Prediction of microvascular invasion based on CT in gastric cancer

Abstract

Abstract Background Microvascular invasion (MVI) is an important step in cancer cell migration and invasion, and it is also a significant factor in predicting tumor recurrence and prognosis. Building a nomogram based on CT image features and clinicopathological data to predict preoperative MVI in gastric cancer (GC). Methods Retrospective study enrolled 358 patients with surgically proven GC. Univariate and multivariate logistic regression analyses were performed to identify the predictors for the model and establish a nomogram for MVI. The performance of the model was evaluated using ROC, accuracy, and C index. Internal validation of the model was conducted using the bootstrap resampling method. Difference in the area under the curve (AUC) between the two models was evaluated using the Delong test. Random forest algorithm is used to extract important risk factors for MVI. Results Mural stratification, Lauren classification and Albumin (Alb) were found to be independent influencing factors for MVI. The nomogram model incorporating these three factors showed significantly better performance compared to the original model that did not include CT parameters (P < 0.05). The AUC of the model was 0.779 (95% CI 0.774–0.868), and the average AUC of the bootstrap sample was 0.813. The sensitivity, specificity, and accuracy of the model were 65.6%, 86.0%, and 70.7%, respectively. Conclusion The nomogram based on CT image features and clinicopathological data demonstrated good predictive value for MVI in GC. This nomogram can provide valuable baseline information for individualized treatment of GC.