Key points are not available for this paper at this time.
Glioblastoma multiforme (GBM) is one of the most aggressive primary glioma brain cancer in adults that has poor prognosis and few medications. Recent developments in Explainable Artificial Intelligence (XAI) have made it possible to develop clinical-interpretable predictive models with high accuracy to help with personalized prognosis and optimization of treatment. This systematic review and meta-analysis summarize the existing research studies on the XAI-based models of GBM prediction and management published between 2020-2025. The study used multiple database search to identify eligible studies that were further screened with pre-determined inclusion criteria. Important performance measures such as area under the curve (AUC) and accuracy were retrieved and interpreted descriptively. The findings show that well-performing machine learning models, including XGBoost, deep neural networks, and ensemble models, produce AUCs as high as 0.9845 and accuracies as high as 95.5%. The most common prognostic factors identified with the use of XAI techniques like SHapley additive exPlanations and local interpretable model-agnostic explanations were IDH1 mutation status, patient age, radiomic texture factors, and therapeutic regimens. In other instances, novel biomarkers, including one representing PPP2RIB identified by PathX-convolutional neural networks were discovered, showing the promise of XAI to reveal new clinical knowledge. Nevertheless, there are limitations to these advances, such as class imbalance, the need for multi-modal dataset combining radiomics and genomics, and lack of standardized integration protocols with clinical care.
Iyiade et al. (Fri,) studied this question.