AI/machine learning based-medical software: case studies on model development and regulatory compliance

Description: The value of AI in health care is more evident now than ever. AI/machine learning based-medical software can improve the accuracy in diagnosis and treatment of diseases, enhancing the delivery of health care for each specific individual. In the coming years, more and more in vitro diagnostic medical device/ medical device software based on or including AI/machine learning algorithms will try to enter different markets.
This technology brings some challenges in the design and development of the medical device and in its regulatory compliance.
The process of implementing machine learning (ML) use cases that can scale continues to remain elusive. Analogous to the role of the software-development lifecycle (SDLC), the machine learning model-development lifecycle (MDLC) guides the activities of ML model development from inception through retirement. With the increased scrutiny of AI based medical devices (machine and deep learning algorithms/models) by the regulatory agencies across the world, it has become very challenging for medical device manufacturers to establish and provide adequate evidence and fulfill expectations for their device clearances across regions. So far, different strategies have been adopted by manufacturers of AI/machine learning based-medical software to comply with the various regulatory agencies and specially with the fact that regulations do not allow the entrance in the market of “unfrozen” AI/machine learning algorithms: allow this modality only for research mode; submit several versions of a device for approval; approve a ‘snapshot’ of the medical device software.
Lack of regulations and standards on this topic add to the already existing unclarity and complications, while these approaches are also limiting this technology to allow compliance with existing regulations.
This session will outline the key phases of the MDLC- including case studies and pilot projects which would reflect the current thinking and expectations from various regulatory agencies. The concepts of Federated learning and how it fulfils the clinical validation with real world performance data for the model development lifecycle, would be an added highlight and focus for this session. Continuing the session, we will present the use case of an innovative startup developing AI medical software for the assessment of image quality and breast density in mammography. We will discuss the strategic approaches that we evaluated for their medical software with an “unfrozen” AI/machine learning algorithm to access the European market, focusing on risk management, clinical and post-market activities.