How Can Large Language Models & Generative AI Transform Medical Device Design?

From medical image interpretation to remote monitoring to digital pathology systems, artificial intelligence is already at the forefront of medical device innovation.

While manufacturers continue to develop AI-based devices, they can also use the technology to create these devices more efficiently and effectively. The subsets of AI that are currently applicable to design labs include large language models (LLM), a form of deep learning that consumes and learns from massive datasets to complete language-based tasks; and generative AI, which generates new content based on prompts that guide LLM behavior.

Because LLMs and generative AI process mountains of information in seconds, they serve as powerful engineering assistants and tutors. Generative AI APIs can also be used to de-risk design, allowing engineers to spot and correct flaws sooner. This AI-based oversight both expedites processes and lowers the odds of rejected regulatory submissions and recalls—both expensive problems to fix.

To uncover the potential of LLMs and generative AI for medical device design, we caught up with Xiaofan Mai, founder and CEO of IntelliU, an AI-based design analysis platform.

Medical device developers/engineers must have a broad technical understanding to meet device design requirements. As devices have advanced to become more software-driven, how have engineers had to expand their skill sets?
Xiaofan Mai: Having a broader and deeper understanding of the product is crucial for successful design and development. There are two types of knowledge needed: technical engineering—whether software or hardware—and product domain knowledge.

Related:Could AI Be Your Design & Engineering Partner During Product Development?

For the former, medical devices have more software, and more digital components in the system. There’s software-enabled monitoring functionalities, mobile apps, and cloud communication. They need to learn cybersecurity and HIPAA requirements. They need system interoperability, design, and system integration knowledge and skills to ensure all components work seamlessly and efficiently together. And of course, there's the AI component on the software side. They need to know how to incorporate algorithms into the diagnostic, therapeutic, and monitoring functions of the device.

Product domain knowledge takes years of experience and focused efforts to acquire, and it’s equally essential. It includes understanding the medical application of the devices, the complete logic of product design, and the step-by-step process of the product’s function in the clinical environment.

Related:Promoting AI’s Value for Patients

How do medical device companies find people with the skillsets necessary to produce advanced technology? Are they hiring specialists, outsourcing to independent contractors, and/or providing training?

XM: All the above. The primary goal, for OEMs especially, is to develop, grow, and maintain key employees and maintain that domain knowledge internally to contain IP within the organization. They need HR strategies to attract critical experts and succession planning to replace employees who retire or leave. Doing that efficiently and effectively is key to growing talent internally.

How does turnover, combined with the need for more diverse skills, create knowledge gaps?

XM: If companies rely on one or two experts to pass on product domain knowledge to the next generation, the newer employees may get partial learning or mislearning. So, there's always a gap. How do you fill that gap? If you have great HR strategies in place, then you know how to keep those talents within your organization. But again, those learning employees can easily jump to any other software company.

The device industry has traditionally been more hardware-driven. It’s software engineers need to have passion for medical field, they need to want to learn that knowledge, and then stay at the medical device company for 5 to 10 years to become an expert. Along the way, they need to keep accumulating knowledge to close those gaps. It’s an exciting time when everybody's trying to fill knowledge gaps, and the fight for top talent is becoming more serious.

Related:Embracing AI: How the Healthcare Industry Can Meet the Moment

How do you see large language models like generative AI, stepping in to help fill some of those knowledge gaps?

XM: It can help in knowledge creation, development, and maintenance. It helps people work more efficiently and effectively, because we can use large language models and Gen AI to process tons of product domain knowledge that exists in every part of the organization. Every design team holds volumes of documents—on user needs, market needs, product requirements, risk management, and testing. We can use generative AI APIs to extract that knowledge and reorganize it to more effectively and more intelligently support the whole product development lifecycle. It won't fully replace the nuanced and critical thinking of human engineers. However, it can augment human capabilities.

FDA regulates AI/ML-based medical devices. What, if any, are the regulatory implications of using LLMs as medical device design assistants?

XM: A tool used to assist engineers in design, developing, testing, or even maybe down the road manufacturing, is treated like any other software engineering tool used today. It’s

automating the manual work. It’s providing intelligent and domain-based suggestions to consider in design, and it helps engineers work more efficiently and effectively. In this type of scenario, there’s no FDA oversight.

However, we do make security a top priority because our tools process design data, which is a valuable IP asset.
IntelliU has developed an AI-based design analysis platform. How does it augment human-led device design?

XM: It expedites the design testing cycle and FDA submission, as well as prevents defects. It can cost at least $10,000 to fix one software defect. A typical project could have up to 100 defects, and you might miss a critical design scenario that could trigger a recall. The total cost of a major recall is around $6 million and a 13% stock drop.

Our platform automates manual design analysis tasks and helps derisk the product, which leads to cost reduction and the protection of patient safety. It also helps maintain domain knowledge because experts are freed up from tedious tasks and can spend more time doing what they love.

AI is advancing at a rapid pace. With this platform, it becomes easier to collect knowledge, process it more efficiently, and reorganize it to help engineers learn and execute. So, the knowledge gap gets smaller, and the time needed to learn gets shorter.  

How else do you see LLMs and generative AI transforming medical device design/development?

XM: The future of Gen AI and large language models in medical device design is incredibly promising. As it evolves, it will become easier for domain experts to communicate with these powerful databases to acquire knowledge.

In the context of designing more and more complex medical device systems, it will become an invaluable tool for end-to-end device development. This includes everything from preclinical studies to commercialization to post-market surveillance. I can't imagine a future where AI isn’t there to help every step of the way.

Mai will be presenting on this topic at MD&M Minneapolis on Wednesday, Oct. 16, at 9 a.m.