Description: With enormous increase in data processing capabilities, more and more algorithms use AI/ML in their implementation. AI/ML based software are used vastly in medical device industry. Today CE marking of AI/ML based software medical device is according to the requirements of EU MDR which requests manufacturer to submit a technical documentation and be prepared for the audit of their QMS. During this presentation, I will talk about notified body expectations from manufacturer of medical devices with AI/ML based software during an audit. I will talk about necessity of having a robust Risk management system that is imposed by General Safety and Performance Requirements (GSPR) 3 of the MDR. I will talk about management of risks related to AI systems based on the state of the art (ISO 14971: 2019) and analysis of software that can contribute to dangerous situations and discuss 7.1 to 7.3 of IEC 62304 linked to the audit of AI systems. I will discuss 3 elements that manufacturer of AI/ML based software should expect to be questioned during MDR audit of their devices: First explainability of their AI , second Verification and validation of their system and finally Quality of their data . For each section I will present requirements/risks that manufacturer should consider such as risks associated with the AI model, risk of not being able to approve the AI decision or risks of not identifying falls decisions. I will discuss whether the algorithm evaluation platform has correctly been validated. Finally if data represent target patient population, metrics used for measuring AI performance, labeling and acceptance value for labeling . I will discuss cyber attack and its impact on data corruption if not detected. I will ask the participants to discuss their implementation of 3 elements and provide their experience during audit of their systems.
Learning Objectives:
learn about important points to consider during MDR Audit of AI/ML based software medical device
learn about importance of implementing risk management as the base of AI systems used as part of medical device or as part of software as medical device
Discuss risks associated with AI systems related to AI models, algorithm evaluation and data quality