基于术前分期CT的影像组学标签预测三阴性乳腺癌

Preoperative staging CT-based radiomics signature in predicting triple-negative breast cancer

目的:探讨基于术前分期CT的影像组学标签在预测三阴性乳腺癌分子分型中的附加价值。方法:回顾性收集2016年1月至2018年5月经手术病理证实且均为临床术前评估分期需行常规胸部CT增强扫描的481例肿块型乳腺浸润性癌患者,按照样本量1∶2随机抽样选取三阴性乳腺癌与非三阴性乳腺癌共计150例患者(90例作为训练组,60例作为验证组)。所有患者均经免疫组织化学检测,获得乳腺癌分子分型。对所有患者基于病灶三维图像提取影像组学特征,并采用Lasso logistic回归模型进行特征降维及筛选,以建立影像组学标签。采用ROC曲线评价影像组学标签对三阴性乳腺癌的鉴别诊断效能。结果:由5个关键影像组学特征构成的影像组学标签与乳腺癌三阴性分子分型相关( P <0.0001)。建立的影像组学标签对于鉴别三阴性乳腺癌具有较好的预测效能,其在训练组和验证组的ROC曲线下面积(AUC)分别为0.766(95% CI:0.743~0.789)和0.758(95% CI:0.718~ 0.798)。结论:基于术前分期CT建立的影像组学标签有助于三阴性与非三阴性乳腺癌的鉴别,这是术前常规胸部增强CT扫描在辅助临床分期之外的附加价值,可为临床治疗决策提供参考。

Objective: To explore the additional value of preoperative staging CT-based radiomics signature in predicting molecular subtype of triple-negative breast cancer. Methods: From January 2016 to May 2018,481 patients with pathological confirmed invasive breast cancer (mass type) were enrolled in the study.All patients were performed routine staging enhanced chest CT before surgery.A total of 150 cases were randomly selected according to the ratio of 1:2 (triple-negative vs non-triple-negative breast cancer).Among the 150 cases,90 cases were divided into the training dataset and 60 cases were divided into the validation dataset.All patients obtained detailed information about molecular subtypes of breast cancer by immunohistochemistry.The radiomic features were extracted based on volume of the interest in all patients and Lasso logistic regression model was used for dimensionality reduction,feature selection and construction of the radiomics signature.The receiver operation curve (ROC) was used to evaluate the performance of the radiomics signature in predicting triple-negative breast cancer. Results: The radiomics signature composed of 5 radiomic features was associated with the triple-negative molecular subtype of breast cancer ( P <0.0001).The constructed radiomics signature obtained good predictive performance in identifying triple-negative breast cancer,with the area under the ROC curve (AUC) of 0.766 (95%CI:0.743~0.789) and 0.758(95%CI:0.718~0.798) in the training and validation dataset,respectively. Conclusion: The constructed radiomics signature based on preoperative staging CT could be helpful in differentiating between triple-negative and non-triple-negative breast cancer,which can provide additional value of routine preoperative staging CT for clinical treatment decisions.