Assistance by Routine CT Features Combined With 3D Texture Analysis in the Diagnosis of BRCA Gene Mutation Status in Advanced Epithelial Ovarian Cancer

PURPOSE: To evaluate the predictive value of routine CT features combined with 3D texture analysis for prediction of BRCA gene mutation status in advanced epithelial ovarian cancer. METHOD: Retrospective analysis was performed on patients with masses occupying the pelvic space confirmed by pathology and complete preoperative images in our hospital, including 37 and 58 cases with mutant type and wild type BRCA, respectively (total: 95 cases). The enrolled patients’ routine CT features were analyzed by two radiologists. Then, ROIs were jointly determined through negotiation, and the ITK-SNAP software package was used for 3D outlining of the third-stage images of the primary tumor lesions and obtaining texture features. For routine CT features and texture features, Mann-Whitney U tests, single-factor logistic regression analysis, minimum redundancy, and maximum correlation were used for feature screening, and the performance of individual features was evaluated by ROC curves. Multivariate logistic regression analysis was used to further screen features, find independent predictors, and establish the prediction model. The established model’s diagnostic efficiency was evaluated by ROC curve analysis, and the histogram was obtained to conduct visual analysis of the prediction model. RESULTS: Among the routine CT features, the type of peritoneal metastasis, mesenteric involvement, and supradiaphragmatic lymph node enlargement were correlated with BRCA gene mutation (P < 0.05), whereas the location of the peritoneal metastasis (in the gastrohepatic ligament) was not significantly correlated with BRCA gene mutation (P > 0.05). Multivariate logistic regression analysis retained six features, including one routine CT feature and five texture features. Among them, the type of peritoneal metastasis was used as an independent predictor (P < 0.05), which had the highest diagnostic efficiency. Its AUC, accuracy, specificity, and sensitivity were 0.74, 0.79, 0.90, and 0.62, respectively. The prediction model based on the combination of routine CT features and texture features had an AUC of 0.86 (95% CI: 0.79–0.94) and accuracy, specificity, and sensitivity of 0.80, 0.76, and 0.81, respectively, indicating a better performance than that of any single feature. CONCLUSIONS: Both routine CT features and texture features had value for predicting the mutation state of the BRCA gene, but their predictive efficiency was low. When the two types of features were combined to establish a predictive model, the model’s predictive efficiency was significantly higher than that of independent features.