基于影像组学的AI预测非小细胞肺癌新辅助免疫化疗病理反应：系统综述与荟萃分析
Radiomics-Based AI for Predicting Neoadjuvant Immunochemotherapy Pathological Response in Non-Small Cell Lung Cancer: Systematic Review and Meta-Analysis.
作者
作者单位
- Department of Thoracic Surgery, Peking Union Medical College Hospital, Chinese Academy of Medical Science, No.1 Shuaifuyuan, Wangfujing, Beijing, Dongcheng District, 100730, China, 86 13621021237.
- Department of Infectious Diseases, Nanfang Hospital, Southern Medical University, Guangzhou, China.
摘要
中文
非小细胞肺癌（NSCLC）仍然是全球癌症相关死亡的主要原因。准确早期预测对新辅助治疗的反应至关重要。我们旨在评估基于影像组学的AI在预测NSCLC新辅助免疫化疗后病理完全缓解（pCR）和主要病理反应（MPR）方面的诊断性能，并将其与传统放射学标准进行比较。对PubMed、Embase、Cochrane Library和Web of Science进行了系统检索，截止日期为2025年10月26日。纳入了使用计算机断层扫描（CT）或正电子发射断层扫描/CT基于AI模型预测pCR或MPR的研究。使用PROBAST（预测模型风险偏倚评估工具）+AI工具评估方法学质量。使用双变量随机效应模型汇总敏感性、特异性和曲线下面积（AUC）。共纳入23项研究，验证集中有2004名患者，以组织病理学为金标准。对于pCR，AI模型实现的汇总敏感性为0.77（95% CI 0.70-0.83），特异性为0.79（95% CI 0.73-0.84），AUC为0.85，在敏感性上显著优于传统标准（0.77 vs 0.42；P<.001）。对于MPR，AI的敏感性为0.80（95% CI 0.72-0.87），特异性为0.83（95% CI 0.73-0.90），AUC为0.88，也优于传统模型（AUC：0.88 vs 0.65，P<.001）。亚组分析显示，基于正电子发射断层扫描/CT的模型对MPR的特异性高于基于CT的模型（0.95 vs 0.80，P=.005）。基于影像组学的AI显示出高诊断准确性，且敏感性优于传统放射学标准，在术前反应评估方面显示出显著潜力。然而，当前研究的异质性和回顾性设计限制了证据。未来需要大规模、前瞻性、多中心试验和多模态数据整合来验证这些发现以进行临床转化。
English
Non-small cell lung cancer (NSCLC) remains the leading cause of cancer-related mortality worldwide. Accurate early prediction of response to neoadjuvant therapy is critical. We aimed to evaluate the diagnostic performance of radiomics-based AI in predicting pathological complete response (pCR) and major pathological response (MPR) following neoadjuvant immunochemotherapy in NSCLC and to compare it against traditional radiological criteria. A systematic search of PubMed, Embase, Cochrane Library, and Web of Science was conducted through October 26, 2025. Studies utilizing computed tomography (CT) or positron emission tomography/CT-based AI models to predict pCR or MPR were included. Methodological quality was appraised using the PROBAST (Prediction Model Risk of Bias Assessment Tool)+AI tool. Sensitivity, specificity, and area under the curve (AUC) were pooled using a bivariate random effects model. Twenty-three studies involving 2004 patients in validation sets were included, with histopathology as the gold standard. For pCR, AI models achieved a pooled sensitivity of 0.77 (95% CI 0.70-0.83), specificity of 0.79 (95% CI 0.73-0.84), and AUC of 0.85, significantly outperforming traditional criteria in sensitivity (0.77 vs 0.42; P<.001). For MPR, AI demonstrated a sensitivity of 0.80 (95% CI 0.72-0.87), specificity of 0.83 (95% CI 0.73-0.90), and AUC of 0.88, also superior to traditional models (AUC: 0.88 vs 0.65, P<.001). Subgroup analysis revealed that positron emission tomography/CT-based models offered higher specificity for MPR than CT-based models (0.95 vs 0.80, P=.005). Radiomics-based AI demonstrates high diagnostic accuracy and superior sensitivity compared to traditional radiological criteria, showing significant potential for preoperative response assessment. However, the heterogeneity and retrospective design of current studies limit the evidence. Future large-scale, prospective, multicenter trials and multimodal data integration are required to validate these findings for clinical translation.
分类与指标
- 研究类型
- 综述Meta
- 病种
- 肺癌
- JCR 分区
- Q1
- 影响因子
- 8.2
- 新锐分区
- 1区