A Nomogram Combined Radiomic and Semantic Features as Imaging Biomarker for Classification of Ovarian Cystadenomas

Objective: To construct and validate a combined Nomogram model based on radiomic and semantic features to preoperatively classify serous and mucinous pathological types in patients with ovarian cystadenoma. Methods: A total of 103 patients with pathology-confirmed ovarian cystadenoma who underwent CT examination were collected from two institutions. All cases divided into training cohort (N = 73) and external validation cohort (N = 30). The CT semantic features were identified by two abdominal radiologists. The preprocessed initial CT images were used for CT radiomic features extraction. The LASSO regression were applied to identify optimal radiomic features and construct the Radscore. A Nomogram model was constructed combining the Radscore and the optimal semantic feature. The model performance was evaluated by ROC analysis, calibration curve and decision curve analysis (DCA). Result: Five optimal features were ultimately selected and contributed to the Radscore construction. Unilocular/multilocular identification was significant difference from semantic features. The Nomogram model showed a better performance in both training cohort (AUC = 0.94, 95%CI 0.86-0.98) and external validation cohort (AUC = 0.92, 95%CI 0.76-0.98). The calibration curve and DCA analysis indicated a better accuracy of the Nomogram model for classification than either Radscore or the loculus alone. Conclusion: The Nomogram model combined radiomic and semantic features could be used as imaging biomarker for classification of serous and mucinous types of ovarian cystadenomas.