Artificial Intelligence-Assisted Sac Diameter Assessment for Complex Endovascular Aortic Repair

Moritz Wegner; Vincent Fontaine; Petroula Nana; Bryan V Dieffenbach; Dominique Fabre; Stéphan Haulon

doi:10.1177/15266028231208159

Artificial Intelligence-Assisted Sac Diameter Assessment for Complex Endovascular Aortic Repair

J Endovasc Ther. 2023 Oct 30:15266028231208159. doi: 10.1177/15266028231208159. Online ahead of print.

Authors

Moritz Wegner¹, Vincent Fontaine², Petroula Nana², Bryan V Dieffenbach³, Dominique Fabre², Stéphan Haulon²

Affiliations

¹ Department of Vascular and Endovascular Surgery, Faculty of Medicine, University Hospital Cologne, University of Cologne, Cologne, Germany.
² Aortic Center, Marie Lannelongue Hospital, Groupe Hospitalier Paris Saint Joseph, Paris-Saclay University, Le Plessis-Robinson, France.
³ Division of Vascular and Endovascular Surgery, Brigham and Women's Hospital, Boston, MA, USA.

PMID: 37902445
DOI: 10.1177/15266028231208159

Abstract

Purpose: Artificial intelligence (AI) using an automated, deep learning-based method, Augmented Radiology for Vascular Aneurysm (ARVA), has been verified as a viable aide in aneurysm morphology assessment. The aim of this study was to evaluate the accuracy of ARVA when analyzing preoperative and postoperative computed tomography angiography (CTA) in patients managed with fenestrated endovascular repair (FEVAR) for complex aortic aneurysms (cAAs).

Materials and methods: Preoperative and postoperative CTAs from 50 patients (n=100 CTAs) who underwent FEVAR for cAAs were extracted from the picture archiving and communication system (PACS) of a single aortic center equipped with ARVA. All studies underwent automated AI aneurysm morphology assessment by ARVA. Appropriate identification of the outer wall of the aorta was verified by manual review of the AI-generated overlays for each patient. Maximum outer-wall aortic diameters were measured by 2 clinicians using multiplanar reconstruction (MPR) and curved planar reformatting (CPR), and among studies where the aortic wall was appropriately identified by ARVA, they were compared with ARVA automated measurements.

Results: Identification of the outer wall of the aorta was accurate in 89% of CTA studies. Among these, diameter measurements by ARVA were comparable to clinician measurements by MPR or CPR, with a median absolute difference of 2.4 mm on the preoperative CTAs and 1.6 mm on the postoperative CTAs. Of note, no significant difference was detected between clinician measurements using MPR or CPR on preoperative and postoperative scans (range 0.5-0.9 mm).

Conclusion: For patients with cAAs managed with FEVAR, ARVA provides accurate preoperative and postoperative assessment of aortic diameter in 89% of studies. This technology may provide an opportunity to automate cAA morphology assessment in most cases where time-intensive, manual clinician measurements are currently required.

Clinical impact: In this retrospective analysis of preoperative and postoperative imaging from 50 patients managed with FEVAR, AI provided accurate aortic diameter measurements in 89% of the CTAs reviewed, despite the complexity of the aortic anatomies, and in post-operative CTAs despite metal artifact from stent grafts, markers and embolization materials. Outliers with imprecise automated aortic overlays were easily identified by scrolling through the axial AI-generated segmentation MPR cuts of the entire aorta.This study supports the notion that such emerging AI technologies can improve efficiency of routine clinician workflows while maintaining excellent measurement accuracy when analyzing complex aortic anatomies by CTA.

Keywords: artificial intelligence; complex aortic repair; fenestrated endovascular aortic repair; imaging; machine learning; patient outcome prediction.