| 王平,裴旭,殷小平.基于增强CT的放射组学模型用于鉴别肾透明细胞癌与非透明细胞癌[J].中国医学影像技术,2019,35(11): |
| 基于增强CT的放射组学模型用于鉴别肾透明细胞癌与非透明细胞癌 |
| Radiomics model based on enhanced CT for identifying renal clear cell carcinoma from non-transparent cell carcinoma |
| 投稿时间：2019-01-15 修订日期：2019-11-03 |
| DOI： |
| 中文关键词: 肾癌 增强CT 放射组学 随机森林 |
| 英文关键词:renal cell carcinoma CT enhancement Random forest Radiomics |
| 基金项目:保定市科技计划项目（18ZF182） |
|
| 摘要点击次数: 1125 |
| 全文下载次数: 352 |
| 中文摘要: |
| 【】目的：建立放射组学在肾癌增强CT图像的应用模型,并评估模型鉴别肾透明细胞癌与非透明细胞癌的准确性。方法：回顾性分析100例经病理证实的增强CT肾癌病例,其中肾透明细胞癌81例,非透明细胞癌19例；使用ITK-SNAP图像分割软件对增强CT图像的所有病灶动脉期进行感兴趣区(ROI)勾画并融合成为感兴趣区容积(VOI)；本次研究应用A.K软件(Artificial Intelligence Kit,GE医疗,美国)一共获得了396个组学特征,对样本的两个分组分别按照7：3的比例随机分为训练集和测试集,使用单因素方差分析+秩和检验、一般线性模型、前10% mutual information 进行特征降维,最终筛选出了8个特征,建立了两个相关特征的模型,所有模型均在训练集和测试集分别进行ROC曲线分析。结果：基于增强CT病灶动脉期筛选的8个特征建立了两个组学模型,分别是随机森林(Random Forest, RF)模型和逻辑回归(Logistic)模型。RF在训练集的AUC为0.988,敏感性为0.93,特异性为,1,精确度为1；在测试集AUC为0.972,敏感性为0.875,特异性为1,精确度为1。逻辑回归模型在训练集AUC为0.865,敏感性为0.667,特异性为0.923,精确度为0.974；在测试集AUC为0.868,敏感性为0.75,特异性为1,精确度为1。结论：与Logistic模型相比, RF模型在鉴别肾透明细胞癌与非透明细胞癌具有更高的准确度。 |
| 英文摘要: |
| 【】Objective: To establish an application model of radiomics for enhanced CT images of renal cell carcinoma and to evaluate the accuracy of the model for distinguishing renal clear cell carcinoma from non-clear cell renal cell carcinoma. Methods: A retrospective analysis of 100 cases of enhanced CT renal cell carcinoma confirmed by pathology, including 81 cases of renal clear cell carcinoma and 19 cases of non-clear cell renal cell carcinoma; By using ITK-SNAP image segmentation software to segment the region of interest (ROI) of all the lesions of the enhanced CT image ,ROI was merged into the volume of interest (VOI); this study used A.K. software (Artificial Intelligence Kit, GE Healthcare, USA) to obtain a total of 396 omics features, and the two groups of samples were the 7:3 ratio randomly divided into training set and testing set. The One-way ANOVA + Rank Sum test, Correlation analysis and the first 10% mutual information methods are used for feature dimensionality reduction. Finally, 8 features are selected and two models are established. Both models were analyzed by ROC curves in the training set and test set, respectively. Results: Two radiomics models were established based on the eight features of enhanced CT lesions in the arterial phase, namely the Random Forest (RF) model and the Logistic Regression model. The RF has an AUC of 0.988, a sensitivity of 0.93, a specificity of 1, and an accuracy of 1 in the training set; A testing set of the AUC of 0.972, a sensitivity of 0.875, a specificity of 1, and an accuracy of 1. The Logistic Regression model had an AUC of 0.865, a sensitivity of 0.667, a specificity of 0.923, and an accuracy of 0.974 in the training set; The AUC was 0.868, the sensitivity was 0.75, the specificity was 1, and the accuracy was 1 in the testing set. Conclusion: To compared with the Logistic model, the RF model has higher accuracy in identifying renal clear cell carcinoma from non-clear cell renal cell carcinoma. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
