Purpose:This systematic review aims to analyze the literature on the application of AI in predicting patient outcomes and treatment-related toxicity in those undergoing SBRT or SRS across heterogeneous tumor sites.Materials and methods:Our review conformed to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses. PubMed, EMBASE and Scopus were systematically searched for English-language human studies evaluating AI for outcome and toxicity prediction in patients undergoing SBRT or SRS for solid tumors. Search terms included (“Stereotactic Body Radiotherapy” OR “SBRT” OR “Stereotactic Radiosurgery” OR “SRS” OR “Stereotactic Ablative Radiotherapy” OR “SABR”) AND (“Artificial Intelligence” OR “AI” OR “Machine Learning” OR “Deep Learning” OR “Radiomics”) AND (“Response Prediction” OR “Response to Treatment” OR “Outcome Prediction”) AND (“Toxicity” OR “Side Effects” OR “Treatment Toxicities” OR “Adverse Events”).Results:The search yielded 29 eligible retrospective studies, published between 2020 and 2025. Eight studies addressed early-stage primary lung cancer, highlighting the potential of AI-based models in predicting radiation-induced pneumonitis, fibrosis and local control. Five studies investigated AI models for predicting hepatobiliary toxicity following SBRT for liver tumors. Sixteen studies involved SRS-treated patients with brain metastases or benign intracranial neoplasms (e.g., arteriovenous malformations, vestibular schwannomas, meningiomas), exploring AI algorithms for predicting treatment response and radiation-induced changes. In the results, AI might have been exploited to both reaffirm already known clinical predictors and to identify novel imaging, dosimetric or biological biomarkers. Examples include predicting radiation pneumonitis in lung cancer, residual liver function in hepatic tumors and local recurrence in brain metastases, thus supporting tailored treatment decisions.Conclusions:Combining AI with SBRT could greatly enhance personalized cancer care by predicting patient-specific outcomes and toxicity. AI models analyze complex datasets, including imaging and clinical data, to identify patterns that traditional methods may miss, thus enabling more accurate risk stratification and reducing variability in treatment planning. With further research and clinical validation, this integration could make radiotherapy safer, more effective and contribute to advancement in precision oncology.
目的：本系统性综述旨在分析关于人工智能在预测接受立体定向体部放疗或立体定向放射外科治疗的不同部位肿瘤患者的预后及治疗相关毒性方面的应用文献。 材料与方法：本综述遵循系统综述和荟萃分析优先报告条目规范。系统检索了PubMed、EMBASE和Scopus数据库中评估人工智能在实体瘤患者接受SBRT或SRS治疗后预后及毒性预测的英文人类研究。检索词包括（“立体定向体部放疗”或“SBRT”或“立体定向放射外科”或“SRS”或“立体定向消融放疗”或“SABR”）与（“人工智能”或“AI”或“机器学习”或“深度学习”或“影像组学”）及（“疗效预测”或“治疗反应”或“预后预测”）与（“毒性”或“副作用”或“治疗毒性”或“不良事件”）的组合。 结果：检索共纳入29项符合标准的回顾性研究，发表时间为2020年至2025年。其中8项研究针对早期原发性肺癌，重点探讨了基于人工智能的模型在预测放射性肺炎、肺纤维化和局部控制方面的潜力。5项研究探索了用于预测肝脏肿瘤SBRT治疗后肝胆毒性的AI模型。16项研究涉及接受SRS治疗的脑转移瘤或良性颅内肿瘤（如动静脉畸形、前庭神经鞘瘤、脑膜瘤）患者，主要探索AI算法在预测治疗反应和放射性改变方面的应用。结果显示，AI技术既可用于验证已知的临床预测因子，也能识别新的影像学、剂量学或生物学标志物。例如预测肺癌的放射性肺炎、肝肿瘤的残余肝功能以及脑转移瘤的局部复发，从而支持个体化治疗决策。 结论：将人工智能与SBRT技术相结合，通过预测患者特异性预后和毒性，可显著提升肿瘤个体化治疗水平。AI模型通过分析包括影像学和临床数据在内的复杂数据集，能够识别传统方法可能忽略的模式，从而实现更精准的风险分层并减少治疗计划的差异性。随着进一步研究和临床验证，这种融合有望使放射治疗更安全、更有效，并推动精准肿瘤学的发展。
肿瘤(癌症)患者之家