival Prediction in Glioblastoma Patients using Advanced MRI Techniques

Glioblastoma (GBM) is an aggressive brain tumor with a dismal prognosis. Recent advances in radiomics and machine learning (ML) applied to magnetic resonance imaging (MRI) have demonstrated promising potential in enhancing clinical decision-making and prognostic accuracy. This systematic review and meta-analysis aimed to evaluate the predictive performance of radiomics and ML techniques applied to pre-treatment MRI data in glioblastoma prognosis.

A comprehensive literature search was conducted across MEDLINE, EMBASE, and the Cochrane Central Register of Controlled Trials up to March 2024 for studies using radiomics or ML techniques applied to pre-treatment MRI scans to predict progression-free survival (PFS) and overall survival (OS) in glioblastoma patients. The primary outcome was the area under the receiver operating characteristic curve (AUC). Study quality was assessed using the QUADAS-2 tool, meta-analysis employed a random-effects model, and heterogeneity was evaluated using the I2 statistic.

Sixteen studies comprising a total of 2,342 patients were included. MRI-based machine learning models demonstrated high predictive performance for glioblastoma prognosis (AUC: 0.71-0.92), with a tendency to outperform radiomics-based approaches (AUC: 0.68-0.88). A meta-analysis of 12 studies yielded a pooled AUC of 0.78 (95% CI: 0.74-0.82; p < 0.001) for PFS prediction with moderate heterogeneity (I2 = 59%). Four studies focused on OS prediction, showing no heterogeneity (I2 = 0%) and a pooled AUC of 0.81 (95% CI: 0.77-0.85; p < 0.001). Subgroup analysis revealed that ML models (AUC: 0.83 [95% CI: 0.78-0.87]) statistically outperformed radiomics-based models (AUC: 0.76 [95% CI: 0.71-0.80]) for PFS prediction (p = 0.02).

Advanced quantitative MRI techniques, particularly radiomics, demonstrate superior performance over conventional MRI in glioblastoma characterization, improving tumor segmentation, subtype differentiation, and prognostication. By capturing intratumoral heterogeneity, radiomic features achieve higher diagnostic accuracy. The use of the DICOM format enhances model performance through standardized data integration. Radiomics shows promise as a non-invasive biomarker for guiding personalized treatment; however, challenges such as observer variability, methodological heterogeneity, and lack of standardization remain. Future integration with molecular data and validation through large prospective studies are essential for clinical implementation.

Radiomics and ML approaches based on pre-treatment MRI are promising tools for predicting survival outcomes in glioblastoma patients, with ML models demonstrating a slight edge over radiomics for PFS prediction. Standardized protocols and larger multi-center studies are warranted to facilitate clinical adoption.