Bo Zhao1, Hai-Tao Zhu, Xiao-Ting Li, Yan-Jie Shi, Kun Cao, Ying-Shi Sun. 1. From the Department of Radiology, Key Laboratory of Carcinogenesis and Translational Research, Ministry of Education, Peking University Cancer Hospital and Institute, Haidian District, Beijing, China.
Abstract
OBJECTIVES: We investigated the value of radiomics data, extracted from pretreatment computed tomography images of the primary tumor (PT) and lymph node (LN) for predicting LN metastasis in esophageal squamous cell carcinoma (ESCC) patients. MATERIALS AND METHODS: A total 338 ESCC patients were retrospectively assessed. Primary tumor, the largest short-axis diameter LN (LSLN), and PT and LSLN interaction term (IT) radiomic features were calculated. Subsequently, the radiomic signature was combined with clinical risk factors in multivariable logistic regression analysis to build various clinical-radiomic models. Model performance was evaluated with respect to the fit, overall performance, differentiation, and calibration. RESULTS: A clinical-radiomic model, which combined clinical and PT-LSLN-IT radiomic signature, showed favorable discrimination and calibration. The area under curve value was 0.865 and 0.841 in training and test set. CONCLUSIONS: A venous computed tomography radiomic model based on the PT, LSLN, and IT radiomic features represents a novel noninvasive tool for prediction LN metastasis in ESCC.
OBJECTIVES: We investigated the value of radiomics data, extracted from pretreatment computed tomography images of the primary tumor (PT) and lymph node (LN) for predicting LN metastasis in esophageal squamous cell carcinoma (ESCC) patients. MATERIALS AND METHODS: A total 338 ESCC patients were retrospectively assessed. Primary tumor, the largest short-axis diameter LN (LSLN), and PT and LSLN interaction term (IT) radiomic features were calculated. Subsequently, the radiomic signature was combined with clinical risk factors in multivariable logistic regression analysis to build various clinical-radiomic models. Model performance was evaluated with respect to the fit, overall performance, differentiation, and calibration. RESULTS: A clinical-radiomic model, which combined clinical and PT-LSLN-IT radiomic signature, showed favorable discrimination and calibration. The area under curve value was 0.865 and 0.841 in training and test set. CONCLUSIONS: A venous computed tomography radiomic model based on the PT, LSLN, and IT radiomic features represents a novel noninvasive tool for prediction LN metastasis in ESCC.