Prediction of persistent type II endoleak after endovascular aortic repair using machine learning based on preoperative clinical data and radiomic

BACKGROUND: Persistent type II endoleak (T2EL) is associated with adverse outcomes following Endovascular aneurysm repair (EVAR) of abdominal aortic aneurysms. PURPOSE: The purpose of this study was to explore the feasibility of predicting persistent T2ELs after EVAR using preoperative clinical data and random forest derived from computed tomography angiography (CTA). MATERIALS AND METHODS: A retrospective study was performed on patients who underwent EVAR from January 2019 to June 2023. Based on the postoperative CTA, patients were divided into different groups according to the existence of persistent T2EL. Preoperative clinical data were collected, and radiomic features were extracted from the segmented thrombus in the aneurysm sac of preoperative CTA. Feature selection was performed before machine learning model training. Six common machine learning algorithms were used to predict persistent T2ELs based on preoperative features. Model performance was compared and evaluated. RESULTS: Among the initial 1006 preoperative clinical and radiomic features, 12 features were selected for machine learning model development. The support vector machines (SVMs) classifier performed well in both the training and test sets, with area under the curve values of 0.994 (95% confidence interval [CI], 0.967–1) and 0.970 (95% CI, 0.901–1), the sensitivity of 0.927 and 0.882, specificity of 0.979 and 0.920, and accuracy of 0.966 and 0.848, respectively. Based on the DeLong test and decision curve analysis, we concluded that the SVM model had superior predictive performance and clinical applicability. CONCLUSIONS: Machine learning algorithms utilizing preoperative features may predict persistent T2EL after EVAR.