###
中国临床研究英文版:2026,39(9):1361-1366
本文二维码信息
码上扫一扫!
循环肿瘤DNA甲基化标志物联合多模态MRI瘤内与瘤周影像组学特征对乳腺癌腋窝淋巴结转移的术前预测
(秦皇岛市第一医院乳腺疾病诊疗中心,河北 秦皇岛 066000)
Integration of circulating tumor DNA methylation biomarkers with intratumoral and peritumoral multi-parametric MRI radiomics for preoperative prediction of axillary lymph node metastasis in breast cancer
(Breast Disease Diagnosis and Treatment Center,Qinhuangdao First Hospital,Qinhuangdao,Hebei 066000,China)
摘要
本文已被:浏览 166次   下载 211次
Received:April 13, 2026   Published Online:September 24, 2026
中文摘要: 目的  探究乳腺癌患者循环肿瘤DNA(ctDNA)甲基化标志物和乳腺肿瘤瘤内及瘤周多模态MRI 的影像组学特征,评估二者的融合策略在乳腺癌腋窝淋巴结转移(ALNM)术前预测中的临床价值。方法  选取2024年1 月至2025 年11 月秦皇岛市第一医院收治的201 例单侧单发乳腺癌患者,根据术后病理结果分为ALNM 组(n=54)和非ALNM 组(n=147)。术前检测患者外周血ctDNA 中SRY盒转录因子17(SOX17)、谷胱甘肽S-转移酶P1(GSTP1)基因甲基化水平,并基于多模态MRI 图像分割瘤内和瘤周感兴趣容积,提取影像组学特征。将这201例患者按7∶3比例随机分配至训练集(n=141)和验证集(n=60)。通过特征筛选,分别建立瘤内、瘤周影像组学模型及联合模型,并结合临床资料和ctDNA 甲基化数据开展多因素logistic 回归分析。绘制受试者工作特征(ROC)曲线,以曲线下面积(AUC)评估各模型的预测效能,并通过DeLong 检验比较各模型AUC 的差异。结果  多因素logistic 回归分析结果显示组织学分级Ⅲ级(OR=2.500,95%CI:1.388~4.500,P=0.002)、脉管浸润(OR=2.000,95%CI:1.225~3.264,P=0.006)、Ki-67高表达(OR=2.200,95%CI:1.270~3.806,P=0.005)、肿瘤最大径>2 cm(OR=2.000,95%CI:1.201~3.329,P=0.008)以及高水平ctDNA 中GSTP1(OR=1.050,95%CI:1.030~1.071,P<0.01)和SOX17(OR=1.040,95%CI:1.024~1.056,P<0.01)甲基化为ALNM 的独立危险因素。影像组学模型中,瘤内模型预测效能最优(训练集AUC=0.856,验证集AUC=0.830),瘤周模型以3 mm 区域表现最佳(训练集AUC=0.842,验证集AUC=0.795);进一步构建的多模态模型比较显示,ctDNA 甲基化标志物模型(训练集AUC=0.773,验证集AUC=0.744)效能有限,综合模型(整合ctDNA 甲基化与影像组学特征)在训练集和验证集中预测效能最佳(训练集AUC=0.924,验证集AUC=0.912)。结论  ctDNA 甲基化标志物联合多模态MRI 瘤内与瘤周影像组学特征在乳腺癌ALNM的预测中具有较高应用价值,综合模型兼具高灵敏度与特异度。
Abstract:Objective To investigate circulating tumor DNA(ctDNA)methylation biomarkers and intratumoral and peritumoral multi-parametric MRI radiomic features in breast cancer patients,and to evaluate the clinical value of their fusion strategies for preoperative prediction of axillary lymph node metastasis(ALNM). Methods A total of 201 patients with unilateral,single primary breast cancer admitted to Qinhuangdao First Hospital between January 2024 and November 2025 were enrolled and divided postoperatively into an ALNM group(n=54)and a non-ALNM group(n=147)based on pathological findings. Preoperatively,methylation levels of SRY-box transcription factor 17(SOX17)and glutathione S - transferase P1 (GSTP1)genes in peripheral blood ctDNA were measured. Intratumoral and peritumoral volumes of interest were segmented on multi-parametric MRI images,from which radiomic features were extracted. All patients were randomly assigned at a 7∶3 ratio to a training cohort(n=141)and a validation cohort(n=60). After feature selection,intratumoral,peritumoral,and combined radiomic models were established,and multivariate logistic regression analysis was performed integrating clinical data and ctDNA methylation profiles.Receiver operating characteristic(ROC)curves were plotted,predictive performance was assessed using area under the curve(AUC),and AUC differences between models were compared via the DeLong test. Results Multivariate logistic regression identified histological grade Ⅲ(OR=2.500,95%CI:1.388-4.500,P=0.002),lymphovascular invasion(OR=2.000,95%CI:1.225-3.264,P=0.006),high Ki-67 expression(OR=2.200,95%CI:1.270-3.806,P=0.005),maximum tumor diameter >2 cm(OR=2.000,95%CI:1.201-3.329,P=0.008),and elevated methylation levels of both GSTP1(OR=1.050,95%CI:1.030-1.071,P<0.01)and SOX17(OR=1.040,95%CI:1.024-1.056,P<0.01)in ctDNA as independent risk factors for ALNM. Among the radiomic models,the intratumoral model achieved the highest performance(training cohort:AUC=0.856;validation cohort:AUC=0.830),while the peritumoral model performed best at a 3-mm region(training cohort:AUC=0.842;validation cohort:AUC=0.795). Further comparisons of multi-parametric models showed that the ctDNA methylation-based model had limited predictive efficacy(training cohort:AUC=0.773;validation cohort:AUC=0.744),whereas the combined model integrating ctDNA methylation and radiomic features yielded the best predictive performance in both cohorts,(training cohort:AUC=0.924;validation cohort:AUC=0.912). Conclusion The combination of ctDNA methylation biomarkers with intratumoral and peritumoral multi -parametric MRI radiomic features demonstrates high clinical value for predicting ALNM in breast cancer,and the integrated model offers both high sensitivity and specificity.
文章编号:     中图分类号:R737.9 R445.2    文献标志码:A
基金项目:河北省医学科学研究课题计划项目(20241606)
引用文本:


Scan with WeChat

Scan with WeChat