As rightly said, “Perfection is not attainable, but if we chase perfection we can catch excellence.”
The urge to create a perfect world, which is error free, has given rise to many inventions, one such being Artificial intelligence (AI) and Machine Learning (ML), which is being harnessed, via the software medical devices of today, for their applications in the medical device industry.
What are AI based software?
AI and ML based software is different from other Software as Medical Device (SaMD) since the former is designed to carry out the processes of human thinking in order to facilitate complex tasks, often providing detailed insights and allowing users to focus on other aspects of operations. AI programs learn from and act on data, possess the ability to acquire information, use logic to process data, form reasonable solutions based on known variables, recognize and correct mistakes, making it possible to offer higher quality products and services.
Adaptive vs. Locked
Software developers are using machine learning to create algorithms that are ‘locked’. Their functions do not change constantly or are not ‘adaptive’, however, their behavior can change over time based on new data.
Some AI/ML technologies include
- An imaging system using algorithms to give skin cancer diagnosis information.
- An electrocardiogram (ECG) device to estimate the probability of a heart attack.
Challenges associated with AI/ML-based SaMD
- The traditional standard of medical device regulations were not designed for adaptive AI/ML technologies.
- Adjustments and enhancements due to AI/ML technologies can result in changes as they acquire new data.
- Changes such as an increase in performance, change in the input and/or changes related to intended use would frequently require manufacturers to generate a new premarket submission and re-engage with FDA to evaluate the safety and effectiveness of the revised device.
- AI/ML-based SaMD exists on a spectrum from locked to continuously adaptive algorithms.
- The rigor of performance evaluation for both locked and continuously adaptive algorithms depend on the test methods, quality, and applicability of the dataset used for testing, and the algorithm’s training method.
How to address the challenges
The aforementioned challenges can be addressed by adopting a Total Product Life Cycle (TPLC) approach that facilitates a rapid cycle of product improvement and allows these devices to continuously improve while providing effective safeguards. For the realization of the full power of the AI/ML Algorithm, there are four general principles that maintain a balance between benefits and risks, for safe and effective AI/ML-based SaMD. These are;
- Quality Systems and Good Machine Learning Practices (GMLP)
- Initial Premarket Assurance of Safety and Effectiveness
- Approach for modifications after initial review with an established SaMD pre- specifications (SPS) and Algorithm Change Protocol (ACP)
- Transparency and real-world performance monitoring of AI/ML-based SaMD
Further, Manufacturers can use a “Predetermined change control plan” to refine SaMD Pre-Specifications (SPS) or “Algorithm Change Protocol” (ACP), based on the real-world learning and training on the intended use of AI/ML SaMD model.
This approach would streamline the process required for modifications that may not necessitate another FDA premarket review.
Amidst the race for perfection, AI/ML technologies bring great hope and can empower the Physicians and patients proving the healthcare industry the ability to solve numerous problems that were thought to be a mystery before. It is important to work with strong consultants and service providers like Elexes in order to bridge the gap between the cross functional nature of software firms and regulatory bodies to expedite the process of getting AI-based SaMDs into the market. For questions/comments, please email firstname.lastname@example.org