| 刘晓东,王新宇,宁刚.MRI影像组学术前预测乳腺浸润性导管癌Ki-67表达[J].中国医学影像技术,2022,38(2):210~214 |
| MRI影像组学术前预测乳腺浸润性导管癌Ki-67表达 |
| MRI radiomics for preoperative predicting Ki-67 expression of breast invasive ductal carcinoma |
| 投稿时间：2021-04-22 修订日期：2021-09-18 |
| DOI：10.13929/j.issn.1003-3289.2022.02.012 |
| 中文关键词: 乳腺肿瘤 磁共振成像 影像组学 Ki-67 |
| 英文关键词:breast neoplasms magnetic resonance imaging radiomics Ki-67 |
| 基金项目: |
|
| 摘要点击次数: 1290 |
| 全文下载次数: 665 |
| 中文摘要: |
| 目的 观察基于MR弥散加权成像（DWI）及表观弥散系数（ADC）图影像组学术前预测乳腺浸润性导管癌Ki-67表达的价值。方法 回顾性分析212例经组织穿刺活检或手术病理证实的乳腺浸润性导管癌患者的乳腺DWI，以分层抽样方法按7∶3比例将其分为训练集（n=148）和验证集（n=64）。分别基于DWI、ADC图及DWI+ADC图提取全肿瘤影像组学特征，采用常数项剔除法、Spearman相关性分析、最小绝对收缩和选择算子算法筛选最优影像组学特征，应用L1正则化logistic回归方法建立预测乳腺浸润性导管癌Ki-67高低表达的模型DWI、模型ADC及模型DWI+ADC，并以验证集数据进行验证。采用受试者工作特征（ROC）曲线评价各模型的预测效能，计算曲线下面积（AUC），并以DeLong检验比较其差异。结果 212例乳腺浸润性导管癌中，176例呈Ki-67高表达，36例Ki-67低表达。分别基于DWI、ADC图、DWI+ADC图提取124、124及248个影像组学特征，并最终筛选出12、7及13个最优影像组学特征。ROC曲线显示，模型DWI、模型ADC及模型DWI+ADC在训练集数据的AUC分别为0.82、0.87、0.94，在验证集数据的AUC分别为0.81、0.84、0.93；DeLong检验显示，模型DWI+ADC的AUC高于模型DWI（Z=－3.44，P<0.01）及模型ADC（Z=－2.79，P=0.01）。结论 术前基于DWI和ADC图影像组学模型均可有效预测乳腺浸润性导管癌Ki-67表达状态，联合模型的诊断效能更优。 |
| 英文摘要: |
| Objective To explore the value of radiomics of MR diffusion weighted imaging (DWI) and apparent diffusion coefficient (ADC) images in preoperative prediction of Ki-67 expression in breast invasive ductal carcinoma. Methods DWI data of 212 patients with breast invasive ductal carcinoma confirmed by pathology were retrospective analyzed. The patients were divided into training set (n=148) and verification set (n=64) at the ratio of 7:3. The radiomics features of whole tumor were extracted based on DWI, ADC images and DWI+ADC images, then the optimal features were selected with constant term elimination method, Spearman correlation analysis, least absolute shrinkage and selection operator algorithm. L1 regularization logistic regression method was used to establish modelDWI, modelADC and modelDWI+ADC to predict high and low expression of Ki-67 in breast invasive ductal carcinomas, respectively. The prediction models were verified in verification set data. The receiver operating characteristic (ROC) curve was used to evaluate the prediction efficiency of each model, and the corresponding area under the curve (AUC) was calculated and compared with DeLong test. Results Among 212 patients, 176 were found with high Ki-67 expression and 36 with low Ki-67 expression breast invasive ductal carcinoma. Totally 124, 124 and 248 radiomics features were extracted, and finally 12, 7, 13 optimal radiomics features were screened based on DWI, ADC images and DWI+ADC images, respectively. ROC curves showed that the AUC of modelDWI, modelADC and modelDWI+ADC in training set data was 0.82, 0.87, 0.94, while in validation set data was 0.81, 0.84, and 0.93, respectively. DeLong test showed that the AUC of modelDWI+ADC was higher than that of modelDWI (Z=-3.44, P<0.01) and modelADC (Z=-2.79, P=0.01). Conclusion Radiomics models based on DWI and ADC images could effectively predict Ki-67 expression status of breast invasive ductal carcinoma preoperatively, and the combining model of DWI and ADC images had better diagnostic efficiency. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
