Artificial intelligence (AI) uses in stereotactic radiosurgery (SRS): outcome prediction with brain metastasis (BM) - A systematic review.
Authors
Affiliations (4)
Affiliations (4)
- Department of Neurosurgery, Stanford University School of Medicine, Stanford, CA, USA.
- Department of Radiation Oncology, Stanford University School of Medicine, Stanford, CA, USA.
- Department of Neurosurgery, Stanford University School of Medicine, Stanford, CA, USA. Electronic address: [email protected].
- Department of Neurosurgery, Stanford University School of Medicine, Stanford, CA, USA. Electronic address: [email protected].
Abstract
Brain metastases (BM) are the most common intracranial tumors in adults and present ongoing challenges in clinical management, particularly after stereotactic radiosurgery (SRS). Recent advances in artificial intelligence (AI) offer the potential to improve outcome prediction and personalize treatment. This systematic review synthesizes current literature on AI applications for prognostic modeling in BM patients undergoing SRS, with a focus on imaging-based machine learning (ML) and deep learning (DL) tools for predicting local control, survival, and treatment-related toxicity. A systematic review was performed in accordance with PRISMA guidelines. PubMed, Web of Science, and Scopus were searched in October 2024 using a targeted query combining terms related to AI, brain metastasis, and outcome prediction. After screening 255 studies and applying strict inclusion/exclusion criteria, 21 studies published between 2018 and 2024 were included. Data was extracted on study design, modeling techniques, input features, validation methods, and predictive performance. Among the 21 included studies, AI models demonstrated strong performance in predicting key outcomes following SRS. Convolutional neural networks (CNNs), recurrent neural networks, support vector machines (SVMs), and ensemble methods were widely used. Reported AUCs ranged from 0.70 to 0.98, with highest performance achieved through multimodal data integration of MRI-derived radiomic features and clinical variables. Predictive features included peritumoral edema texture, tumor margin heterogeneity, and clinical indicators such as lesion number, tumor size, and extracranial disease progression. Longitudinal modeling using serial imaging (e.g., Conv-GRU) enhanced dynamic outcome prediction. Several studies demonstrated generalizability across institutions, supporting the robustness of these models. AI models, particularly those incorporating radiomics and clinical data, show high accuracy in predicting outcomes for BM patients treated with SRS. These tools offer substantial promise for risk stratification, early identification of treatment failure, and personalized care planning. Further multicenter validation and integration into clinical workflows are warranted to fully realize the benefits of AI in SRS for brain metastases.