Artificial Intelligence for the Preoperative Molecular Characterization of Adult-Type Diffuse High-Grade Gliomas: A Systematic Review and Meta-Analysis.
Authors
Affiliations (8)
Affiliations (8)
- School of Public Health and Preventive Medicine, Monash University, Melbourne, VIC, Australia. [email protected].
- Department of Neurosurgery, College of Medicine, University of Florida, Gainesville, FL, USA.
- Department of Neurosurgery, Johns Hopkins School of Medicine, Baltimore, MD, USA.
- College of Medicine, Xi'an Jiaotong University, Xi'an, China.
- Department of Neurosurgery, Rutgers New Jersey Medical School, Rutgers University, New Brunswick, NJ, USA.
- Department of Neurosurgery, Robert Wood Johnson Medical School, Robert Wood Johnson University Hospital, Rutgers University, New Brunswick, NJ, USA.
- Anatomy Department, Universidad de los Andes School of Medicine, Bogotá, Colombia.
- College Department of Radiology, Hospital of the University of Pennsylvania, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA.
Abstract
Molecular markers are central to diagnosis, prognosis, and treatment selection in adult-type diffuse high-grade gliomas (HGG), but tissue genotyping is invasive and may be affected by intratumoral heterogeneity. We evaluated the diagnostic accuracy of AI-based MRI radiogenomics for preoperative molecular characterization of HGG. Following PRISMA 2020, we searched PubMed, EMBASE, Scopus, and bioRxiv on March 14, 2026, for studies using AI on preoperative MRI to predict molecular markers in HGG. Risk of bias was assessed using QUADAS-2; reporting and methodological quality were assessed with CLEAR and RQS. Diagnostic accuracy was pooled using bivariate random-effects models, with heterogeneity, prediction intervals, Deeks' test, and meta-regression evaluated. Nineteen studies were included (18 contributing 22 model estimates). Diagnostic performance was highest for ATRX loss (pooled sensitivity 0.932, specificity 0.915, and DOR 146.33) and IDH mutation (sensitivity 0.851, specificity 0.836, and DOR 35.43). MGMT promoter methylation demonstrated moderate performance (sensitivity 0.758, specificity 0.750, and DOR 10.20) with high heterogeneity (I<sup>2</sup> = 82%). EGFR and PTEN showed limited utility (DORs 4.54 and 6.36). Performance fell markedly when restricted to externally validated cohorts (MGMT DOR 4.35 vs. 10.20 overall). Risk of bias was high in the index test domain for 79% of studies, with no significant publication bias (Deeks' P = 0.109). AI-based MRI radiogenomics is a promising preoperative adjunct, particularly for IDH typing, but standardized, transparent, externally validated studies are needed. Pooled estimates dominated by internal validation likely overestimate real-world performance.