对比增强MRI的影像组学分析预测低级别胶质瘤的DNA拷贝数亚型及结局
回顾性分析153例连续患者的对比增强MR图像,提取1167个影像组学特征,构建整合影像组学与影像学特征的列线图预测低级别胶质瘤DNA拷贝数亚型。CN2亚型总生存期和无进展生存期最短,三个亚型模型的AUC分别为0.924、0.829和0.837。
要点
- CN2亚型低级别胶质瘤患者的无进展生存期和总生存期最短。
- 整合影像组学与影像学特征的列线图可区分CN2亚型,AUC为0.924。
- CN1和CN3亚型列线图的AUC分别为0.829和0.837,校准良好。
核心结论
- CN2亚型低级别胶质瘤患者无进展生存期和总生存期最短;整合影像组学和影像学特征的模型在预测DNA拷贝数亚型和临床结局方面表现出良好性能。
研究设计与样本
- 回顾性分析153例连续患者的资料,从对比增强MR图像提取1167个影像组学特征,采用LASSO逻辑回归筛选特征并构建影像组学标签,采用多变量逻辑回归构建三个拷贝数相关影像学模型。
诊断标准与关键数值
- CN2亚型列线图纳入影像组学标签(AUC:0.891,OR:2.345;P=0.001)、结节外生长(OR:14.413;P<0.001)和宽度(OR:0.194;P=0.027),区分CN2亚型的AUC为0.924(95%CI:0.869-0.979)。
- CN1亚型列线图纳入影像组学标签(AUC:0.730,OR:2.408;P=0.001)、出血(OR:0.100;P<0.001)、边界不清(OR:4.433;P=0.001)和体积≥60cm³(OR:4.195;P=0.002),AUC为0.829(95%CI:0.765-0.892)。
- CN3亚型列线图纳入影像组学标签(AUC:0.660,OR:2.518;P=0.003)、坏死/囊变(OR:6.975;P=0.008)、出血(OR:3.723;P=0.024)、边界不清(OR:0.124;P<0.001)和额叶肿瘤(OR:4.870;P<0.001),AUC为0.837(95%CI:0.767-0.909)。
- 十折交叉验证平均准确率:CN2亚型92.8%,CN1亚型72.6%,CN3亚型79.0%。
临床应用与阅片要点
- 原文结论指出,整合影像组学和影像学特征的模型在预测DNA拷贝数亚型和临床结局方面表现出良好性能。
- 决策曲线分析表明所有影像学模型具有临床实用性。
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)。
英文原文(摘要)· 点击展开
RATIONALE AND OBJECTIVES: DNA copy-number (CN)2-subtype impairs outcomes in patients with lower-grade gliomas (LGG). We aimed to determine the value of preoperative nomograms integrating radiomic and radiographic (RR) features in predicting DNA copy-number subtype.
METHODS: Data of 153 consecutive patients were retrospectively analyzed. A total of 1167 radiomics features were extracted from contrast-enhanced MR images. LASSO logistic regression was performed to choose the key features and construct a radiomics signature. Three CN-related RR model were built with multivariate logistic regression.
RESULTS: CN2-subtype was associated with shortest median PFS(p <0.001) and OS (p <0.001). The radiomics nomogram, which incorporated the signature (AUC:0.891, OR: 2.345; p = 0.001), extranodular growth (OR: 14.413; p <0.001) and width (OR: 0.194; p = 0.027), distinguished CN2-subtype with an AUC of 0.924(95%CI: 0.869-0.979).The radiomics nomogram, which incorporated the signature (AUC:0.730, OR: 2.408; p = 0.001), hemorrhage (OR: 0.100; p <0.001), poorly-defined margin (OR:4.433; p = 0.001) and volume>=60cm3 (OR: 4.195; p = 0.002) were associated with CN1-subtype (AUC:0.829,95%CI:0.765-0.892).The radiomics nomogram, which incorporated the signature (AUC:0.660, OR: 2.518; p = 0.003), necrosis/cystic(OR:6.975; p = 0.008), hemorrhage (OR:3.723; p = 0.024), poorly-defined margin (OR:0.124; p <0.001) and frontal lobe tumors (OR: 4.870; p <0.001) were associated with CN3-subtype (AUC: 0.837,95%CI: 0.767-0.909).All three RR models showed good discrimination and calibration. Decision curve analysis indicated that all RR models were clinically useful. The average accuracy of the ten-fold cross validation was 92.8% for CN2-subtype, 72.6% for CN1-subtype and 79.0% for CN3-subtype.
CONCLUSION: The shortest PFS and OS was observed in LGG patients with CN2-subtype. The RR models, integrating radiomic and radiographic features, demonstrates good performance for predicting DNA copy-number subtype and clinical outcomes.
引用
- Wan, Zhou, Zhang, et al. Radiomic Analysis of Contrast-Enhanced MRI Predicts DNA Copy-Number Subtype and Outcome in Lower-Grade Gliomas. Acad Radiol. 2021. PMID: 34916150. doi:10.1016/j.acra.2021.10.014
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。