A new whitepaper explores reasons for the gap between research and practice in healthcare. Human factors, technical challenges and infrastructural aspects of artificial intelligence (AI) in the health domain are discussed to create a starting point for researchers, practitioners and policy makers.
The whitepaper is an initiative by VU Campus Center for AI & Health. Three institutes, Vrije Universiteit Amsterdam, Amsterdam UMC and Amsterdam University of Applied Sciences collaborate in multidisciplinary consortia to tackle AI challenges from all angles. The paper addresses the mechanisms behind current implementation barriers. Such collaboration is essential to create innovative, human-centred AI with tangible benefits for patients, relatives, healthcare professionals and society.
Potential to provide relief in healthcare
The societal need for health and healthcare innovation is clear. Rising demand and staff shortages put pressure on professionals, while AI can support decision-making and take over administrative tasks. ''With an ageing population, a record number of cancer diagnoses and increasing mental health issues, by 2050 nearly one in three Dutch workers would need to be in healthcare. AI can ease this burden on many levels'', Mark Hoogendoorn explains, AI scientist at VU Amsterdam, collaborating under the umbrella of VU Campus Center for AI and Health.
Success depends on human factors
The successful integration of AI in healthcare depends on a few key human factors: co-creation, AI knowledge and skills among healthcare professionals, and user accountability and acceptance of AI models. These factors directly affect the acceptance, usability and effectiveness of AI in healthcare and ultimately patient outcomes. Researchers stress AI as a complement, not a replacement. ''In breast cancer screening, for example, combining an AI system with the expert knowledge of a radiologist leads to improved accuracy as compared to either the individual radiologist or a standalone AI system'', says Hoogendoorn.
AI reliability beyond metrics
But the reliability of AI is challenged by dataset complexity, size and completeness, which limits generalisability. ''For me as a healthcare professional it is imminent that the AI model is trustworthy, robust and that I am able to understand why the model provides a certain output'', says Edwin Geleijn, physiotherapist at AUMC. Real-world evaluations therefore remain crucial.
Challenges in AI lifecycle
Beyond development, deployment and monitoring raise challenges such as legal issues, data sharing, technical integration and workflow alignment. Such as legal issues around the use of data and AI in clinical practice, data availability and sharing between institutions, technical implementation and integration in clinical workflow.
Collaborating for reliable AI in healthcare
Promising directions lie in both technical and organisational innovations: better AI development, improved data management, training for professionals, and practical monitoring. Addressing human, technological and infrastructural challenges is essential. The researchers look forward to advancing these innovations with the broader AI healthcare community.