Applying patient characteristics, stent-graft selection, and pre-operative computed tomographic angiography data to a machine learning algorithm: Is endoleak prediction possible?

Introduction: This study aims to predict endoleak after endovascular aneurysm repair (EVAR) using machine learning (ML) integration of patient characteristics, stent-graft configuration, and a selection of vessel lengths, diameters and angles measured using pre-operative computed tomography angiography (CTA).

Methods: We evaluated 1-year follow-up CT scans (arterial and delayed phases) in patients who underwent EVAR for the presence or absence of an endoleak. We also obtained data on the patient characteristics, stent-graft selection, and preoperative CT vessel morphology (diameter, length, and angle). The extreme gradient boosting (XGBoost) for the ML system was trained on 30 patients with endoleaks and 81 patients without. We evaluated 5217 items in 111 patients with abdominal aortic aneurysms, including the patient characteristics, stent-graft configuration and vascular morphology acquired using pre-EVAR abdominal CTA. We calculated the area under the curve (AUC) of our receiver operating characteristic analysis using the ML method.

Results: The AUC, accuracy, 95% confidence interval (CI), sensitivity, and specificity were 0.88, 0.88, 0.79-0.97, 0.85, and 0.91 for ML applying XGBoost, respectively.

Conclusions: The diagnostic performance of the ML method was useful when factors such as the patient characteristics, stent-graft configuration and vessel length, diameter and angle of the vessels were considered from pre-EVAR CTA. Implications for practice: Based on our findings, we suggest that this is a potential application of ML for the interpretation of abdominal CTA scans in patients with abdominal aortic aneurysms scheduled for EVAR.