Automated Morphological Phenotyping Identifies Anatomic Patterns Associated with Early Type I Endoleak after Endovascular Aortic Repair.

Abstract

Early type I endoleak (T1EL) following endovascular aortic repair (EVAR) remains a clinically important complication. Risk stratification is traditionally based on binary morphological thresholds, but their ability to predict early failure is limited. This study aimed to use fully automated volume segmentation (FAVS) to identify anatomic phenotypes associated with early T1EL. This was a multicentre, retrospective cohort study. Pre-operative computed tomography angiography of patients undergoing EVAR was analysed using FAVS, with matched National Vascular Registry clinical data. Unsupervised Gaussian mixture modelling was used to identify anatomic phenotypes. The primary outcome was early (within thirty days) T1EL. Phenotypes were compared for endoleak rates and distribution of hostile anatomy. Performance of traditional hostile anatomy constructs was evaluated for comparison and was assessed using receiver operating characteristic and area under the curve analysis (AUC). Among 1 003 patients, early T1EL occurred in 62 (6.2%). Traditional hostile anatomy constructs demonstrated poor discrimination for early T1EL (AUC 0.533). Six distinct morphological phenotypes were identified, with significantly different early T1EL rates (4.2 - 21.6%; p = .001). All phenotypes contained anatomies classified as both hostile and non-hostile by conventional criteria (38.9 - 83.4%), with significant differences in neck thrombus and calcification burden (p < .001). Overall model discrimination was modest (AUC 0.697, 95% confidence interval 0.632 - 0.761). High dimensional morphological phenotyping using FAVS identifies anatomic patterns associated with early failure, supporting a move toward morphology driven EVAR planning.