| 谢玉海,韩剑剑,马培旗,王小雷,马文俊,张宁宁,曹雪花,魏天贤,陈楠.基于多中心数字化乳腺X线摄影影像组学预测乳腺癌人表皮生长因子受体-2过表达[J].中国医学影像技术,2023,39(3):365~369 |
| 基于多中心数字化乳腺X线摄影影像组学预测乳腺癌人表皮生长因子受体-2过表达 |
| Radiomics based on multi-center digital mammography for predicting human epidermal growth factor receptor-2 overexpression of breast cancer |
| 投稿时间：2022-08-12 修订日期：2023-01-15 |
| DOI：10.13929/j.issn.1003-3289.2023.03.011 |
| 中文关键词: 乳腺肿瘤 乳房X线摄影 人表皮生长因子受体-2 影像组学 |
| 英文关键词:breast neoplasms mammography human epidermal growth factor receptor-2 radiomics |
| 基金项目:皖南医学院科研项目（JXYY202139）。 |
|
| 摘要点击次数: 2967 |
| 全文下载次数: 774 |
| 中文摘要: |
| 目的 观察基于多中心数字化乳腺X线摄影（DM）影像组学预测乳腺癌人表皮生长因子受体-2（HER-2）过表达的价值。方法 回顾性分析来自机构1（n=336）、2（n=142）、3（n=130）共608例经病理证实乳腺癌患者的DM资料。按照7∶3比例将来自机构1、2共478例分为训练集（334例，其中92例HER-2阳性、242例HER-2阴性）和验证集（144例，40例HER-2阳性、104例HER-2阴性）；以来自机构3的130例（33例HER-2阳性及97例HER-2阴性）为外部验证集。于显示病变面积较大的内外斜（MLO）位或头足（CC）位DM图像中勾画病灶ROI，提取及筛选最佳影像组学特征，以支持向量机（SVM）构建影像组学模型，用于预测乳腺癌HER-2过表达；采用受试者工作特征（ROC）曲线评估模型的诊断效能，绘制校准曲线及决策曲线，评价其校准度及临床获益。结果 共筛选出3个最佳影像组学特征，以之构建的SVM影像组学模型预测训练集、验证集及外部验证集乳腺癌HER-2过表达的曲线下面积（AUC）分别为0.824、0.775及0.812。校准曲线显示，该模型在训练集、验证集及外部验证集的平均绝对误差分别为0.08、0.05和0.04；决策曲线分析显示，阈值概率为0.2~0.8时，模型的临床实用性较好。结论 基于多中心DM影像组学可有效预测乳腺癌HER-2过表达。 |
| 英文摘要: |
| Objective To investigate the value of multi-center digital mammography (DM) radiomics for predicting human epidermal growth factor receptor-2 (HER-2) overexpression of breast cancer. Methods DM data of 608 patients with pathologically confirmed breast cancer from institution 1 (n=336), 2 (n=142) and 3 (n=130) were retrospectively analyzed. Totally 478 patients from institution 1 and 2 were randomly divided into training set (n=334, 92 cases of HER-2 positive and 242 cases of HER-2 negative) and validation set (n=144, 40 cases of HER-2 positive and 104 cases of HER-2 negative) at the ratio of 7:3, while 130 patients from institution 3 (33 cases of HER-2 positive and 97 cases of HER-2 negative) were taken as external validation set. The mediolateral oblique (MLO) or cranial caudal (CC) DM images showed larger lesion areas were selected to delineate lesion ROI, and the best radiomics features were extracted and screened. Then support vector machine (SVM) were performed to construct a radiomics model for predicting HER-2 overexpression of breast cancer. Receiver operating characteristic (ROC) curve was used to evaluate the diagnostic efficacy of the radiomics model, while calibration curve and decision curve were drawn to evaluate its calibration degree and clinical value. Results Three best radiomics features were screened out and used to construct the radiomics model, and its area under the curve (AUC) for predicting HER-2 overexpression of breast cancer in the training set, validation set and external validation set was 0.824, 0.775 and 0.812, respectively. The calibration curve showed that the mean absolute error of the radiomics model in the training set, validation set and external validation set was 0.08, 0.05 and 0.04, respectively. The decision curve analysis showed that the threshold probability was 0.2-0.8, the radiomics model had good clinical applicability. Conclusion Multi-center DM radiomics could effectively predict HER-2 overexpression of breast cancer. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
