Ultrasound based radiomics nomogram combined with clinical parameters to predict lymphovascular space invasion in endometrioid adenocarcinoma

This study aims to assess the predictive ability of a radiomics nomogram incorporating clinical features and ultrasound radiomics signature in determining the presence of lymphovascular space invasion (LVSI) in endometrioid adenocarcinoma (EAC) before surgical intervention. This retrospective, single-center study included 171 patients diagnosed with EAC. Stratified random sampling was utilized to divide the data into a training group for model construction, and a test group for assessing the model's reliability, with a ratio of 7:3. Ultrasound radiomics features were extracted from the ultrasound images. Then, the Z-score method and the least absolute shrinkage and selection operator (LASSO) were used to select significant features, and the ultrasound radiomics score (Rad-score) was constructed. A comprehensive prediction model was established based on the multivariate logistic regression analysis, and a nomogram was drawn. The model diagnostic performance was assessed via the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity on training and test sets. Model comparisons were performed using the Delong test. Univariate and multivariate logistic regression analyses found that the independent risk factors of LVSI in EAC were preoperative histological grade and Rad-score. In the comprehensive prediction model, the AUC, sensitivity, specificity, and accuracy for the training set were 0.83 (95% CI: 0.76-0.91), 0.96, 0.64, and 0.71; for the test set, the values were 0.75 (95% CI: 0.57-0.93), 0.75, 0.70, and 0.71, respectively. The ultrasound radiomics and comprehensive prediction models showed good prediction efficiency, characterized by high sensitivity with moderate specificity for LVSI in EAC. In the preoperative evaluation of LVSI in EAC, the comprehensive prediction model can obtain high sensitivity with moderate specificity.