结直肠癌肝转移患者CT机器学习和影像组学分析预测RAS突变状态
回顾性纳入2018年1月至2021年5月接受肝转移灶手术切除且有术前增强CT和RAS检测的28例患者共77个肝转移灶,提取851个影像组学特征。线性回归模型结合z-score归一化后12个稳健特征预测RAS突变,AUC 0.953、准确率98%、敏感度96%、特异度100%。
要点
- CT门脉期影像组学特征经z-score归一化后用线性回归模型预测RAS突变,AUC达0.953。
- 最优单变量预测特征为original_shape_Maximum2DDiameter和wavelet_HLL_glcm_InverseVariance,准确率80%。
- 归一化后的线性回归模型性能优于所测试的机器学习分类器。
核心结论
- CT影像组学分析中采用归一化方法可预测结直肠癌肝转移患者的RAS突变状态。
研究设计与样本
- 回顾性研究,纳入2018年1月至2021年5月接受肝转移灶手术切除、有病理证实的肝转移、术前增强CT图像质量良好且以RAS评估为标准参考的患者,共28例患者77个肝转移灶,平均年龄60岁(范围40-80岁);由两名专家放射科医生在CT门脉期逐层分割每个肝转移灶,先独立后共识,使用PyRadiomics从Slicer 3D平台提取851个影像组学特征,并进行平衡处理及组间和组内相关系数计算。
诊断标准与关键数值
- 单变量分析中,两种归一化方法下最佳预测特征均为original_shape_Maximum2DDiameter和wavelet_HLL_glcm_InverseVariance,准确率80%、AUC≥0.75、敏感度≥80%、特异度≥70%(P值<<0.01);多变量分析中,z-score归一化后线性回归模型组合12个稳健特征性能最佳,AUC 0.953、准确率98%、敏感度96%、特异度100%、PPV 100%、NPV 96%(P值<<0.01)。
临床应用与阅片要点
- 在术前CT门脉期图像上逐层分割肝转移灶并提取影像组学特征,经z-score归一化后使用线性回归模型可能有助于预测RAS突变状态。
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
PURPOSE: To assess the efficacy of machine learning and radiomics analysis by computed tomography (CT) in presurgical setting, to predict RAS mutational status in colorectal liver metastases.
METHODS: Patient selection in a retrospective study was carried out from January 2018 to May 2021 considering the following inclusion criteria: patients subjected to surgical resection for liver metastases; proven pathological liver metastases; patients subjected to enhanced CT examination in the presurgical setting with a good quality of images; and RAS assessment as standard reference. A total of 851 radiomics features were extracted using the PyRadiomics Python package from the Slicer 3D image computing platform after slice-by-slice segmentation on CT portal phase by two expert radiologists of each individual liver metastasis performed first independently by the individual reader and then in consensus. Balancing technique was performed, and inter- and intraclass correlation coefficients were calculated to assess the between-observer and within-observer reproducibility of features. Receiver operating characteristics (ROC) analysis with the calculation of area under the ROC curve (AUC), sensitivity (SENS), specificity (SPEC), positive predictive value (PPV), negative predictive value (NPV) and accuracy (ACC) were assessed for each parameter. Linear and non-logistic regression model (LRM and NLRM) and different machine learning-based classifiers were considered. Moreover, features selection was performed before and after a normalized procedure using two different methods (3-sigma and z-score).
RESULTS: Seventy-seven liver metastases in 28 patients with a mean age of 60 years (range 40-80 years) were analyzed. The best predictors, at univariate analysis for both normalized procedures, were original_shape_Maximum2DDiameter and wavelet_HLL_glcm_InverseVariance that reached an accuracy of 80%, an AUC ≥ 0.75, a sensitivity ≥ 80% and a specificity ≥ 70% (p value < < 0.01). However, a multivariate analysis significantly increased the accuracy in RAS prediction when a linear regression model (LRM) was used. The best performance was obtained using a LRM combining linearly 12 robust features after a z-score normalization procedure: AUC of 0.953, accuracy 98%, sensitivity 96%, specificity of 100%, PPV 100% and NPV 96% (p value < < 0.01). No statistically significant increase was obtained considering the tested machine learning both without normalization and with normalization methods.
CONCLUSIONS: Normalized approach in CT radiomics analysis allows to predict RAS mutational status in colorectal liver metastases patients.
引用
- Granata V, Fusco R, Setola SV, et al. Machine learning and radiomics analysis by computed tomography in colorectal liver metastases patients for RAS mutational status prediction. Radiol Med. 2024. PMID: 38761342. doi:10.1007/s11547-024-01828-5
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。