Discrimination of benign and malignant ovarian sex cord-stromal tumors through the analysis of clinical features, MR imaging, and MR-based radiomics.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiology, The Third Affiliated Hospital of Zhengzhou University, Zhengzhou, China.
- Department of Research Collaboration, R&D center, Hangzhou Deepwise & League of PHD Technology Co., Ltd, Hangzhou, China.
- Tianjian Laboratory of Advanced Biomedical Sciences, Institute of Advanced Biomedical Sciences, Zhengzhou University, Zhengzhou, China.
Abstract
This study aims to compare diagnostic performance of clinical parameters, conventional MRI, and radiomic signatures, and to develop a multimodal nomogram for differentiating benign from malignant ovarian sex cord-stromal tumors (SCSTs). Retrospectively enrolled 113 patients with pathologically confirmed 123 SCSTs (42 malignant and 81 benign). Univariate and multivariate logistic regression analysis identified the independent predictors for clinical and conventional MRI models. Radiomics features were extracted from fat-suppressed T2-weighted imaging (FS-T2WI) and diffusion-weighted imaging (DWI) sequences, with extreme gradient boosting (XGBoost) used to build radiomics models. A combined nomogram integrated Rad-scores with the independent clinical and conventional MRI predictors. Model performance was evaluated by receiver operating characteristic (ROC) analysis and decision curve analysis (DCA) in development and validation cohorts. Among traditional methods, the traditional model achieved the highest AUCs (training: 0.880 [<i>95% CI</i>: 0.807-0.946]; validation: 0.908 [<i>95% CI</i>: 0.797-0.989]). Among radiomics-based models, T2&DWI model achieved the highest AUCs (training: 0.887 [<i>95% CI</i>: 0.802-0.955]; validation: 0.843 [<i>95% CI</i>: 0.684-0.963]). The combined model demonstrated optimal predictive performance across all models, with AUCs of 0.945 (<i>95% CI</i>: 0.892-0.985) in training and 0.914 (<i>95% CI</i>: 0.798-0.988) in validation cohorts, respectively. Furthermore, DCA confirmed greater clinical net benefit of the combined model across threshold probabilities compared to other models. While both traditional model and MRI-based radiomics model demonstrate significant potential for SCSTs stratification, our final combined nomogram achieves optimal predictive performance and net benefit. This tool may enhance personalized therapeutic management for SCSTs patients.