The value of a nomogram based on clinical data and contrast enhanced CT radiomics in the preoperative prediction of Epstein-Barr virus-associated gastric carcinoma

Abstract

Objective: To explore the value of a nomogram based on clinical data and enhanced CT radiomics in the prediction of Epstein-Barr virus-associated gastric carcinoma(EBVaGC). Methods: The data of 136 patients, including 100 males and 36 females, aged [M (Q1, Q3)] 65 (53, 71) years, with gastric cancer confirmed by surgery and pathology were retrospectively analyzed. According to Epstein-Barr virus-encoded small RNA (EBER) in situ hybridization, those patients were divided into Epstein-Barr virus (EBV) positive group (n=32) and EBV negative group (n=104). All patients underwent multi-phase enhanced CT scanning before surgery and randomly assigned to the training group (n=95) and validation group (n=41) in a ratio of 7︰3. MaZda software was used to extract radiomics features of enhanced CT images. The intra-group correlation coefficient (ICC), variance analysis and minimum absolute shrinkage and selection algorithm (LASSO) regression were used to reduce the dimensionality of the radiomics features, and then the radiomics score (Radscore) was calculated. The nomogram model was based on combined clinical data, morphological features and Radscore. The predictive power of the nomogram was evaluated according to the area under the receiver operating characteristic curve (AUC), and the net clinical benefit of the nomogram was evaluated by the decision curve and calibration curves were drawn according to the data of the training group and the validation group to analyze the consistency of the nomogram model. Results: After selection, six optimal radiomics features were obtained, including Mean, Skewness, S(1, 0) Sum entropy, S(1, 1) Contrast, 99% percentile and S(2, 2)Angular second moment. Radscore of EBV positive group were higher than that of the EBV negative group (training group: 3.78±0.83 vs 2.80±0.98; validation group: 3.81±0.47 vs 2.94±0.95) (both P<0.05) both in the training group and validation group. The AUC of the radiomics model in training group and validation group were 0.773(95%CI:0.612-0.962)and 0.792(95%CI:0.597-0.927)respectively,and the sensitivity and specificity were 63.6% and 93.1%, 70.0% and 87.1%, respectively. The AUC of the nomogram model based on clinical data and radiomics in the training group and the validation group were 0.883(95%CI:0.644-0.984) and 0.851(95%CI:0.715-0.996), respectively. The nomogram model showed superior predictive performance (both P<0.05). Conclusion: The nomogram model based on clinical data and radiomics has better efficacy in the prediction of Epstein-Barr virus associated gastric cancer.