Xin Chen1, Mengjie Fang2, Di Dong2, Xinhua Wei3, Lingling Liu3, Xiangdong Xu3, Xinqing Jiang3, Jie Tian4, Zaiyi Liu5. 1. The Second School of Clinical Medicine, Southern Medical University, 1023 Shatai Nan Road, Guangzhou, 510515, China; Department of Radiology, Guangdong General Hospital, Guangdong Academy of Medical Sciences, 106 Zhongshan Er Road, Guangzhou, 510080, China; Department of Radiology, Guangzhou First People's Hospital, The Second Affiliated Hospital of South China University of Technology, 1 Panfu Road, Guangzhou, China. 2. University of Chinese Academy of Sciences, 95 Zhongguancun Dong Road, Beijing, 100190, China. 3. Department of Radiology, Guangzhou First People's Hospital, The Second Affiliated Hospital of South China University of Technology, 1 Panfu Road, Guangzhou, China. 4. Department of Radiology, Guangdong General Hospital, Guangdong Academy of Medical Sciences, 106 Zhongshan Er Road, Guangzhou, 510080, China; University of Chinese Academy of Sciences, 95 Zhongguancun Dong Road, Beijing, 100190, China. Electronic address: jie.tian@ia.ac.cn. 5. The Second School of Clinical Medicine, Southern Medical University, 1023 Shatai Nan Road, Guangzhou, 510515, China; Department of Radiology, Guangdong General Hospital, Guangdong Academy of Medical Sciences, 106 Zhongshan Er Road, Guangzhou, 510080, China. Electronic address: zyliu@163.com.
Abstract
RATIONALE AND OBJECTIVES: Poorly differentiated non-small cell lung cancer (NSCLC) indicated a poor prognosis and well-differentiated NSCLC indicates a noninvasive nature and good prognosis. The purpose of this study was to build and validate a radiomics signature to predict the degree of tumor differentiation (DTD) for patients with NSCLC. MATERIALS AND METHODS: A total of 487 patients with pathologically diagnosed NSCLC were retrospectively included in our study. Five hundred ninety-one radiomics features were extracted from each tumor from the contrast-enhanced computed tomography images. A minimum redundancy maximum relevance algorithm and a logistic regression model were used for dimension reduction, feature selection, and radiomics signature building. The performance of the radiomics signature was assessed using receiver operating characteristic analysis, and the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy were calculated to quantify the association between a signature and DTD. An independent validation set contained 184 consecutive patients with NSCLC. RESULTS: A nine-radiomics-feature-based signature was built and it could differentiate low and high DTDs in the training set (AUC = 0.763, sensitivity = 0.750, specificity = 0.665, and accuracy = 0.687), and the radiomics signature had good discrimination performance in the validation set (AUC = 0.782, sensitivity = 0.608, specificity = 0.752, and accuracy = 0.712). CONCLUSIONS: A radiomics signature based on contrast-enhanced computed tomography imaging is a potentially useful imaging biomarker for differentiating low from high DTD in patients with NSCLC.
RATIONALE AND OBJECTIVES: Poorly differentiated non-small cell lung cancer (NSCLC) indicated a poor prognosis and well-differentiated NSCLC indicates a noninvasive nature and good prognosis. The purpose of this study was to build and validate a radiomics signature to predict the degree of tumor differentiation (DTD) for patients with NSCLC. MATERIALS AND METHODS: A total of 487 patients with pathologically diagnosed NSCLC were retrospectively included in our study. Five hundred ninety-one radiomics features were extracted from each tumor from the contrast-enhanced computed tomography images. A minimum redundancy maximum relevance algorithm and a logistic regression model were used for dimension reduction, feature selection, and radiomics signature building. The performance of the radiomics signature was assessed using receiver operating characteristic analysis, and the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy were calculated to quantify the association between a signature and DTD. An independent validation set contained 184 consecutive patients with NSCLC. RESULTS: A nine-radiomics-feature-based signature was built and it could differentiate low and high DTDs in the training set (AUC = 0.763, sensitivity = 0.750, specificity = 0.665, and accuracy = 0.687), and the radiomics signature had good discrimination performance in the validation set (AUC = 0.782, sensitivity = 0.608, specificity = 0.752, and accuracy = 0.712). CONCLUSIONS: A radiomics signature based on contrast-enhanced computed tomography imaging is a potentially useful imaging biomarker for differentiating low from high DTD in patients with NSCLC.