Final document of the IMDRF AI/ML-enabled Working Group setting out 10 guiding principles for Good Machine Learning Practice (GMLP), intended to promote the development of safe, effective, and high-quality AI-enabled medical devices across the total product life cycle. The principles cover: understanding the intended use and leveraging multidisciplinary expertise; good software engineering, medical device design, and security practices; clinical evaluation using datasets representative of the intended patient population; independence of training and test datasets; fit-for-purpose reference standards; model choice and design tailored to the available data and intended use; assessment focused on human-AI interactions and the performance of the human-AI team rather than the device in isolation; testing under clinically relevant conditions; clear, essential information for users; and monitoring of deployed models and management of re-training risks. The document presents the principles as a call to action for standards organizations, regulators, and other bodies to further advance GMLP, and notes that generative AI may heighten their importance.

Edition 1 of the IMDRF Artificial Intelligence Medical Devices (AIMD) Working Group document on key terms and definitions for Machine Learning-enabled Medical Devices (MLMD). It establishes terms and definitions across the total product life cycle to promote consistency, support global harmonization, and provide a foundation for future guidelines on MLMD. Section 5 provides key definitions relevant to machine learning used in medical devices (including the definition of MLMD), Section 6 gives definitions from technical standards (e.g., bias, continuous learning, reference standard, reinforcement learning), and Section 7 discusses common machine learning terms. Most terms were previously defined in GHTF documents or internationally recognized AI standards, while some were developed or discussed by the AIMD Working Group. The document also notes that the term "bias" is used differently in data science and in legal discussions.