Yongping Zhou1, Jitao Wang2, Shuang-Lin Zhang3, Hao Wang1, Yong Yan1, Xiumin Qi4, Shubo Chen2, Fang-Ming Chen5. 1. Department of Hepatobiliary Surgery, The Affiliated Wuxi No. 2 People's Hospital of Nanjing Medical University, Wuxi, China. 2. Department of Hepatobiliary Surgery, Xingtai Institute of Cancer Control, The Affiliated Xingtai People's Hospital of Hebei Medical University, Xingtai, China. 3. Department of Radiology, The Affiliated Wuxi No. 2 People's Hospital of Nanjing Medical University, Wuxi, China. 4. Department of Pathology, the Affiliated Wuxi No. 2 People's Hospital of Nanjing Medical University, Wuxi, China. 5. Department of Radiology, The Affiliated Wuxi No. 2 People's Hospital of Nanjing Medical University, Wuxi, China. fmchencoil@126.com.
Abstract
BACKGROUND: The current radiologic criteria for assessing intraoperative superior mesenteric-portal vein (SMPV) involvement (i.e., presence of tumor-SMPV contact >180° or venous deformity) in pancreatic ductal adenocarcinoma (PDAC) are highly specific but insufficiently sensitive. Therefore, development of improved markers for a more accurate prediction is essential. This study aimed to develop a risk score model to estimate SMPV involvement in PDAC using radiomics analysis of computed tomography (CT) images. METHODS: Data from two institution-based cohorts of PDAC patients undergoing preoperative CT scans were used to develop (n = 173) and validate (n = 156) a radiomics-based risk score of SMPV involvement using clinical and imaging variables. A radiomics signature was developed based on 2436 radiomic features extracted from the semi-automatic three-dimensional segmentation ofn CT images. The SMPV involvement risk score was built using multivariate logistic regression and compared with the current radiologic criteria. RESULTS: The study surgically identified SMPV involvement in 59 (34.1%) and 57(36.5 %) patients with PDAC in the development and validation cohorts, respectively. A 12-feature-based radiomics signature achieved areas under receiver operating characteristics curves (AUCs) of 0.89 or greater for estimating SMPV involvement. Multivariate regression identified the radiomics signature and SMPV deformity as associated with SMPV involvement. The risk score model had significantly improved AUC (0.928 vs. 0.768; P < 0.001) and sensitivity (84.2% vs. 66.7%; P = 0.025) in the radiologic evaluation. CONCLUSIONS: The novel risk score in this study, combining radiomics signature and venous deformity, demonstrated promising performance for estimating SMPV involvement preoperatively for patients with PDAC.
BACKGROUND: The current radiologic criteria for assessing intraoperative superior mesenteric-portal vein (SMPV) involvement (i.e., presence of tumor-SMPV contact >180° or venous deformity) in pancreatic ductal adenocarcinoma (PDAC) are highly specific but insufficiently sensitive. Therefore, development of improved markers for a more accurate prediction is essential. This study aimed to develop a risk score model to estimate SMPV involvement in PDAC using radiomics analysis of computed tomography (CT) images. METHODS: Data from two institution-based cohorts of PDAC patients undergoing preoperative CT scans were used to develop (n = 173) and validate (n = 156) a radiomics-based risk score of SMPV involvement using clinical and imaging variables. A radiomics signature was developed based on 2436 radiomic features extracted from the semi-automatic three-dimensional segmentation ofn CT images. The SMPV involvement risk score was built using multivariate logistic regression and compared with the current radiologic criteria. RESULTS: The study surgically identified SMPV involvement in 59 (34.1%) and 57(36.5 %) patients with PDAC in the development and validation cohorts, respectively. A 12-feature-based radiomics signature achieved areas under receiver operating characteristics curves (AUCs) of 0.89 or greater for estimating SMPV involvement. Multivariate regression identified the radiomics signature and SMPV deformity as associated with SMPV involvement. The risk score model had significantly improved AUC (0.928 vs. 0.768; P < 0.001) and sensitivity (84.2% vs. 66.7%; P = 0.025) in the radiologic evaluation. CONCLUSIONS: The novel risk score in this study, combining radiomics signature and venous deformity, demonstrated promising performance for estimating SMPV involvement preoperatively for patients with PDAC.