The predictive potential of contrast-enhanced computed tomography based radiomics in the preoperative staging of cT4 gastric cancer

Quantitative Imaging in Medicine and Surgery

article

Bo Liu¹  Dengyun Zhang¹  He Wang¹  Hexiang Wang²  Pengfei Zhang³  Dawei Zhang¹  Qun Zhang⁴  Jian Zhang¹ 

[1] Department of Gastrointestinal Surgery , the Affiliated Hospital of Qingdao University;Department of Radiology , the Affiliated Hospital of Qingdao University;Department of Radiology , the Weihai Wendeng District People’s Hospital;Multidisciplinary Centre , the Institute of High Energy Physics of the Chinese Academy of Sciences

关键词: Gastric cancer;    contrast-enhanced computed tomography (CE-CT);    radiomics;    T staging;   

DOI  :  10.21037/qims-22-286

学科分类：外科医学

来源: AME Publications

【 摘 要 】

Background: The accuracy of preoperative staging is crucial for cT4 stage gastric cancer patients. The aim of this study was to develop the radiomics model and evaluate its predictive potential for differentiating preoperative cT4 stage gastric cancer patients into pT4b and no-pT4b patients. Methods: A multicenter retrospective analysis of 704 gastric cancer patients with preoperative contrast-enhanced computed tomography (CE-CT) staging cT4 between January 2008 and December 2021. These patients were divided into the training cohort (478 patients, the Affiliated Hospital of Qingdao University) and validation cohort (226 patients, the Weihai Wendeng District People’s Hospital). According to the pathological stage of the tumors, the patients were divided into pT4b or no-pT4b stage. In the training cohort, the clinical and radiomics features were analyzed to construct the clinical model, tri-phase radiomics signatures and nomogram. Two kinds of methods were employed to achieve dimensionality reduction: (I) the least absolute shrinkage and selection operator (LASSO); and (II) the minimum redundancy maximum relevance (mRMR) algorithms. We utilized Logistic regression, support vector machine (SVM), Decision tree and Adaptive boosted tree (AdaBoost) algorithms as the machine learning classifiers. The nomogram was constructed on the clinical characteristics and the Rad-score. The performance of the models was evaluated by receiver operating characteristic (ROC) area under the curve (AUC), Decision Curve Analysis (DCA) curve and calibration curve. Results: The 345 pT4b and 359 no-pT4b stage patients were included in this study. In the validation cohort, the AUC of the clinical model was 0.793 (95% CI: 0.732–0.855). The tri-phase radiomics features combined with the SVM algorithm was the best radiomics signature with an AUC of 0.862 (95% CI: 0.812–0.912). The nomogram was the best predictive model of all with an AUC of 0.893 (95% CI: 0.834–0.927). In the training and validation cohorts, the calibration curves and DCA curves of the nomogram showed satisfactory result. Conclusions: CE-CT-based radiomics nomogram offers good accuracy and stability in differentiating preoperative cT4 stage gastric cancer patients into pT4b and non-pT4b stages, which has a great clinical relevance for selecting the course of treatment for cT4 stage gastric cancer patients.

【 授权许可】

All Rights reserved