AI-Assisted Carotid Ultrasound for Nonexpert Use in Primary Care Cardiovascular Prevention: User-Centered Iterative Development Study.
Authors
Affiliations (8)
Affiliations (8)
- Department of Nursing, Umeå University, Umea, Västerbotten, Sweden.
- Department of Diagnostics and Intervention, Umeå University, Umea, Västerbotten, Sweden.
- Västerbotten County, Umea, Västerbotten, Sweden.
- Department of Public Health and Clinical Medicine, Umeå University, Umea, Västerbotten, Sweden.
- Department of Epidemiology and Global Health, Umeå University, Umea, Västerbotten, Sweden.
- Umeå University, Umea, Västerbotten, Sweden.
- Department of Public health and Clinical Medicine, Umeå University, Umea, Västerbotten, Sweden.
- Department of Health and Medical Care, Region Norrbotten, Luleå, Norrbotten, Sweden.
Abstract
Carotid ultrasound is traditionally confined to specialist settings because examination quality and diagnostic reliability are highly dependent on the operator's technical skill and experience. Recent advances in AI-assisted ultrasound may enable task shifting to primary care staff and support the integration of personalized visual risk communication into routine preventive care. Visualization of subclinical atherosclerosis has been shown to strengthen cardiovascular risk communication and support preventive engagement. However, limited evidence exists on how such technology can be practically integrated into routine primary care workflows and what conditions are required to support use by nonexpert operators. This study aimed to describe an iterative user-centered design process enabling nonexpert use of AI-assisted carotid ultrasound for visualizing subclinical atherosclerosis in primary care cardiovascular disease prevention. We conducted a user-centered, iterative study within the Västerbotten Intervention Programme in northern Sweden. Fourteen nurse assistants from 4 primary care centers representing urban, rural, and remote rural settings tested an AI-assisted carotid ultrasound prototype with real-time guidance and artery segmentation in their primary care work environment, while 15 nurse assistants participated in semistructured interviews. Structured system-interaction observations were conducted during prototype testing. Interview data were analyzed using inductive thematic analysis informed by Braun and Clarke's approach, while data from structured system-interaction observations were analyzed pragmatically to identify usability and workflow issues relevant to the iterative development process. The thematic analysis of interview data identified 4 themes describing nurse assistants' experiences and conditions for implementation: technology that fits the everyday primary care work; system guidance as a prerequisite for effective use; practical, peer-supported learning for confidence in clinical use; and confidence in the patient encounter. Observations identified recurring usability challenges related to ergonomics, probe handling, interface navigation, image acquisition, and time pressure. These findings led to iterative refinements, including redesigned interface navigation, simplified data entry, added audio feedback, enhanced probe support, and revised training materials. This user-centered study identified key practical, organizational, and educational conditions for integrating AI-assisted carotid ultrasound into routine preventive care by nonexpert staff in primary care. The findings suggest that successful implementation will depend not only on technical capability but also on ergonomic and workflow fit, intuitive system guidance, peer-supported learning, and confidence in patient communication. These findings provide a foundation for future validation and implementation studies.