Co-designing an AI-assisted carotid ultrasound method for non-expert use in primary care: The PRIMVIZA project

Background Effective cardiovascular disease (CVD) prevention in primary care requires communication strategies that make individual risk understandable. Visualising subclinical atherosclerosis can enhance motivation for lifestyle change. In the VIPVIZA randomised trial, personalised pictorial feedback from expert-performed carotid ultrasound improved lifestyle, long-term risk factor trajectories, and adherence to preventive medication. However, ultrasound remains operator- and site-dependent, limiting use in primary care. Advances in AI-assisted ultrasound create opportunities for task shifting, but successful implementation requires methods developed with intended users. PRIMVIZA aims to translate the VIPVIZA approach into primary care by developing an AI-supported carotid ultrasound method anchored in the on-going Västerbotten Intervention Programme (VIP) in northern Sweden. Objective To describe an iterative process of co-designing methods for use of an AI-assisted, non-expert-dependent carotid ultrasound system for visualisation of subclinical atherosclerosis, aimed at improving CVD prevention in primary care. Methods A tailored user-centred co-design process was conducted at four primary care centres representing urban, rural, and remote settings. A multidisciplinary research team contributed expertise in medicine, nursing, medical engineering, and health economics. Iterative cycles included prototype testing, structured system-interaction observations, and semi-structured interviews with nurse assistants (n = 14) who conduct standardised VIP assessments. Feedback targeted training needs, workflow alignment, probe handling, interface clarity, ergonomics, and role perceptions. Findings from each cycle guided revisions to the hardware setup, software logic, graphical interface, and basic technical training. Results Nurse assistants found the prototype intuitive, and AI-generated guidance supportive for identifying the carotid artery. Early iterations revealed challenges related to probe grip, image stability, readiness indicators, and equipment positioning. Participants highlighted the need for concise instructions and varied training formats and described the prospect of new responsibilities and a more entrusted role as stimulating and professionally developing. They also assessed the system as feasible to integrate into existing workflows. Feedback informed iterative adjustments to the interface, ergonomics, and training content, addressing recurring barriers and preparing the prototype for patient-based testing. Conclusion An iterative co-design process enabled systematic refinement of an AI-assisted carotid ultrasound prototype for non-expert use in primary care. The system and training package are now ready for feasibility testing in real patient examinations within the VIP setting.