Fully automatic volume segmentation using deep learning approaches to assess aneurysmal sac evolution after infrarenal endovascular aortic repair

J Vasc Surg. 2022 Sep;76(3):620-630.e3. doi: 10.1016/j.jvs.2022.03.891. Epub 2022 May 23.

Abstract

Objective: Endovascular aortic repair (EVAR) surveillance relies on serial measurements of the maximal diameter despite significant inter- and intraobserver variability. Volumetric measurements are more sensitive; however, their general use has been hampered by the time required for their implementation. An innovative, fully automated software (PRAEVAorta; Nurea, Bordeaux, France), using artificial intelligence, had previously demonstrated fast and robust detection of the characteristics of infrarenal abdominal aortic aneurysms on preoperative imaging studies. In the present study, we assessed the robustness of these data on post-EVAR computed tomography (CT) scans.

Methods: We compared fully automatic and semiautomatic segmentation manually corrected by a senior surgeon (E.D.) using a dataset of 48 patients (48 early post-EVAR CT scans with 6466 slices and 101 follow-up CT scans with 13,708 slices).

Results: The analyses confirmed the excellent correlation of the post-EVAR volumes and surfaces and the proximal neck and maximum aneurysm diameters measured using the fully automatic and manually corrected segmentation methods (Pearson's coefficient correlation, >0.99; P < .0001). A comparison between the fully automatic and manually corrected segmentation methods revealed a mean Dice similarity coefficient of 0.950 ± 0.015, Jaccard index of 0.906 ± 0.028, sensitivity of 0.929 ± 0.028, specificity of 0.965 ± 0.016, volumetric similarity of 0.973 ± 0.018, and mean Hausdorff distance/slice of 8.7 ± 10.8 mm. The mean volumetric similarity reached 0.873 ± 0.100 for the lumen and 0.903 ± 0.091 for the thrombus. The segmentation time was nine times faster with the fully automatic method (2.5 minutes vs 22 minutes per patient with the manually corrected method; P < .0001). A preliminary analysis also demonstrated that a diameter increase of 2 mm can actually represent a >5% volume increase.

Conclusions: PRAEVAorta enabled a fast, reproducible, and fully automated analysis of post-EVAR abdominal aortic aneurysm sac and neck characteristics, with a comparison between different time points. It could become a crucial adjunct for EVAR follow-up through the early detection of sac evolution, which might reduce the risk of secondary rupture.

Keywords: Abdominal aortic aneurysm; Artificial intelligence; Automatic segmentation; Deep learning; Endovascular aneurysm repair; Thrombus; Volume.

MeSH terms

  • Aorta, Abdominal / surgery
  • Aortic Aneurysm, Abdominal* / complications
  • Aortic Aneurysm, Abdominal* / diagnostic imaging
  • Aortic Aneurysm, Abdominal* / surgery
  • Artificial Intelligence
  • Blood Vessel Prosthesis Implantation* / adverse effects
  • Blood Vessel Prosthesis Implantation* / methods
  • Deep Learning*
  • Endovascular Procedures* / adverse effects
  • Humans
  • Retrospective Studies
  • Treatment Outcome