Validation study comparing Artificial intelligence for fully automatic aortic aneurysms Segmentation and diameter Measurements On contrast and non-contrast enhanced computed Tomography (ASMOT).
Authors
Affiliations (5)
Affiliations (5)
- Department of Vascular and Endovascular Surgery, Besançon University Hospital, Besançon, France. Electronic address: [email protected].
- Department of Vascular Surgery, Bordeaux University Hospital, Bordeaux, France. Electronic address: [email protected].
- Nurea, F-33000, Bordeaux, France. Electronic address: [email protected].
- Department of Vascular and Endovascular Surgery, Besançon University Hospital, Besançon, France; Université de Franche-Comté, SINERGIES, F-25000 Besançon, France. Electronic address: [email protected].
- Department of Vascular and Endovascular Surgery, Besançon University Hospital, Besançon, France; Université de Franche-Comté, SINERGIES, F-25000 Besançon, France. Electronic address: [email protected].
Abstract
Accurate aortic diameter measurements are essential for diagnosis, surveillance, and procedural planning in aortic disease. Semi-automatic methods remain widely used but require manual corrections, which can be time-consuming and operator-dependent. Artificial intelligence (AI)-driven fully automatic methods may offer improved efficiency and measurement accuracy. This study aims to validate a fully automatic method against a semi-automatic approach using computed tomography angiography (CTA) and non-contrast CT scans. A monocentric retrospective comparative study was conducted on patients who underwent endovascular aortic repair (EVAR) for infrarenal, juxta-renal or thoracic aneurysms and a control group. Maximum aortic wall-to-wall diameters were measured before and after repair using a fully automatic software (PRAEVAorta2®, Nurea, Bordeaux, France) and compared to measurements performed by two vascular surgeons using a semi-automatic approach on CTA and non-contrast CT scans. Correlation coefficients (Pearson's R) and absolute differences were calculated to assess agreement. A total of 120 CT scans (60 CTA and 60 non-contrast CT) were included, comprising 23 EVAR, 4 thoracic EVAR, 1 fenestrated EVAR, and 4 control cases. Strong correlations were observed between the fully automatic and semi-automatic measurements in both CTA and non-contrast CT. For CTA, correlation coefficients ranged from 0.94 to 0.96 (R<sup>2</sup> = 0.88-0.92), while for non-contrast CT, they ranged from 0.87 to 0.89 (R<sup>2</sup> = 0.76-0.79). Median absolute differences in aortic diameter measurements varied between 1.1 mm and 4.2 mm across the different anatomical locations. The fully automatic method demonstrated a significantly faster processing time, with a median execution time of 73 seconds (IQR: 57-91) compared to 700 (IQR: 613-800) for the semi-automatic method (p < 0.001). The fully automatic method demonstrated strong agreement with semi-automatic measurements for both CTA and non-contrast CT, before and after endovascular repair in different aortic locations, with significantly reduced analysis time. This method could improve workflow efficiency in clinical practice and research applications.