Utilizing Radiomics as Predictive Factor in Brain Metastasis Treated With Stereotactic Radiosurgery: Systematic Review and Radiomic Quality Assessment

Abstract

Radiomics and machine learning (ML) are increasingly utilized to predict treatment response by uncovering latent information in medical images. This study systematically reviews radiomics studies on brain metastasis treated with stereotactic radiosurgery (SRS) and quantifies their radiomic quality score (RQS). A systematic search on Scopus, Web of Science, and PubMed was conducted to identify original studies on radiomics for predicting treatment response, adhering to predefined patient, intervention, comparator, and outcome (PICO) criteria. No restrictions were placed on language or publication date. Two independent reviewers assessed eligible studies, and the RQS was calculated based on Lambin’s guidelines. The Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) 2020 guidelines were followed. Seventeen studies involving 2744 patients met the inclusion criteria out of 200 identified. All studies were retrospective and utilizing various MRI scanners models with different field strength. The average RQS across studies was low (39.2%), with a maximum score of 19 points (52.7%). Radiomic-based models demonstrated superior predictive accuracy compared to clinical or visual assessment, with AUC values ranging from 0.74 to 0.92. Integration of clinical features such as Karnofsky performance status, dose, and isodose line further improved model performance. Deep learning models achieved the highest predictive accuracy, with AUC of 0.92. Radiomics demonstrate significant potential in predicting treatment outcomes with high accuracy, offering opportunities to advance personalized management for BM. To facilitate clinical adoption, future studies must prioritize adherence to standardized guidelines and robust model validation to ensure reproducibility.