Introduction Aortic aneurysms, particularly abdominal aortic aneurysms (AAA), represent a major source of cardiovascular morbidity and mortality. Early detection, precise segmentation for volumetric assessment, and vigilant surveillance for endoleaks following endovascular aneurysm repair (EVAR) are critical for improving outcomes. Artificial intelligence (AI), including deep learning and radiomics, has demonstrated promise in automating these tasks, potentially reducing radiologist workload and enhancing accuracy. This systematic review evaluates the role of AI across the aortic aneurysm care continuum: detection, segmentation, and postoperative surveillance. Methods We systematically reviewed retrospective cohorts, multicenter validations, and pilot clinical trials evaluating AI applications in aortic aneurysm management. Data were extracted on sensitivity, specificity, Dice similarity coefficient (DSC), area under the receiver operating characteristic curve (AUC), and computational efficiency. Outcomes were stratified into three domains: (1) aneurysm detection, (2) aneurysm segmentation for volumetric monitoring, and (3) endoleak detection post‐EVAR. Pooled descriptive analysis was performed to benchmark AI performance across modalities (CTA, MRA, ultrasound). Results AI‐assisted detection achieved high diagnostic performance, with pooled sensitivity and specificity >90% for AAA detection on CT angiography. Radiomics‐based classifiers demonstrated superior discrimination of aneurysm presence compared to diameter‐based thresholds, achieving AUCs of 0.93‐0.96. Segmentation studies employing U‐Net and 3D CNN architectures achieved DSCs ranging 0.85‐0.92 internally and ∼0.80 on external datasets, enabling accurate volumetric follow‐up and growth assessment. Processing times varied from 30 seconds for lightweight CNNs to 2‐3 minutes for more complex attention‐based networks. Endoleak surveillance represented an emerging domain: preliminary DL models trained on post‐EVAR CTA achieved sensitivities of 0.88‐0.91 and specificities of 0.85‐0.90, with notable challenges in differentiating slow‐flow endoleaks from postoperative artifacts. Across studies, pooled analysis showed a modest performance decline in external validations (sensitivity drop ∼4%, DSC decline ∼0.05), underscoring the importance of multi‐center training. AI integration consistently reduced radiologist reading times and improved reproducibility of measurements, with volumetric growth detection outperforming traditional diameter‐based monitoring. Conclusion AI has demonstrated strong potential across the management spectrum of aortic aneurysms, achieving high accuracy in detection, robust segmentation for volumetric follow‐up, and promising results in endoleak surveillance. While internal validations report excellent benchmarks, modest declines on external testing highlight the need for harmonized datasets and prospective multicenter validation. Integration of AI into clinical workflows could enhance early detection, personalize surveillance, and improve long‐term outcomes in patients with aortic aneurysms. Future work should focus on explainable AI models and standardized reporting to facilitate widespread adoption.
M.M. El-Sayed (2025) studied this question.