Volume Measurements for Surveillance after Endovascular Aneurysm Repair using Artificial Intelligence.

Objective: Surveillance after endovascular aneurysm repair (EVAR) is suboptimal due to limited compliance and relatively large variability in measurement methods of abdominal aortic aneurysm (AAA) sac size after treatment. Measuring volume offers a more sensitive early indicator of aneurysm sac growth or regression and stability, but is more time consuming and thus less practical than measuring maximum diameter. This study evaluated the accuracy and consistency of the artificial intelligence (AI) driven software PRAEVAorta 2 and compared it with an established semi-automated segmentation method.

Methods: Post-EVAR aneurysm sac volumes measured by AI were compared with a semi-automated segmentation method (3mensio software) in patients with an infrarenal AAA, focusing on absolute aneurysm volume and volume evolution over time. The clinical impact of both methods was evaluated by categorising patients as showing either AAA sac regression, stabilisation, or growth comparing the 30 day and one year post-EVAR computed tomography angiography (CTA) images. Inter- and intra-method agreement were assessed using Bland-Altman analysis, the intraclass correlation coefficient (ICC), and Cohen's κ statistic.

Results: Forty nine patients (98 CTA images) were analysed, after excluding 15 patients due to segmentation errors by AI owing to low quality CT scans. Aneurysm sac volume measurements showed excellent correlation (ICC = 0.94, 95% confidence interval [CI] 0.88 - 0.99) with good to excellent correlation for volume evolution over time (ICC = 0.85, 95% CI 0.75 - 0.91). Categorisation of AAA sac evolution showed fair correlation (Cohen's κ = 0.33), with 12 discrepancies (24%) between methods. The intra-method agreement for the AI software demonstrated perfect consistency (bias = -0.01 cc), indicating that it is more reliable compared with the semi-automated method.

Conclusion: Despite some differences in AAA sac volume measurements, the highly consistent AI driven software accurately measured AAA sac volume evolution. AAA sac evolution classification appears to be more reliable than existing methods and may therefore improve risk stratification post-EVAR, and could facilitate AI driven personalised surveillance programmes. While high quality CTA images are crucial, considering radiation exposure is important, validating the software with non-contrast CT scans might reduce the radiation burden.