A systematic review and meta-analysis of deep learning and radiomics in predicting MGMT promoter methylation status in glioblastoma: Efficacy, reliability, and clinical implications

• Published in: Displays Vol. 89; p. 103072
• Main Authors: Chen, Yu, Liao, Yuehui, Li, Panfei, Jin, Wei, Fang, Jingwan, Huang, Junwei, Feng, Yaning, Xie, Changxiong, Li, Ruipeng, Jin, Qun, Lai, Xiaobo
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2025
• Subjects: Deep learning; Glioblastoma (GBM); Meta-analysis; MGMT promoter methylation; Radiomics; Deep learning; MGMT promoter methylation; Glioblastoma (GBM); Radiomics; Meta-analysis
• ISSN: 0141-9382
• DOI: 10.1016/j.displa.2025.103072

•Non-Invasive Prediction: Reviews MRI-based radiomics and deep learning models for non-invasive MGMT methylation prediction in GBM.•Meta-Analysis: A pooled AUC of 0.86 demonstrates strong performance, but heterogeneity exists due to differences in protocols and models.•External Validation: A lower mean AUC (0.69) in external validation emphasizes the need for diverse datasets.•Standardization Challenges: Variability in MRI protocols and feature extraction methods limits comparability and clinical applicability.•Future Directions: Calls for multicenter studies, data integration, and explainable AI to enhance reliability and trust. O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation is a critical predictive biomarker for assessing temozolomide response in glioblastoma (GBM). Deep learning (DL) and radiomics offer promising non-invasive alternatives for evaluating MGMT promoter methylation status. To evaluate the diagnostic performance and methodological rigor of published deep learning and radiomic models for predicting MGMT promoter methylation. A comprehensive literature search was conducted across PubMed, Ovid Embase, EBSCOhost Cumulative Index to Nursing and Allied Health Literature (EBSCO CINAHL), Web of Science, IEEE Xplore, and ACM Digital Library databases through December 31, 2024. Studies using magnetic resonance imaging (MRI)-based radiomic features and DL algorithms to classify MGMT promoter methylation status in GBM patients were included. The review protocol was registered with PROSPERO (CRD42021279221), and study selection adhered to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Diagnostic Test Accuracy Studies (PRISMA-DTA) guidelines. Methodological quality was assessed using the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis + Artificial Intelligence (TRIPOD + AI) checklist and the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. A meta-analysis of diagnostic performance was performed using Stata v.17.1. The pooled area under the curve (AUC) was 0.86 (95 % CI: 0.83–0.89), reflecting strong diagnostic performance. However, external validation studies revealed a significantly lower mean AUC of 0.69, indicating potential overfitting. High heterogeneity (I2 > 90 %) was attributed to variations in imaging protocols, feature extraction techniques, and data sources. While radiomics and DL-based models show potential for non-invasive MGMT promoter methylation prediction, their clinical applicability is hindered by a lack of standardized datasets and robust external validation. Future studies should focus on addressing these limitations to enhance reliability and generalizability.