An automated deep learning pipeline for assessing aortic remodeling after frozen elephant trunk repair: a single-center pilot study

Abstract Computed tomography (CT) is the gold standard for assessing aortic remodeling following aortic dissection treatment. Clinical studies utilizing CT data are needed to evaluate the effectiveness of interventions such as the frozen elephant trunk (FET) procedure. However, large-scale studies are currently limited by the time-consuming and labor-intensive nature of manual image analysis. To address this, we developed an AI-based automated pipeline using open datasets to assess morphological changes in aortic dissection. We evaluated the pipeline’s efficacy using preoperative and postoperative CT scans from 14 patients who underwent FET repair. Two surgeons independently measured the aorta and true lumen manually. To separate segmentation error from plane selection error, one surgeon repeated the delineation on the plane selected by the pipeline. Because measurements were clustered within patients, agreement was assessed using linear mixed-effects models and repeated-measures Bland–Altman analysis. Agreement at a single time point was good to excellent (intraclass correlation coefficients: 0.90 to 0.95). On the identical plane, AI segmentation agreed closely with manual delineation (Dice: 0.951 for the aorta, 0.913 for the true lumen), with systematic bias arising largely from plane selection. This pipeline is feasible for future cohort-level research.