| 李若馨,温幸林,刘锦辉,冷晓玲,李昱凝.超声影像组学联合显微超声造影(SR CEUS)用于乳腺癌组织学分级[J].中国医学影像技术,2026,42(6):854~858 |
| 超声影像组学联合显微超声造影(SR CEUS)用于乳腺癌组织学分级 |
| Ultrasound radiomics combined with super resolution contrast-enhanced ultrasound (SR CEUS) for histological grading of breast cancer |
| 投稿时间：2025-12-06 修订日期：2026-05-01 |
| DOI：10.13929/j.issn.1003-3289.2026.06.009 |
| 中文关键词: 乳腺肿瘤 肿瘤,组织学类型 影像组学 显微超声造影 |
| 英文关键词:breast neoplasms neoplasms by histologic type radiomics super resolution contrast-enhanced ultrasound |
| 基金项目:国家自然科学基金(82360362)。 |
|
| 摘要点击次数: 161 |
| 全文下载次数: 40 |
| 中文摘要: |
| 目的 探讨超声影像组学联合显微超声造影(SR CEUS)用于乳腺癌组织学分级的价值。方法 回顾性纳入 229例乳腺癌,依据病理结果分高、中及低分化组(57、107、65例),并按7∶3比例随机分为训练集(n=160)与验证集(n=69)。行SR CEUS定量分析,获取血管密度(VD)、速度方差(Vel Var)等参数。于常规超声图中勾画肿瘤ROI并提取影像组学特征,以Pearson相关分析、最小绝对收缩和选择算子(LASSO)算法筛选特征,分别采用随机森林(RF)、支持向量机(SVM)、逻辑回归(LR)、K邻近法(KNN)、极限梯度提升(XGBoost)及分类梯度提升(CatBoost)机器学习方法建立影像组学模型;以单及多因素logistic回归分析筛选乳腺癌组织学级别的独立预测因子并构建SR CEUS模型,基于预测效能最佳的影像组学模型与SR CEUS独立预测因子构建联合模型。绘制受试者工作特征(ROC)曲线,以曲线下面积(AUC)及Obuchowski指数评价模型效能。结果 RF模型为预测效能最佳的影像组学模型,其在验证集的微观平均和宏观平均AUC分别为0.901和0.908、Obuchowski指数为0.901。以VD及Vel Var作为独立预测因子构建的SR CEUS模型在验证集的微观平均和宏观平均AUC分别为0.839和0.830、Obuchowski指数为0.824。联合模型在验证集的微观平均和宏观平均AUC分别为0.918和0.927、Obuchowski指数为0.924,预测效能最高。结论 超声影像组学联合SR CEUS用于乳腺癌组织学分级具有较高价值。 |
| 英文摘要: |
| Objective To investigate the value of ultrasound radiomics combined with super resolution contrast-enhanced ultrasound (SR CEUS) for histological grading of breast cancer. Methods Totally 229 patients with pathologically confirmed breast cancer were retrospectively enrolled and categorized into well-, moderately- and poorly-differentiated groups (57, 107 and 65 cases, respectively) according to pathological findings, also randomly divided into training set (n=160) and validation set (n=69) at a ratio of 7∶3. SR CEUS parameters, including vessel density (VD) and velocity variance (Vel Var) were quantitatively analyzed. Tumor ROI were delineated on conventional ultrasound images, and radiomics features were extracted and screened using Pearson correlation analysis, the least absolute shrinkage and selection operator (LASSO) algorithm. Machine learning methods, including random forest (RF), support vector machine (SVM), logistic regression (LR), K-nearest neighbors (KNN), extreme gradient boosting (XGBoost) and category gradient boosting (CatBoost) were used to construct radiomics models. The independent predictive factors of histological grade of breast cancer were screened with univariate and multivariate logistic regression analysis, and SR CEUS model was constructed. Finally a combined model was established combining the best radiomics model and SR CEUS independent predictive factors. Receiver operating characteristic (ROC) curves were drawn, the area under the curve (AUC) and Obuchowski index were used to evaluate the efficacy of the above models. Results RF model was the optimal radiomics model, with micro-average and macro-average AUC of 0.901 and 0.908 and Obuchowski index of 0.901 in validation set, respectively. Constructed based on independent predictive factors including VD and Vel Var, SR CEUS model had micro-average and macro-average AUC of 0.839 and 0.830 and Obuchowski index of 0.824 in validation set, respectively. The micro-average and macro-average AUC of combined model was 0.918 and 0.927, and Obuchowski index was 0.924 in validation set, which showed the best efficiency. Conclusion Ultrasound radiomics combined with SR CEUS had high value for histological grading of breast cancer. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
