AI-supported medical device certification: "We want electronic approvals"

The KIMEDS project aims to develop an AI-supported safety solution across the entire lifecycle of medical software to shorten the time from development to certification of innovative medical technology. The goal is to provide best medical care for patients, prompting a close project collaboration between science, industry, and regulatory authorities.

Professor Jochen Hampe is a physician who is joined by computer scientist Dr. Georg Heidenreich to lead the KIMEDS project. MEDICA-tradefair.com asked the two experts why the certification process for medical devices calls for artificial intelligence.

Professor Hampe, Dr. Heidenreich, more efficient certification processes are one of the core challenges of the medical device industry. Why is that?

Dr. Georg Heidenreich: The basic problem is the increasing complexity of medical devices, which simultaneously increases the regulatory requirements of the technical documentation for the approval process. That is why we considered using digitization and the development of an integrated, AI-based product safety monitoring approach that takes us from application to approval.

Prof. Jochen Hampe: If the existing process continues the way it does right now, medical technology will fall short of the current state-of-the-art. It means patients will receive poor care unnecessarily and – I am stating this from a physician’s perspective - medical teams will have to use outdated medical equipment. The current state necessitates artificial intelligence and the digitization of the approval process to where portions of the assessment can be handled by artificial intelligence. Ultimately, this benefits society as a whole.

Since it is evolving so fast, we must first define a dynamic state-of-the-art technology and create methods that reflect how our dynamic development of knowledge and our knowledge of safety properties can be integrated into the regulatory process.

How many researchers are involved in the project and what is the timeline?

Hampe: KIMEDS is a 36-month joint project by the Federal Ministry of Education and Research, abbreviated BMBF, which is funded with 1.46 million euros. The consortium consists of the Federal Institute for Drugs and Medical Devices, iSAX GmbH & Co. KG, the International Center for Computational Logic at the Technical University of Dresden, Siemens Healthineers, TÜV Süd and the Else Kröner Fresenius Center for Digital Health. The project collaboration will always include five or six scientists from the various partner institutions.

What is the purpose of the project?

Hampe: It is primarily a pilot project that assesses whether our concept is ready for a full-scale implementation. Our interdisciplinary team is a major strength in this setting. We have people stemming from the industry sector like Georg Heidenreich who are really knowledgeable about this process. We also have the relevant public stakeholders on board. Our key partners are the colleagues from the International Center for Computational Logic at the TU Dresden, who are experts in AI, knowledge technologies, and logic in computer science. Everything we do that involves AI in this setting must subsequently be comprehensible and explainable. After all, Georg, what happens if something goes wrong with a medical device?

Heidenreich: You have a risk management file to recreate an ex-post scenario via the documentation and the device. We want to digitize this file, which to date has been a simple hard copy that the authorities receive. This would also enable the authorities to reenact the scenario. Conversely, when manufacturers design devices, they can use libraries and past experiences to compile their risk management file. Artificial technology plays a role because it can run certain tests. We don’t assume that AI will be able to provide a comprehensive assessment of the safety risk in three years. However, by then we will have developed the method that makes this possible and a network-based platform that will remain active beyond the end of the project, collects knowledge, and continues to write the AI ​​methods. The project will definitely deliver initial AI-based plausibility checks. Within the emerging network, we will further delve into the content, enabling a check for logical consequences to represent certain device types.

How should we envision the system you are developing?

Hampe: We initially learn from a pilot case, an X-ray machine where we benefit from the expertise of our partner Siemens Healthineers. We plan to design the system to essentially manage any risk. It is meant to allow modular safety and risk documentation. Many modern medical devices are developed as a monolithic architecture, which is particularly the case for risk documentation.

Heidenreich: The approval process is also monolithic, which is something that is impossible to manage for auditors. The Medical Device Regulation (MDR), the new regulation from Brussels, is far more complex as it pertains to the approval documents. Now everything must be re-approved.

Hampe: It’s something we simply must improve because the process no longer works for any of the parties involved - the regulatory agencies, the manufacturers, the medical device users. Everyone has reached their breaking point and is looking for powerful alternatives to the current processes.

Heidenreich: We want an electronic approval process. Thanks to an AI-based solution, the future scenario should look something like "enter the medical device documentation and out comes the CE mark". That is our vision for the next 20 years.