Implications of MDR for Medical Devices Incorporating Artificial Intelligence and Machine Learning

Artificial intelligence (AI) and machine learning (ML) in medical devices is an important part of the healthcare industry with the potential to improve patient care, as well as administrative processes by automating tasks and achieving faster results. According to the latest report by PwC, AI will contribute an additional $15.7 trillion to the world economy by 2030, with the greatest impact being in the field of healthcare. Real-world applications of AI and ML in medical devices include imaging systems used for diagnostic information, smart electrocardiograms estimating the probability of a heart attack, and AI-assisted stethoscopes that patients can use at home.

Artificial Intelligence can be broadly defined as the:

“science and engineering of making intelligent machines, especially intelligent computer programs using models based on statistical analysis of data, expert systems that primarily rely on if-then statements, and machine learning.”

Machine Learning is an:

“artificial intelligence technique that can be used to design and train software algorithms to learn from and act on data, essentially creating adaptive algorithms that have the potential to continuously optimize device performance in real-time to improve patient outcomes.”

AI in medical devices is a relatively new technology that is constantly evolving, and therefore, is an uncharted territory that presents challenges for regulatory bodies. The EU directives and regulations (e.g. MDD, MDR), and the harmonised standards (e.g. EN IEC 62304) do not have concrete guidelines on medical devices which incorporate AI and ML. The existing general regulatory requirements for such devices include:

• Demonstration of safety and performance of the medical device
• The devices must be validated against the intended purpose and stakeholder requirements and verified against the specifications (including MDR Annex I 17.2)
• They must ensure that the software has been developed in a way that ensures repeatability, reliability, and performance (including MDR Annex I 17.1), including describing the methods used for verification
• If the clinical evaluation is based on a comparator device, this device must be sufficiently technical equivalent, which explicitly includes the evaluation of the software algorithms (MDR Annex XIV, Part A, paragraph 3)
• Before development, manufacturers must determine and ensure the competence of the people involved (ISO 13485:2016 7.3.2 f)

The greatest benefits of AI/ML in software resides in its ability to learn from real-world use and experience, and its capability to improve its performance. The ability for AI/ML software to learn from real-world feedback (training) and improve its performance (adaptation) makes these technologies uniquely situated among software as a medical device (SaMD).

In the spring of 2019, the FDA released a proposed regulatory framework for AI/ML-based Software as Medical Device (SaMD) as a first step to regulate such devices. The aim being a potential approach to mandate premarket review for AI/ML-driven software modifications. The proposal discusses its experience in the premarket approval of “locked” algorithms. However, “adaptive” AI/ML-based SaMD with the ability to continuously learn and adapt to real-world data have not been able to be regulated using the traditional pathways. Suggestions were made that a new total product lifecycle (TPLC) regulatory approach is required to keep up with the pace of these highly iterative, autonomous devices while continually providing an effective and safe regulatory framework.

The proposal also translates risk categories by using the International Medical Device Regulators Forum (IMDRF) SaMD risk categorization framework, which combines the seriousness of the medical condition, and the significance of the information provided by the AI/ML-based SaMD to the healthcare decision. The risk-based approach to categorize SaMD is based on intended use. The four categories, ranging from lowest (I) to highest risk (IV), reflect the risk associated with the clinical situation and device use.

The proposal anticipates that modifications to devices incorporating AI/ML may involve algorithm architecture modifications and re-training with new data sets. The types of modifications fall into three broad categories:

• Performance – clinical and analytical performance
• Inputs used by the algorithm and their clinical association to the SaMD output
• Intended use – The intended use of the device for the state of the healthcare condition

The proposal addresses the necessity of a TPLC approach by assessing the quality of a manufacturer’s software development, testing, and performance monitoring of the device. This approach enables the continual evaluation and monitoring of a software product from its premarket development to post-market performance.

The FDA’s proposed TPLC approach is based on the following general principles that balance the benefits and risks, and provide access to safe and effective AI/ML based SaMD:

• Establish clear expectations on quality systems and good ML practices (GMLP)
• Conduct premarket review for those SaMD that require a premarket submission to demonstrate reasonable assurance of safety and effectiveness, and establish clear expectations for manufacturers of AI/ML-based SaMD to continually manage patient risks throughout the lifecycle
• Expect manufacturers to monitor the AI/ML device and incorporate a risk management approach, and other approaches outlined in “Deciding When to Submit a 510(k) for a Software Change to an Existing Device.” This will entail guidance in development, validation, and execution of the algorithm changes (SaMD Pre-Specifications and Algorithm Change Protocol)
• Enable increased transparency to users and the FDA using post-market real-world performance reporting for maintaining continued assurance of safety and effectiveness

The future for AI/ML based medical devices is exciting and requires a shift in perspective to maximize the safety and efficacy of AI/ML in health care. It currently poses significant challenges for manufacturers and regulatory agencies, therefore requiring collaboration between them to develop integrated systems and a novel streamlined approach in regularizing such advanced tools.