Applications for Artificial Intelligence in Ultrasound Education: A Scoping Review (Preprint)

Abstract

<sec> <title>BACKGROUND</title> Many medical students and residents undergo dedicated ultrasound training as part of formal curricula. Yet, institutions face several barriers to successful implementation of ultrasound training, including shortages of credentialed faculty. In recent years, artificial intelligence (AI) has been used to guide the acquisition and interpretation of ultrasound images. </sec> <sec> <title>OBJECTIVE</title> This scoping review therefore aimed to identify the types of AI interventions that have been used to support ultrasound skill acquisition by undergraduate and graduate medical trainees and evaluate the impact of AI tools on educational outcomes such as trainee performance, confidence, and learning efficiency. Secondarily, this study also assessed the challenges and limitations of current AI technologies in ultrasound education. </sec> <sec> <title>METHODS</title> A literature search of Medline-OVID, Web of Science, and PubMed was conducted to inform this scoping review, with eligibility criteria outlined as follows: • Population: Must involve undergraduate or graduate medical trainees • Intervention: Must involve an AI intervention in which trainees directly interacted and utilized AI for purposes such as ultrasound skill teaching, diagnosis, structure detection, image acquisition, etc. • Comparator: No comparator group, expert, non-AI, before and after, etc. • Outcomes: Any outcome reported in the literature • Exclusion: Studies that describe development or validation of AI technologies without directly assisting learning of ultrasound by undergraduate or graduate medical trainees 181 total citations were originally identified across the three searched databases with 31 studies deemed eligible for inclusion in the review. A data collection sheet was designed to collect key data including author(s), year of publication, study location (country), study design, study aims, type of AI intervention, categories of purpose (diagnosis, anatomy recognition), participants, testing set, comparator, and main findings. </sec> <sec> <title>RESULTS</title> AI had a positive impact on trainees’ ability to diagnose multiple pathologies, improved scan quality, and efficiently detected anatomical structures. AI also helped trainees achieve high diagnostic agreement with that of experts. Varied results were found from studies comparing AI-assisted ultrasound instruction to other training methods in terms of trainees’ diagnostic performance, confidence levels, scan time, and trust in AI. </sec> <sec> <title>CONCLUSIONS</title> The majority of articles included in this review showcased AI’s potential to enhance the capabilities of less experienced ultrasound trainees. In its current state, AI may prove to be a valuable complement to traditional ultrasound curricula, but widespread implementation of AI into ultrasound curricula and practice requires further research on AI’s impact on long-term ultrasound skill retention and applications across diverse patient populations and pathologies. </sec>

Ähnliche Arbeiten