Lifecycle Management Framework for IVDR and EU AI Act Compliant Machine Learning Enabled Medical Device Software

Abstract

The application of Machine Learning (ML) in In Vitro Diagnostic (IVD) medical software presents significant possibilities for improving diagnostic accuracy, but also introduces additional regulatory hurdles within the European Union (EU) due to the interplay between the In Vitro Diagnostic Regulation (IVDR) and the AI Act. This thesis undertakes a comparative gap analysis of two legislative frameworks to identify and address legal differences and overlaps. The IVDR is governed by scientific validity, safety, and clinical performance, but lacks transparency regarding algorithmic bias and verifiable machine learning components. In contrast, the AI Act brings more structure by providing a comprehensive risk-based approach for high-risk AI systems. However, it lacks detailed clinical validation guidance within IVD boundaries. To mitigate these regulatory hurdles, this thesis proposes an integrated lifecycle management framework that incorporates AI Act requirements into the existing lifecycle framework, complying with the IVDR. The framework supports complete compliance with operability restrictions on the development of ML-based IVD medical software. This study makes a significant academic and practical contribution by laying the groundwork for an open-access regulatory framework that enables startup companies to navigate complex compliance requirements efficiently.