Digital Twins in Neuro-Oncology: A Systematic Review of Current Implementations, Technical Strategies, and Clinical Applications

Purpose: To perform a systematic review evaluating current digital twin (DT) implementations, highlighting clinical relevance and technical strategies, identifying opportunities to advance personalized, predictive care in neuro-oncology. Materials and Methods: PubMed, Scopus, and Web of Science databases were systematically screened for English-language original research articles publisfrom inception through June 2025 focused on DT development, validation, or patient-specific computational models in neuro-oncology. Extracted variables included computational frameworks, data sources, clinical or predictive tasks, and reported outcomes. Risk of bias and applicability were assessusing the Prediction model Risk Of Bias ASsessment Tool (PROBAST), which revealed well-defined predictors and outcomes but frequent concerns regarding participants and analysis. Results: Of the 73 articles reviewed, 21 met eligibility criteria. DTs simulated tumor growth, radiation response, immune interactions, and drug transpMost models (n = 20) relied on mechanistic or biophysical frameworks, with increasing adoption of artificial intelligence–driven and hybrid approachestotal of 12 studies focused on glioblastomas or high-grade gliomas, and 17 relied primarily on MRI data. Tumor-growth and treatment-response simulations were the most common DT applications. Only six studies provided publicly available code, and closed-loop calibration was reported in eight studPredictive accuracy and correlation with clinical data were generally high, but real-time integration, multimodal data fusion, and external validation werlimited. Conclusion: DTs showed promise for advancing personalized neuro-oncology, with demonstrated potential in modeling tumor behavior and optimizing therapies. Applications relied mainly on mechanistic artificial intelligence methods. Despite strong predictive performance, reproducibility, multimodal integration, and external validation remained limited, reflecting method heterogeneity.