An empirical study on the use of AI for requirements engineering in Healthcare Projects

Abstract

This thesis addresses the challenge of managing complex and evolving requirements in healthcare software projects, a domain characterized by stringent regulatory demands, sensitive patient data, and diverse stakeholder perspectives. Traditional Requirements Engineering (RE) methods, reliant on manual analysis and subjective judgment, often struggle to capture, validate, and maintain high-quality requirements under these conditions. At the same time, advances in Artificial Intelligence (AI), particularly in Natural Language Processing (NLP), Machine Learning (ML), and Large Language Models (LLMs), offer promising tools for automating and enhancing RE tasks. Yet, the practical effectiveness and domain suitability of these AI techniques in healthcare remain underexplored. To bridge this gap, the study employs a mixed-methods approach. A systematic literature review synthesizes research on AI applications in healthcare RE, identifying key benefits such as automated requirement extraction, ambiguity detection, and classification, as well as limitations, including vocabulary mismatches and ethical concerns. Building on these findings, expert interviews and practitioner surveys were conducted with clinicians, administrators, developers, and regulatory specialists. These empirical investigations validated the literature insights, highlighting real-world barriers such as model interpretability, trust, and compliance risks, while revealing opportunities for domain-specific AI adaptation. Based on this dual evidence base, a comprehensive framework was developed. It advocates for AI models trained on healthcare-specific corpora, embedded ethical and legal filters, interactive feedback loops for stakeholder validation, and interdisciplinary collaboration among AI engineers, clinicians, and legal experts. Pilot guidelines and training strategies are proposed to support successful tool adoption. Initial evaluations suggest that this framework can reduce elicitation time, improve requirement precision, and strengthen compliance confidence, without sacrificing human oversight. Overall, this thesis advances both theory and practice by providing actionable strategies and a validated roadmap for integrating AI into Requirements Engineering in healthcare, ultimately aiming to enhance the safety, quality, and regulatory adherence of health IT systems.