A Systematic Review and Bibliometric Analysis of Applications of Artificial Intelligence and Machine Learning in Vascular Surgery

Abstract

Background: Artificial intelligence (AI) and machine learning (ML) have seen increasingly intimate integration with medicine and healthcare in the last 2 decades. The objective of this study was to summarize all current applications of AI and ML in the vascular surgery literature and to conduct a bibliometric analysis of published studies.

Methods: A comprehensive literature search was conducted through Embase, MEDLINE, and Ovid HealthStar from inception until February 19, 2021. Reporting of this study was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Title and abstract screening, full-text screening, and data extraction were conducted in duplicate. Data extracted included study metadata, the clinical area of study within vascular surgery, type of AI/ML method used, dataset, and the application of AI/ML. Publishing journals were classified as having either a clinical scope or technical scope. The author academic background was classified as clinical, nonclinical (e.g., engineering), or both, depending on author affiliation.

Results: The initial search identified 7,434 studies, of which 249 were included for a final analysis. The rate of publications is exponentially increasing, with 158 (63%) studies being published in the last 5 years alone. Studies were most commonly related to carotid artery disease (118, 47%), abdominal aortic aneurysms (51, 20%), and peripheral arterial disease (26, 10%). Study authors employed an average of 1.50 (range: 1-6) distinct AI methods in their studies. The application of AI/ML methods broadly related to predictive models (54, 22%), image segmentation (49, 19.4%), diagnostic methods (46, 18%), or multiple combined applications (91, 37%). The most commonly used AI/ML methods were artificial neural networks (155/378 use cases, 41%), support vector machines (64, 17%), k-nearest neighbors algorithm (26, 7%), and random forests (23, 6%). Datasets to which these AI/ML methods were applied frequently involved ultrasound images (87, 35%), computed tomography (CT) images (42, 17%), clinical data (34, 14%), or multiple datasets (36, 14%). Overall, 22 (9%) studies were published in journals specific to vascular surgery, with the majority (147/249, 59%) being published in journals with a scope related to computer science or engineering. Among 1,576 publishing authors, 46% had exclusively a clinical background, 48% a nonclinical background, and 5% had both a clinical and nonclinical background.

Conclusions: There is an exponentially growing body of literature describing the use of AI and ML in vascular surgery. There is a focus on carotid artery disease and abdominal aortic disease, with many other areas of vascular surgery under-represented. Neural networks and support vector machines composed most AI methods in the literature. As AI/ML continue to see expanded applications in the field, it is important that vascular surgeons appreciate its potential and limitations. In addition, as it sees increasing use, there is a need for clinicians with expertise in AI/ML methods who can optimize its transition into daily practice.