Performance assessment of an artificial intelligence algorithm for opportunistic screening of abdominal aortic aneurysms

Abstract

Purpose

Abdominal aortic aneurysm (AAA) is a common incidental finding on CT imaging performed in the acute care setting. Artificial intelligence (AI) algorithms have been developed to automatically measure aortic lumen size and thus facilitate AAA detection. However, few studies have evaluated the performance of such tools in a large clinical setting. This retrospective study aimed to evaluate the performance of a commercially-available AI algorithm for the opportunistic screening of incidental AAA on non-optimized CT imaging.

Methods

CT examinations of the abdomen and pelvis performed in the emergency setting of a tertiary academic center between July 2020 and May 2021 were retrospectively processed by the AI algorithm, while natural language processing software (NLP) was used to analyze the initial radiology report. Exams which were positive for the presence of AAA on imaging by AI analysis, but negative by NLP of their corresponding report, were designated as potential discrepancies and independently reviewed by an ED radiologist.

Results

4023 abdominal and pelvic CT examinations were analyzed. 98.3 % (3955) cases were negative for presence of AAA by NLP assessment of their respective report, with 16 of these cases flagged by AI as discrepancies potentially positive for AAA. 31 % (5/16) of these cases were determined by secondary review to be truly positive for previously undocumented AAA. The enhanced detection rate with AI assistance was 7.4 %.

Discussion

Artificial intelligence algorithms demonstrate the potential to improve detection rates of incidental abdominal aortic aneurysms on CT imaging, particularly in high throughput workflows such as the emergency department.