Machine Learning Applications for the Prediction of Bone Cement Leakage in Percutaneous Vertebroplasty

• DOI: 10.3389/fpubh.2021.812023
• Publication Date: 2021-12-10T05:43:20Z
• Authors: Wenle Li; Jiaming Wang; Wencai Liu; Chan Xu; Wanying Li; Kai Zhang; Shibin Su; Rong Li; Zhaohui Hu; Qiang Liu; Ruogu Lu; Chengliang Yin
• Keywords: Bone cement; Percutaneous Vertebroplasty; Gradient boosting; Elastic net regularization

Background: Bone cement leakage is a common complication of percutaneous vertebroplasty and it could be life-threatening to some extent. The aim this study was develop machine learning model for predicting the risk in patients with osteoporotic vertebral compression fractures undergoing vertebroplasty. Furthermore, we developed an online calculator clinical application. Methods: This retrospective including 385 patients, who had fracture disease underwent surgery at Department Spine Surgery, Liuzhou People's Hospital from June 2016 2018. Combing patient's characteristics variables, applied six (ML) algorithms predictive models, logistic regression (LR), Gradient boosting (GBM), Extreme gradient (XGB), Random Forest (RF), Decision Tree (DT) Multilayer perceptron (MLP), which predict bone leakage. We tested results ten-fold cross-validation, calculated Area Under Curve (AUC) models selected highest AUC as excellent performing build web calculator. Results: showed that Injection volume cement, Surgery time Multiple were all independent predictors by using multivariate analysis observation subjects. Heatmap revealed relative proportions 15 variables. In prediction, ML ranged 0.633 0.898, while RF 0.898 used best Web ( https://share.streamlit.io/liuwencai0/pvp_leakage/main/pvp_leakage ) estimate each patient Conclusion: It achieved good prediction occurrence our model. concluded based on can help orthopedist make more individual rational strategies.