CT-Based Radiomics and Machine Learning for Predicting Cement Leakage After Percutaneous Vertebroplasty: A Retrospective Study.

Authors

Abstract

Cement leakage is a frequent complication after percutaneous vertebroplasty (PVP) for osteoporotic vertebral compression fractures (OVCFs). Accurate preoperative risk assessment may facilitate clinical decision-making and procedural planning. This retrospective study included 145 patients who underwent PVP between October 2020 and October 2025. Patients were divided into a training cohort (n=101) and an internal validation cohort (n =44). CT-based radiomic features were extracted and selected using a multistep pipeline. Logistic regression (LR), support vector machine (SVM), and k-nearest neighbors (KNN) were used to construct radiomics, clinical-only, Radscore-based, and combined models. Model performance was evaluated using AUC, calibration, decision curve analysis, and SHAP analysis. In the internal validation cohort, the feature-level LR and SVM radiomics models achieved the highest observed AUC (0.793), whereas the KNN model achieved an AUC of 0.759. The combined models showed the highest observed AUCs within the LR and SVM frameworks, while the Radscore-based model performed slightly better than the combined model within the KNN framework. A statistically significant improvement was observed only for the LR-based combined model compared with the clinical-only model. CT-based radiomics may provide complementary information to clinical variables for predicting cement leakage after PVP. However, the differences among models and classifiers should be interpreted cautiously because of the limited sample size and internal validation design. CT-based radiomics may have potential for preoperative risk stratification of cement leakage after PVP. External validation is required before clinical application.

Tags

Topics