Yu-Ting Peng1, Chuan-Yang Zhou1, Peng Lin1, Dong-Yue Wen1, Xiao-Dong Wang1, Xiao-Zhu Zhong1, Deng-Hua Pan1, Qiao Que1, Xin Li2, Ling Chen2, Yun He3, Hong Yang4. 1. Department of Medical Ultrasonics, First Afliated Hospital of Guangxi Medical University, No. 6 Shuangyong Road, Nanning 530021, Guangxi Zhuang, China. 2. GE Healthcare, Shanghai, China. 3. Department of Medical Ultrasonics, First Afliated Hospital of Guangxi Medical University, No. 6 Shuangyong Road, Nanning 530021, Guangxi Zhuang, China. Electronic address: 228388072@qq.com. 4. Department of Medical Ultrasonics, First Afliated Hospital of Guangxi Medical University, No. 6 Shuangyong Road, Nanning 530021, Guangxi Zhuang, China. Electronic address: yanghong@gxmu.edu.cn.
Abstract
RATIONALE AND OBJECTIVES: The purpose of this study was to establish and validate radiomics signatures based on ultrasound (US) medicine images to assess the biological behaviors of intrahepatic cholangiocarcinoma (ICC) in a noninvasive manner. MATERIALS AND METHODS: This study consisted of 128 ICC patients. We focused on evaluating six pathological features: microvascular invasion, perineural invasion, differentiation, Ki-67, vascular endothelial growth factor, and cytokeratin 7. Region of interest (ROI) of ICC was identified by manually plotting the tumor contour on the grayscale US image. We extracted radiomics features from medical US imaging. Then, dimensionality reduction methods and classifiers were used to develop radiomic signatures for evaluating six pathological features in ICC. Finally, independent validation datasets were used to assess the radiomic signatures performance. RESULTS: We extracted 1076 quantitative characteristic parameters on the US medicine images. Based on extracted radiomics features, the best performing radiomic signatures for evaluating microvascular invasion features were produced by hypothetical test + support vector machine (SVM), perineural invasion subgroup were least absolute shrinkage and selection operator + principal component analysis + support vector machine, differentiation subgroup were hypothetical test + decision tree, Ki-67 subgroup were hypothetical test + logistic regression, vascular endothelial growth factor subgroup were hypothetical test + Gradient Boosting Decision Tree (GBDT), and cytokeratin 7 subgroup were hypothetical test + bagging, respectively. CONCLUSION: Through the high-throughput radiomics analysis based on US medicine images, we proposed radiomics signatures that have moderate efficiency in predicting the biological behaviors of ICC noninvasively.
RATIONALE AND OBJECTIVES: The purpose of this study was to establish and validate radiomics signatures based on ultrasound (US) medicine images to assess the biological behaviors of intrahepatic cholangiocarcinoma (ICC) in a noninvasive manner. MATERIALS AND METHODS: This study consisted of 128 ICC patients. We focused on evaluating six pathological features: microvascular invasion, perineural invasion, differentiation, Ki-67, vascular endothelial growth factor, and cytokeratin 7. Region of interest (ROI) of ICC was identified by manually plotting the tumor contour on the grayscale US image. We extracted radiomics features from medical US imaging. Then, dimensionality reduction methods and classifiers were used to develop radiomic signatures for evaluating six pathological features in ICC. Finally, independent validation datasets were used to assess the radiomic signatures performance. RESULTS: We extracted 1076 quantitative characteristic parameters on the US medicine images. Based on extracted radiomics features, the best performing radiomic signatures for evaluating microvascular invasion features were produced by hypothetical test + support vector machine (SVM), perineural invasion subgroup were least absolute shrinkage and selection operator + principal component analysis + support vector machine, differentiation subgroup were hypothetical test + decision tree, Ki-67 subgroup were hypothetical test + logistic regression, vascular endothelial growth factor subgroup were hypothetical test + Gradient Boosting Decision Tree (GBDT), and cytokeratin 7 subgroup were hypothetical test + bagging, respectively. CONCLUSION: Through the high-throughput radiomics analysis based on US medicine images, we proposed radiomics signatures that have moderate efficiency in predicting the biological behaviors of ICC noninvasively.
Authors: Lang Chen; Furong Zeng; Lei Yao; Tongdi Fang; Mengting Liao; Jing Long; Liang Xiao; Guangtong Deng Journal: Cancer Med Date: 2020-01-05 Impact factor: 4.452