基于18F-FDG PET/CT的瘤内栖息地异质性预测高级别浆液性卵巢癌新辅助化疗早期反应
研究纳入258例接受[18F]FDG PET/CT后行新辅助化疗的高级别浆液性卵巢癌患者,按RECIST 1.1分为反应组与非反应组,利用nnU-Net自动分割原发肿瘤并以K-means、DBSCAN和Otsu中的最佳方法划分栖息地亚区,提取影像组学特征构建亚区模型、传统模型和联合模型。亚区模型在测试集的预测效能显著优于传统模型(AUC 0.83 vs 0.71,P=0.007),联合模型与亚区模型无统计学显著差异(AUC 0.89 vs 0.83,P=0.077)。
要点
- 基于栖息地分析的PET/CT影像组学模型可有效预测HGSOC患者NACT早期反应。
- 亚区模型测试集AUC 0.83,显著优于传统模型AUC 0.71(P=0.007)。
- 建模影像组学特征与NACT前Ki67及NACT前至IDS后Ki67变化值显著相关(P<0.05)。
核心结论
- 基于栖息地分析的PET/CT影像组学模型可有效预测HGSOC患者NACT早期反应,其可解释性来自肿瘤增殖活性,并为个体化治疗决策提供可靠影像生物标志物。
研究设计与样本
- 共纳入258例接受[18F]FDG PET/CT后行NACT的HGSOC患者,按RECIST 1.1分为反应组和非反应组。
- 原发肿瘤病灶由nnU-Net自动分割,肿瘤栖息地亚区采用K-means、DBSCAN和Otsu中的最佳方法分割。
- 基于影像组学提取异质性特征并经递归特征消除筛选,构建亚区模型、传统模型和联合模型。
- 模型性能采用受试者工作特征曲线和DeLong检验评估,Pearson相关分析用于探索影像组学特征与病理参数的关系。
主要发现
- 自动分割模型Dice系数为0.82,在验证队列中表现优异。
- 测试数据集中亚区模型预测效能显著优于传统模型(AUC 0.83 vs 0.71,P=0.007)。
- 联合模型与亚区模型之间无统计学显著差异(AUC 0.89 vs 0.83,P=0.077)。
- 参与建模的影像组学特征分别与NACT前Ki67(pre-Ki67)及NACT前至IDS后Ki67变化值(ΔKi67)显著相关(P<0.05)。
关键数值与推荐
- 反应分组标准为RECIST 1.1。
- 亚区模型测试集AUC 0.83,传统模型AUC 0.71,两者比较P=0.007。
- 联合模型测试集AUC 0.89,与亚区模型比较P=0.077。
- 自动分割模型Dice系数0.82。
临床怎么用
- 原文结论认为基于栖息地分析的PET/CT影像组学模型可有效预测HGSOC患者NACT早期反应,并可提供影像生物标志物用于个体化治疗决策。
局限性
- 原文未报告局限性。
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)。
原文
OBJECTIVE: To explore the predictive value of intratumoral habitat heterogeneity for the early therapeutic response to neoadjuvant chemotherapy (NACT) in patients with high-grade serous ovarian cancer (HGSOC). MATERIALS AND METHODS: A total of 258 patients with HGSOC receiving [18F]fluorodeoxyglucose ([18F]FDG) PET/CT followed by NACT were enrolled and classified into a response group and a non-response group according to RECIST 1.1. The primary tumor lesion was automatically segmented by the nnU-Net. Habitat subregions of the tumor were segmented by the best method in K-means, DBSCAN, and Otsu. Heterogeneity features were extracted based on radiomics and selected by recursive feature elimination to develop the subregion model, the traditional model, and the union model. The performance of the models was evaluated by the receiver operating characteristic curve and DeLong test, and Pearson correlation analysis was used to explore the relationships between radiomics features and pathological parameters. RESULTS: The automatic segmentation model had a Dice coefficient of 0.82 and showed excellent performance in the validation cohort. The predictive efficacy of the subregion model in the test dataset was significantly better than that of the traditional model [Area Under the Curve (AUC): 0.83 vs. 0.71, P = 0.007. There was no statistically significant difference between the union model and the subregion model (AUC: 0.89 vs. 0.83, P = 0.077). Radiomics features involved in the modeling were significantly correlated with Ki67 before NACT (pre-Ki67) and the change value of Ki67 before NACT-after IDS(ΔKi67) respectively (P < 0.05). CONCLUSION: The PET/CT radiomics model based on habitat analysis can effectively predict the early response to NACT in patients with HGSOC, which can be explained by the tumor proliferative activity and provide reliable imaging biomarkers for individualized treatment decisions.
引用
- Liu Z, Mouni D, Zhang S, et al. Predicting the early response to neoadjuvant chemotherapy in high-grade serous ovarian cancer by intratumoral habitat heterogeneity based on 18F-FDG PET/CT. Eur J Nucl Med Mol Imaging. 2025. PMID: 40762796. doi:10.1007/s00259-025-07480-z
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。