基于CT影像组学预测HR+/HER2-晚期乳腺癌CDK4/6抑制剂联合内分泌治疗疗效
本研究回顾性纳入102例接受一线CDK4/6抑制剂联合内分泌治疗的HR+/HER2-晚期乳腺癌患者,从基线增强CT多转移灶图像中提取影像组学特征,分别采用Cox回归和XGBoost构建PFS预测及疗效预测模型。联合影像组学-临床模型在训练集和测试集PFS预测C指数分别为0.85和0.76,显著优于单一模型。
要点
- 基线增强CT多转移灶影像组学特征联合临床模型可初步预测一线CDK4/6抑制剂联合内分泌治疗的PFS。
- 联合模型测试集6、12、18个月时间依赖性AUC分别达0.938、0.785、0.836。
- 该模型对最佳总体反应预测AUC为0.80,早期疗效反应预测AUC为0.70。
核心结论
- 在该探索性队列中,影像组学模型在预测HR+/HER2-晚期乳腺癌CDK4/6抑制剂联合内分泌治疗疗效方面展现出初步潜力。
- 联合影像组学-临床模型预测PFS的C指数在训练集为0.85(95%CI 0.73-0.98),测试集为0.76(95%CI 0.54-0.99)。
- 联合模型用于PFS预测的表现显著优于单纯临床模型和单纯影像组学模型(均P<0.05)。
研究设计与样本
- 回顾性研究,筛选481例晚期HR+/HER2-乳腺癌患者,最终纳入102例一线治疗患者。
- 患者随机分为训练集82例与测试集20例,测试集样本量仅20例,结论仅为探索性。
- 影像组学特征提取自基线增强CT图像,覆盖多个转移灶。
- 特征筛选采用LASSO与递归特征消除,PFS预测采用Cox比例风险模型,疗效反应预测采用XGBoost。
- 模型性能评估指标为一致性指数、AUC及决策曲线分析。
关键影像表现/征象
- 影像组学特征来源于基线增强CT上多个转移灶的图像,原文未描述具体可读的肉眼影像征象。
诊断标准与关键数值
- PFS预测联合模型训练集C指数为0.85(95%CI 0.73-0.98),测试集C指数为0.76(95%CI 0.54-0.99)。
- 联合模型测试集6、12、18个月时间依赖性AUC分别为0.938、0.785、0.836;训练集分别为0.895、0.897、0.960。
- 最佳总体反应模型AUC为0.80,早期治疗反应模型AUC为0.70。
- 联合模型PFS预测性能显著优于单纯临床模型和单纯影像组学模型(均P<0.05)。
- 原文未提供模型敏感度、特异度、影像组学评分的具体截断值或风险分层阈值。
临床应用与阅片要点
- 对于拟行一线CDK4/6抑制剂联合内分泌治疗的HR+/HER2-晚期乳腺癌,基线增强CT影像组学或可辅助预估PFS获益。
- 阅片时应注意该模型基于多转移灶同步提取特征,单一病灶测量和肉眼征象不足以替代影像组学评分。
- 该影像组学模型为研究性工具,测试集仅20例,尚不能直接指导个体化治疗决策。
局限性
- 回顾性单中心设计,测试集仅20例,样本量小,统计效能有限。
- 原文明确结论为探索性,需在更大规模、多中心、独立队列中验证后方可考虑临床应用。
- 未提供模型的外部验证结果,泛化能力尚不确定。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
INTRODUCTION: The combination of CDK4/6 inhibitors and endocrine therapy is the first-line standard regimen for advanced HR⁺/HER₂⁻ breast cancer. However, due to tumor heterogeneity, some patients derive limited benefit. This study aimed to develop and validate a radiomics-based model to predict treatment efficacy.
MATERIALS AND METHODS: This retrospective study screened 481 patients with advanced HR+/HER2- breast cancer. After applying inclusion and exclusion criteria, 102 first-line patients were enrolled and randomly split into training (n = 82) and testing (n = 20) sets. Radiomic features were extracted from baseline contrast-enhanced CT images of multiple metastatic sites. LASSO and recursive feature elimination were used for feature selection. Cox proportional-hazards and XGBoost algorithms were employed to develop models for predicting progression-free survival (PFS) and treatment response, respectively. Model performance was evaluated using the concordance index (C-index), area under the curve (AUC), and decision curve analysis (DCA).
RESULTS: The combined radiomics-clinical model achieved C-index values of 0.85 (95% CI: 0.73-0.98) and 0.76 (95% CI: 0.54-0.99) in the training and testing sets, respectively, for PFS prediction, significantly outperforming both the clinical-only and radiomics-only models (all P < 0.05). In the test set, time-dependent AUC values at 6, 12, and 18 months were 0.938, 0.785, and 0.836, respectively (training set: 0.895, 0.897, and 0.960). For response prediction, the best overall response model achieved an AUC of 0.80, and the early treatment response model achieved an AUC of 0.70.
CONCLUSION: In this exploratory cohort, radiomics models show preliminary promise in predicting the efficacy of CDK4/6 inhibitor plus endocrine therapy in advanced breast cancer. However, these findings require validation in larger, multicenter, independent cohorts before clinical implementation.
引用
- Wei, Lin, Cao, et al. Radiomics Model Predicts Efficacy of First-Line CDK4/6 Inhibitor Combined with Endocrine Therapy in HR+/HER2- Advanced Breast Cancer. Acad Radiol. 2026. PMID: 42557206. doi:10.1016/j.acra.2026.07.041
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。