User-Centered Iterative Development of AI-Guided Carotid Ultrasound for Cardiovascular Prevention in Primary Care by Nonexperts

By
Anna Sjöström
Christer Grönlund
Karolina Jonzén
Patrik Wennberg
Åsa Hörnsten
Thorbjörn Lundberg
Anni-Maria Pulkki-Brännström
Luisa Ebeling
Albin Dahlin Almevall
October 8, 2026

Journal Of Medical Internet Research

Overview

The article describes the rationale for developing AI-guided carotid ultrasound for use by nonexperts in primary care, where conventional examinations are operator-dependent and commonly performed by specialized personnel. Prior research cited in the article suggests automated cIMT measurement can be reproducible with novice operators, while plaque detection remains variable; evidence on integrating AI plaque tools into routine workflows is limited. Current guidance discourages population screening for asymptomatic carotid stenosis, so the potential role of these tools must be distinguished from routine screening.

Background

Carotid ultrasound quality and diagnostic reliability depend on operator expertise, which can limit access outside specialized settings. AI systems may provide real-time guidance and image analysis, potentially supporting task shifting, but implementation also depends on training, organizational support, and user confidence. Automated cIMT measurement has shown reproducibility with novice operators, whereas manual plaque detection has substantial interobserver variability, particularly for smaller plaques. The source notes that evidence remains limited on the use of AI-assisted plaque assessment by nonexperts in routine clinical workflows.

Data Highlights

The supplied article context contains no numerical results for the user-centered development study. The related-article context reports that, in a pilot cardiac ultrasound study, cardiologists agreed with advanced practice providers' interpretations of major findings in more than 96% of examinations; AI-guided examinations required fewer cardiologist corrections than examinations without real-time AI guidance. No sample size or further numerical outcomes are provided.

Key Findings

Carotid ultrasound is operator-dependent, and in Sweden examinations are generally conducted in hospitals by highly trained biomedical scientists.
AI-assisted systems may provide automated scanning guidance and real-time image analysis, potentially enabling nonexpert users to perform clinically relevant examinations.
Prior research cited in the article found automated cIMT measurements remained reproducible when obtained by novice operators.
Manual carotid plaque detection has substantial intersonographer variability, particularly for smaller plaques; automated plaque detection and segmentation are less established than automated cIMT assessment.
The article identifies limited evidence on integrating AI-assisted carotid ultrasound into routine workflows and on its use by nonexpert operators.
Current guidance in the supplied web context discourages screening asymptomatic adults for carotid artery stenosis; the USPSTF recommends against it, and the 2024 AHA/ASA guideline classifies routine screening as not recommended.

Clinical Implications

AI guidance could support task shifting of carotid ultrasound to primary care, but the source identifies training, organizational support, and user confidence as implementation requirements. The supplied guidance context does not support using such technology to conduct population screening for asymptomatic carotid stenosis.

Conclusion

AI-guided carotid ultrasound is being explored as a way to support nonexpert acquisition and improve consistency, particularly in light of operator dependence and variability in plaque detection. The supplied material describes a rationale and evidence gaps, rather than establishing routine clinical effectiveness or a role in population screening.