Artificial Intelligence in medical devices: challenges and opportunities

The sector of Artificial Intelligence (AI) is continuously evolving, and in the next few years the impact of this new technology will be more and more significant in several spheres.

The use of software based on Artificial Intelligence in the health setting is spreading widely also in the field of medical devices (MDs).

The use of software in this sector is not new and Regulation (EU) 2017/745 (MDR) deals with this in several parts. In particular, the MDS defines the software as active devices and dedicates rule 11 of Annex VIII to their classification.

Software is also widely treated in the General Requirements on Safety and Performance (GSPR); in addition, some MDCGs were published on this regard (for instance MDCG 2019-11 regarding the classification of this specific type of MD).

Types of Software as Medical Devices

Three types of software can be defined as medical devices: SiMD (Software in a Medical Device), SaMD (Software as a Medical Device) and software as accessories in a medical device.

In the first case, the software is part of the medical device, cooperating with it in performing its functions (e.g. diagnosis, cure or treatment).

On the other hand, in the SaMDs, the software itself performs the function (“standalone software”). Finally, in the last case, the software is not a part of the medical device, but it works as an accessory.

Differently from classic algorithms (the so called “locked”) that operate with prefixed instructions on a single set of data as input to produce given specific output, some of this software can store data to adapt itself dynamically and change its own algorithm so that future calculations and generated output are more accurate, thus resulting more flexible and adaptable.

For these reasons, the AI potentials in this sector are huge (from the use of Artificial Intelligence for medical diagnosis to a better prevention of diseases), but at the same time it can pose many risks (privacy violation, performance accuracy, ethical issues regarding the processing of sensitive data).

The legal framework of Artificial Intelligence

For the regulating point of view, the fact that the algorithm “learns” and evolves poses a relevant challenge for the regulation of these products.

It must be established, for instance, if the regulatory authorisation of medical devices equipped with Artificial Intelligence should only consider the “basic” version of the algorithm or if its changes and adaptations also need assessment (an issue known as the “update problem”).

Having such ability to evolve, in fact, it needs to be demonstrated how the software maintains unchanged the intended use established by the manufacturers.

To face AI opportunities and challenges, the European Union needs to develop a full and clear legal framework enabling manufacturers to manage their devices appropriately.

In this perspective, in 2021 the European Commission published a proposal of regulation on artificial intelligence with the aim of defining a horizontal regulating approach capable of ensuring that the AI system placed on the EU market are safe and reliable.

The regulatory future of Artificial Intelligence shall be set avoiding the creation of ambiguous scenarios regarding aspects such as definitions and obligations of the concerned parties, the classification of the devices and the type of risk strategy to be applied.

To reach this goal in the framework of the MDs, a careful integration should be carried out between the horizontal regulations (i.e. the future AI regulation and the GDPR) and the MDR.

Moreover, new standards and common specifications shall be developed, providing AI managing tools for the different industrial sectors. In this sense, an ISO/IEC JTC 1/SC 42 joint technical committee has already been set up at international level.

The development of such technical and legal framework will have a strong impact on the manufacturers of MDs equipped with artificial intelligence but it will also be an opportunity to exploit all potentials of this new technology.

Written by Simone Pippa on 27/03/2023

Foto di Colin Behrens da Pixabay