Development and Validation of CT-Based Radiomic Nomograms for Prediction of Survival and Chemotherapeutic Benefits in Gastric Cancer

Wenjuan Zhang,Jiangliang Ren,Tiezhu Ren,Juan Deng,ShengLin Li,Xiaoai Ke,Haiting Yang,Peng Zhang,Bo Long,Junlin Zhou

DOI: https://doi.org/10.2139/ssrn.4174462

2022-01-01

Abstract:Background: We aimed to develop CT-based radiomics models for predicting survival and chemotherapeutic benefits after gastric cancer (GC) resection.Methods: Three-hundred-and-twenty-two consecutive GC patients underwent feature extraction from preoperative diagnostic CT images. A radiomics signature reflecting the phenotypes of the primary tumor and adjacent peritoneum for predicting disease-free survival (DFS) was generated based on selected features with stability and the Boruta algorithm. Two radiomics-based nomograms, one integrating preoperative and one integrating preoperative and postoperative clinical risk factors, were constructed using multivariable Cox regression analysis. Performance was evaluated using Harrell’s concordance indices (C-index), time-independent receiver operating characteristic analysis, calibration, and clinical usefulness.Findings: Five key features were selected to construct a radiomics signature significantly associated with DFS (P<0.001). The preoperative radiomics-based nomogram (pre-nomogram) incorporated the radiomics signature and two clinical variables (cT and cN stage). The postoperative radiomics-based nomogram (post-nomogram), incorporating the radiomics signature and three clinical variables (cN stage, surgical approach, and serum CEA), exhibited better prognostic ability for the estimation of DFS than other rival models, with C-indices of 0.791 (95%CI, 0.751-0.831) and 0.742 (95%CI, 0.670-0.814) in the training and validation cohorts, respectively. The nomogram provided two risk strata with high or low risk of survival; further analysis showed that patients in the high-risk group exhibited a favorable response to adjuvant chemotherapy, while those in the low-risk group did not.Interpretation: The proposed CT-based radiomics nomogram could serve as powerful predictors of DFS in patients with GC and identify suitable candidates for chemotherapy to guide personalized treatment.Funding Information: This work was supported by the National Natural Science Foundation of China (82102151), the Open Fund project of Key Laboratory of Medical Imaging of Gansu Province (GSYX202006).Declaration of Interests: One of the co-authors, Jialiang Ren, is an employee of GE Healthcare. The other authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article.Ethics Approval Statement: Ethical approval was obtained for this retrospective analysis (Ethics Committees at Lanzhou University Second Hospital [approval date: March 11, 2017]). The requirement for informed consent was waived.