Prediction of cervical stromal invasion using ultrasound radiomics: from conventional ultrasound to intelligent diagnosis.

Rationale and objectives: Accurate preoperative assessment of cervical stromal invasion (CSI) in endometrial cancer (EC) is crucial for surgical planning, but conventional imaging has limited diagnostic performance. This study aims to develop and validate a non-invasive ultrasound radiomics model to improve the preoperative prediction accuracy of CSI. Materials and methods: We retrospectively analyzed 294 patients with pathologically confirmed EC, randomly assigned (7:3) to training (n=205) and test (n=89) cohorts. Regions of interest were manually segmented on 2D ultrasound images using 3D-Slicer, from which 837 radiomics features were extracted. Following screening for reproducibility (ICCs ≥ 0.75) and univariate analysis, the features were finally reduced to 25 via LASSO regression, and the radiomic score (Radscore) was calculated. Similarly, six clinical predictors (menopause duration, parity, CA-125, tumor size, endometrial-myometrial junction, vascularity grade) were identified via univariate analysis and LASSO regression and combined into a clinical score (C_score). Logistic regression was used to build radiomics, clinical, and combined nomogram models. Performance was assessed using AUC, calibration curves, and decision curve analysis (DCA). Results: The radiomics model achieved the highest AUCs: 0.975 (95% CI: 0.958-0.991) in the training cohort and 0.906 (95% CI: 0.848-0.965) in the test cohort, significantly outperforming the clinical model (AUC: 0.947 and 0.832) and the nomogram model (AUC: 0.945 and 0.838). Furthermore, it showed good calibration and provided substantial net clinical benefit across a wide range of threshold probabilities on DCA. Conclusion: Ultrasound radiomics is a promising non-invasive tool for preoperatively predicting CSI in EC, with potential to enhance personalized treatment planning.