Développement d'un cadre pour l'évaluation éthique de l'intelligence artificielle en santé

Thesis Abstract

The rapid integration of artificial intelligence (AI) technologies into healthcare systems has raised critical ethical concerns, accentuating the urgent need for a comprehensive evaluative framework to ensure responsible AI deployment aligned with ethical principles and patient safety. This study aims to develop a robust, contextually adaptable framework for the ethical evaluation of AI applications in health settings, addressing the gaps in existing guidelines which often lack operational clarity and contextual specificity. To achieve this, the research is structured around three main objectives to identify key ethical considerations in AI healthcare tools, to formulate a multidimensional evaluative model incorporating stakeholder perspectives, and to validate the framework through empirical testing in clinical environments. A mixed-methods research design was employed, integrating qualitative and quantitative approaches. The qualitative component involved semi-structured interviews with 35 healthcare professionals, AI developers, bioethicists, and patient advocates, conducted across five major hospitals and AI development centers. Thematic analysis was applied to identify recurrent ethical concerns and contextual factors influencing AI decision-making processes. Concurrently, a quantitative survey was administered to 200 clinicians and AI engineers, utilizing a structured questionnaire designed to assess perceptions of ethical risks and the relevance of identified criteria. The data collected from both phases were integrated to inform the creation of the framework. The theoretical foundation of the study draws upon the Responsible Innovation Theory and the Ethical Principles of Biomedical Ethics (autonomy, beneficence, non-maleficence, and justice). These theories underpin the development of a multi-layered evaluative model, which encompasses technical, clinical, socio-cultural, and legal dimensions. To validate the framework, a pilot test was conducted focusing on three AI-powered diagnostic tools operating in cardiology, with a sample of 50 patients and corresponding healthcare providers. Data analysis involved structural equation modeling (SEM) to assess the framework's coherence and contextual applicability, alongside thematic analysis of stakeholder feedback to refine evaluative criteria. Expected findings include a set of clearly defined, operationalizable ethical assessment indicators tailored for AI healthcare applications, along with a structured decision-support matrix that guides stakeholders through ethical considerations during development, deployment, and ongoing monitoring. The study anticipates identifying significant correlations between stakeholder perceptions and the practical applicability of ethical criteria, thus emphasizing the importance of inclusive, contextual evaluation mechanisms. This research contributes substantially to the existing body of knowledge by bridging theoretical ethical principles with practical evaluation tools, offering a validated framework that supports ethically responsible AI implementation in health sectors. It advances the discourse by emphasizing stakeholder engagement and context-specific assessment, filling notable gaps in current guidelines that are often generic and difficult to operationalize. Furthermore, the developed framework provides a foundation for policymakers, healthcare providers, and AI developers to integrate ethical evaluation systematically into AI lifecycle management. The main conclusion drawn from this study is that a multidimensional, stakeholder-informed evaluative framework can significantly enhance the ethical robustness of AI applications in healthcare, promoting transparency, accountability, and trust. It recommends that health institutions adopt the framework for routine ethical assessment, expand validation across diverse clinical settings, and incorporate ongoing stakeholder feedback mechanisms. Future research should examine the longitudinal impact of applying this framework on AI-related ethical outcomes and explore its adaptation to emerging AI innovations such as personalized medicine and AI-driven health data management.