Trustworthy Artificial Intelligence in Healthcare: a proposed framework

Trustworthy Artificial Intelligence (TwAI) is not sufficiently defined within healthcare to enable legal manufacturers and developers to identify and operationalise, with a high degree of assurance, the General Safety and Performance Requirements. This includes patients’ fundamental rights which are necessary to meet European regulatory requirements and stakeholder expectations. Medical Device Software developers are tasked with identifying appropriate performance measures for robustness of Artificial Intelligence-enabled Medical Devices based on the use case and risks identified by the development team. Risk identification is heavily reliant on the adequacy of their risk management process and the competency of personnel. As medical device software defects continue to be identified in the post-market phase, further research and guidance is necessary. Medical device manufacturers have reported a lack of understanding and harmonisation of the requirements that make up TwAI in healthcare, starting with ethical requirements. With standards development still in progress, research is necessary to ensure a common framework is applied, having appropriate development processes, methods, and measures for delivery of TwAI in healthcare. Therefore, we conducted a review of trust for two reasons: 1) to attempt to gain a collective understanding of TwAI in healthcare; 2) to propose a framework that considers the regulatory landscape for Artificial Intelligence-enabled Medical Devices that is sufficiently detailed to move beyond theoretical concepts. This research brings together the state of the art for artificial intelligence in healthcare for a harmonised approach and extends the literature by presenting a proposed framework to be operationalised by legal manufacturers and developers who place medical devices on the market for clinical use.