| 宋德领,崔书君,杨飞,马永青,张玉姣,郭亚哲,朱月香.基于动态对比增强MRI影像组学模型预测乳腺癌新辅助化疗后非病理完全缓解[J].中国医学影像技术,2021,37(4):547~551 |
| 基于动态对比增强MRI影像组学模型预测乳腺癌新辅助化疗后非病理完全缓解 |
| Radiomics model based on dynamic contrast-enhanced MRI for predicting breast cancer non-pathological complete response after neoadjuvant chemotherapy |
| 投稿时间：2020-03-09 修订日期：2020-09-19 |
| DOI：10.13929/j.issn.1003-3289.2021.04.016 |
| 中文关键词: 乳腺肿瘤 磁共振成像 影像组学 新辅助治疗 |
| 英文关键词:breast neoplasms magnetic resonance imaging radiomics neoadjuvant therapy |
| 基金项目: |
|
| 摘要点击次数: 3319 |
| 全文下载次数: 1064 |
| 中文摘要: |
| 目的 观察基于动态对比增强MRI （DCE-MRI）影像组学模型预测乳腺癌新辅助化疗（NAC）后非病理完全缓解（non-pCR）的价值。方法 回顾性分析144例经病理证实并接受NAC的乳腺癌患者，按照7[DK （]∶[DK）]3比例将其分入训练组（n=99）和验证组（n=45），比较2组病理完全缓解（pCR）与non-pCR患者临床病理指标的差异。以MaZda软件提取4个周期NAC后DCE-MRI所示病灶纹理特征，以最小绝对收缩与选择算子（LASSO）算法及十折交叉验证法筛选最优特征参数，建立影像组学标签。采用多因素Logistic回归法构建包含影像组学标签和差异具有统计学意义的临床病理指标在内的联合预测模型，以受试者工作特征（ROC）曲线评价影像组学标签及模型预测乳腺癌患者NAC non-pCR的效能。结果 训练组36例pCR、63例non-pCR，验证组分别为15例和30例。2组pCR与non-pCR患者间孕激素受体（PR）、人类表皮生长因子受体2（HER2）和细胞增殖核抗原（Ki-67）表达差异均有统计学意义（P均<0.05）。共筛选出8个最优特征参数建立影像组学标签，以之预测训练组和验证组患者NAC后non-pCR的曲线下面积（AUC）分别为0.85和0.84；而以联合预测模型预测训练组患者NAC后non-pCR的AUC、敏感度、特异度分别为0.90、88.89%及83.33%，验证组分别为0.89、83.33%及86.67%。结论 基于DCE-MRI的影像组学模型对预测乳腺癌NAC后non-pCR具有一定价值。 |
| 英文摘要: |
| Objective To explore the value of the radiomics model based on dynamic contrast-enhanced MRI (DCE-MRI) for predicting non-pathological complete response (non-pCR) of breast cancer after neoadjuvant chemotherapy (NAC). Methods Data of 144 breast cancer patients confirmed by pathology who received NAC were retrospective analyzed. The patients were divided into training group (n=99) and validation group (n=45) in a ratio of 7 ∶ 3. Clinicopathological data of pathological complete response (pCR) and non-pCR were compared between groups. MaZda software was used to extract the texture features of breast cancers on DCE-MRI after 4 periods of NAC. The least absolute shrinkage and selection operator (LASSO) regression method were used to screen the optimal features with the method of 10 fold cross-validation, so as to establish the radiomics signatures. A combined prediction model including the radiomics signatures and clinicopathological indicators being statistically different between pCR and non-pCR were constructed with multivariable Logistic regression. Receiver operating characteristic (ROC) curve method was used to evaluate the diagnostic efficacy of radiomics signatures and combined prediction model in predicting non-pCR of breast cancer after NAC. Results There were 36 cases of pCR and 63 cases of non-pCR in training group, 15 cases of pCR and 30 cases of non-pCR in validation group. Significant differences of progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2) and nuclear-associated antigen (Ki-67) expressionwere found between pCR and non-pCR patients in both groups (all P<0.05). Eight optimal feature parameters were selected for establishment of radiomics signature. The area under the curve (AUC) of radiomics signatures for predicting non-pCR after NAC was 0.85 and 0.84 in training group and validation group, respectively. The AUC, sensitivity and specificity of the combined prediction model for predicting non-pCR after NAC in training group was 0.90, 88.89% and 83.33%, while in validation group was 0.89, 83.33% and 86.67%, respectively. Conclusion Radiomics model based on DCE-MRI had certain value in predicting non-pCR of breast cancer after NAC. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
