| 摘要: |
| 目的 探讨基于腹腔镜超声的纹理分析鉴别诊断肾透明细胞癌和非透明细胞癌的价值。方法 回顾性纳入华中科技大学同济医学院附属同济医院超声科2012年12月至2020年6月经病理证实的肾细胞癌患者的腹腔镜超声检查资料。根据病理结果分为肾透明细胞癌和非透明细胞癌两组。在腹腔镜二维超声图像上通过ITK-SNAPE软件手工勾画感兴趣区,然后采用Pyradiomics包提取纹理特征。使用组内和组间相关系数来选择具有良好稳定性和可重复性的特征。使用最大相关最小冗余(mRMR)和最小绝对收缩和选择算子(LASSO)算法进行特征选择和建模,采用ROC曲线评估模型诊断效能。结果 共纳入83例肾细胞癌病灶(肾透明细胞癌组66例,非透明细胞癌组17例)。基于6个纹理特征构建的预测模型鉴别肾透明细胞癌和非透明细胞癌的ROC 曲线下面积,敏感度,特异度和准确度分别为0.860、0.765、0.864、0.843。结论 基于腹腔镜超声的纹理分析可以准确鉴别肾透明细胞癌与非透明细胞癌。 |
| 关键词: 纹理分析 腹腔镜超声 肾透明细胞癌 肾细胞癌 |
| DOI： |
| 投稿时间：2021-08-31修订日期：2021-10-07 |
| 基金项目:湖北省自然科学基金资助项目(2020CFB597) |
|
| Texture Analysis Based on Laparoscopic Ultrasound Images in Discrimination of Clear Cell Renal Cell Carcinoma and non-Clear Cell Renal Cell Carcinoma |
| Wang Ting,Guan Wei,Li Fan,Yu Yang,Deng You-bin,Deng Xuan |
| (Department of Medical Ultrasound, Tongji Hospital,Tongji Medical College,Huazhong University of Science and Technology) |
| Abstract: |
| Objective To investigate the feasibility of texture analysis based on laparoscopic ultrasound images for distinguishing clear cell renal carcinoma (ccRCC) from non-clear cell renal carcinoma (non-ccRCC). Methods From December 2012 to June 2020, the laparoscopic ultrasonography data of patients with renal cell carcinoma confirmed by pathology in Tongji Hospital of Tongji Medical College of Huazhong University of Science and Technology were analyzed retrospectively. The region of interest (ROI) was drawn manually by ITK-SNAPE software on the laparoscopic B-mode ultrasound images, and the python-based Pyradiomics package was used to extracted the texture features. Intra-group and inter-group correlation coefficients were used to select features with good stability and reproducibility. The maximum correlation minimum redundancy (mRMR) and minimum absolute contraction and selection operator (LASSO) algorithms were used to select the features and develop the model. The diagnostic performance of the model was evaluated by ROC curves. Results A total of 83 RCC lesions were included There were 66 cases of ccRCC and 17 cases of non-ccRCC according to the pathological results. The area under the curve of the predictive model based on the six texture features was 0.860, with sensitivity 0.765, specificity 0.864 and accuracy 0.843 for distinguishing ccRCC from non-ccRCC Conclusion The texture analysis based on laparoscopic ultrasound images can be used for distinguishing ccRCC from non-ccRCC. |
| Key words: Texture analysis Laparoscopic ultrasonography Clear cell renal cell carcinoma Renal cell carcinoma |
