Constructs Influencing Patient Perceptions of Use of AI in Medical Imaging Analysis: Systematic Review.
Authors
Affiliations (2)
Affiliations (2)
- Frazer Institute, Faculty of Health, Medicine and Behavioural Sciences, The University of Queensland, Brisbane, Queensland, Australia.
- Centre for Online Health, The University of Queensland, Brisbane, Queensland, Australia.
Abstract
The use of artificial intelligence (AI) in medical imaging has been growing exponentially. Understanding patient perceptions and factors influencing their views of AI is critical to develop adequate strategies to support implementation and acceptance. This study aims to investigate the constructs that influence patients' perceptions and acceptance of AI's use in the analysis of their medical images to support screening and diagnosis. A systematic review was conducted to meet the research objective. Relevant articles were found by searching 5 databases. Data were extracted using an iteratively refined framework and synthesized narratively due to heterogeneity in study designs, populations, health care contexts, and outcomes. A total of 59 relevant studies were included in the review. Patient acceptance of AI in medical image analysis emerged from multiple interacting factors. The most consistently reported determinant in 48 studies was that AI implementation should prioritize human-in-the-loop models, positioning AI as supportive tools, working in conjunction with health care providers rather than as an autonomous decision-maker. Other factors identified were performance of the AI, clarity of accountability, trust, and ethical factors. Patients' individual characteristics such as demographics and health history were also noted to influence acceptance indirectly. The review findings were used to draft a conceptual model to draw attention to the complex relationship among the identified factors. This review informed the development of a conceptual model illustrating the complex and interactive factors shaping patient acceptance of AI in medical imaging, which can be tested prospectively in future studies. Our results highlight that patients' likelihood of accepting AI cannot be attributed to a few factors. Instead, promoting acceptance will require a holistic approach where multiple factors are considered simultaneously and adapted for each use case.