| 周宇堃,甄俊平,靳波,边文瑾,杨洁,樊知昌.基于T1WI及IDEAL-T2WI影像组学模型鉴别腮腺多形性腺瘤和腺淋巴瘤[J].中国医学影像技术,2020,36(5):675~679 |
| 基于T1WI及IDEAL-T2WI影像组学模型鉴别腮腺多形性腺瘤和腺淋巴瘤 |
| Differentiating parotid gland pleomorphic adenoma from adenolymphoma based on T1WI and IDEAL-T2WI radiomics models |
| 投稿时间：2019-09-17 修订日期：2020-03-24 |
| DOI：10.13929/j.issn.1003-3289.2020.05.008 |
| 中文关键词: 腮腺 腺瘤,多形性 腺淋巴瘤 影像组学 机器学习 磁共振成像 |
| 英文关键词:parotid gland adenoma,pleomorphic adenolymphoma radiomics machine learning magnetic resonance imaging |
| 基金项目: |
|
| 摘要点击次数: 3250 |
| 全文下载次数: 681 |
| 中文摘要: |
| 目的 观察T1WI、IDEAL-T2WI影像组学特征及机器学习模型鉴别腮腺多形性腺瘤（PA）与腺淋巴瘤（AL）的价值。方法 回顾性分析33例腮腺PA和25例AL患者。运用放射组学云平台，于轴位T1WI和IDEAL-T2WI上手动勾画病灶ROI，提取影像组学特征；以方差阈值法、SelectKBest及Lasso算法筛选最优特征。采用随机森林、逻辑回归算法建立机器学习模型，绘制ROC曲线，分析T1WI、IDEAL-T2WI、IDEAL-T2WI联合T1WI建立模型的诊断效能。结果 T1WI、IDEAL-T2WI、IDEAL-T2WI联合T1WI分别得到6、9、12个有效特征。基于IDEAL-T2WI联合T1WI建立随机森林模型的诊断效能最高，AUC为0.87，95%CI（0.59，1.00），准确率0.83。结论 基于T1WI、IDEAL-T2WI影像组学特征及机器学习模型可有效鉴别诊断腮腺PA和AL。 |
| 英文摘要: |
| Objective To explore the value of radiomics features and machine learning models based on T1WI and IDEAL-T2WI in differential diagnosis of parotid gland pleomorphic adenoma (PA) from adenolymphoma (AL). Methods Clinical and imaging data of 58 patients with parotid tumors, including 33 with PA and 25 with AL were retrospectively analyzed. Axial T1WI and IDEAL-T2WI were manually segmented, and radiomics features were extracted using the radcloud software. Effective radiomics features were selected with the variance threshold method, SelectKBest and Lasso algorithm. The machine learning models were established by using random forest and Logistic regression algorithm, and the ROC curves were drawn to analyze the diagnostic performance. The ability of T1WI, IDEAL-T2WI and image combination in diagnosis of PA from AL were analyzed. Results T1WI, IDEAL-T2WI and IDEAL-T2WI combined with T1WI obtained 6, 9 and 12 effective features. The random forest model based on IDEAL-T2WI combined with T1WI sequences had the highest diagnostic performance, with AUC of 0.87 (95% CI[0.59,1.00]) and accuracy of 0.83. Conclusion Radiomics features and machine learning models based on T1WI and IDEAL-T2WI can provide important references for differentiation of PA and AL. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
