A preoperative pathological staging prediction model for esophageal cancer based on CT radiomics
Haojun Li,
Shuoming Liang,
Mengxuan Cui,
Weiqiu Jin,
Xiaofeng Jiang,
Simiao Lu,
Jicheng Xiong,
Hainan Chen,
Ziwei Wang,
Guotai Wang,
Jiming Xu,
Linfeng Li,
Yao Wang,
Haomiao Qing,
Yongtao Han,
Xuefeng Leng
Affiliations
Haojun Li
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Shuoming Liang
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Mengxuan Cui
Yidu Cloud Technology Inc
Weiqiu Jin
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Xiaofeng Jiang
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Simiao Lu
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Jicheng Xiong
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Hainan Chen
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Ziwei Wang
School of Medicine, University of Electronic Science and Technology
Guotai Wang
School of Mechanical and Electrical Engineering, University of Electronic Science and Technology of China
Jiming Xu
Yidu Cloud Technology Inc
Linfeng Li
Yidu Cloud Technology Inc
Yao Wang
Yidu Cloud Technology Inc
Haomiao Qing
Department of Radiology, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Yongtao Han
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Xuefeng Leng
Department of Thoracic Surgery, Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China
Abstract Background Accurate and comprehensive preoperative staging is one of the most important prognostic factors for the management of esophageal cancer (EC). We aimed to develop and validate predictive models using radiomics from preoperative contrast-enhanced Computed Tomography (CT) images to assess pathological staging in EC patients. Methods This study retrospectively included 161 patients who underwent esophagectomy at Sichuan Cancer Hospital from July 2018 to February 2023. Pathological staging outcomes encompassed overall TNM staging, T and N staging, and tumor progressions (vascular invasion and perineural invasion). Radiomics features were extracted from segmented regions of tumors. A radiomic signature (Rad-signature) for each outcome was developed using a fivefold cross-validation least absolute shrinkage and selection operator (LASSO) regression model within the training cohort and subsequently validated in the test cohort for predictive accuracy. Results Out of the 851 radiomics features extracted, two were selected to formulate the Rad-signature for each staging outcome. These signatures showed a significant correlation with their respective outcomes in both the training set and the testing set. Furthermore, the Rad-signature exhibited favorable predictive performance for advanced pTNM staging, advanced pT staging, vascular invasion and perineural invasion, with AUC of 0.721 [95%CI, 0.570–0.872], 0.900 [95%CI 0.805–0.995], 0.824 [0.686–0.961], and 0.737 [0.586–0.887], respectively. However, the predictive performance of the Rad-signature for pN staging is moderate (AUC = 0.693 [0.534–0.852]), indicating needs for additional data modalities. Conclusions This study established a non-invasive preoperative radiomics model that demonstrated good predictive performance in determining the pTNM staging, pT staging, vascular invasion, and perineural invasion for EC patients. These results could inform personalized treatment strategies and improve outcomes for EC patients.
