Suppose you want to implement an imaginary artificial intelligence (AI) system in the healthcare sector. This paper examines the regulatory framework required for this in terms of development, clinical research and authorisation, commercialisation, and compliance with the healthcare system. The subject of this investigation is the so-called artificial narrow intelligence ("ANI") systems used in the healthcare sector. These systems use machine learning ("ML") or deep learning methods to process real databases in healthcare. Examples of such systems that are detective, i.e., based on historical and Computational Measurable Dynamics or learning along predefined parameters and protocols, include imaging diagnostics, personalised precision medicine, telemedicine for measurement analysis and communication, decision support systems (DSS) for operational analysis and healthcare management, and solutions to accelerate virtual clinical research. For reasons of scope, this paper focuses on the healthcare sector in particular, and general, issues related to AI are addressed only to the extent necessary to shed light on the specificities of healthcare.
Software and program code: Logically, AI is a software that can mimic human capabilities. One can differentiate between the types of AI, as the case may be, the technology on which the application is based, the industry and sector in which it is used, or what functions it has. In the context of healthcare, let us now narrow down the scope to machine learning methods that support the healthcare system through data analysis, image recognition and language technology solutions.
Medical device: The MDR (Medical Devices Regulation)[2] defines the term as follows: "a medical device is any instrument, apparatus, appliance, software, implant, reagent, material or other article intended by the manufacturer to be used, alone or in combination, for human beings for one or more of the following specific medical purposes, such as the diagnosis, prevention, monitoring, prognosis, treatment, or alleviation of disease, or the diagnosis of injury or disability". Additional medical purposes include the testing, replacement, or modification of an anatomical, physiological and/or pathological process or condition. The processing of information by testing samples (organs, tissues, and blood) from the human body - in vitro (i.e., outside the body, e.g., in a test tube or a cup) is also included. The definition of a medical device accessory also complements the concept, although not serving a medical purpose in itself, when used in conjunction with a medical device, the accessory enables or facilitates the use of the medical device. An active device is a device that works outside the human body, such as software.
Within the medical device category, a distinction should be made between medical devices and medical assistive devices[3], which are medical devices or technical nursing equipment for the personal use of the patient, but do not require the constant presence of a qualified medical professional and are typically used for diagnostic, therapeutic, rehabilitation or nursing purposes.
While not yet in force, the draft Regulation of the European Parliament and the Council is laying down harmonised rules for Artificial Intelligence (hereinafter "AI Code")[4] which defines AI system as a software developed using specific techniques that can generate outputs, such as content, predictions, recommendations, or decisions, with respect to human-defined objectives, thereby influencing the environment in relation to which they interact with. The itemised list that may be extended[5] includes supervised or unsupervised machine learning and deep learning methods, logic programming, the knowledge base, an inference engine, a system of experts, statistical approaches, and estimation methods. The definition has already been given several criticisms, mainly from big tech companies that want a more precise definition or narrowing of the broad definition, such as Huawei and IBM.[6]
On an international scale, the WHO has issued a recommendation[7] specifically addressing the control and ethical considerations of using AI for health purposes. In its definition, it refers to the OECD Council of Artificial Intelligence Recommendation,[8] which defines AI as a machine system capable of making predictions, recommendations and decisions for purposes set by humans and designed with varying degrees of autonomy. The WHO Recommendation lists technologies which are machine learning applications that relate to similarity (pattern) recognition, language technology processes, sound and signal recognition and professional systems.[9]
On a national level, in Hungary, previously, the Act on Healthcare essentially provided the same, although slightly narrower definition of medical devices, which was subsequently, in May 2021, incorporated into the text of the legislation with identical content.[10],[11] On the regulatory level, additional definitions can also be found depending on what the device is made of, how or what purpose it is used for, what purpose it serves, but from the point of view of artificial intelligence, these detailed rules in this article are not evaluated in terms of definition, because it is determined by the purpose of use, which in some cases not even known at the moment.
Accordingly to the aforementioned it is easy to see that there is no superintelligence, self-aware robots, or systems that can replace or take human decisions. In healthcare, AI systems need more time to become autonomous, they are designed to prepare and assist human decisions, but still operating under human control.
How does an idea or development become a usable AI system in healthcare? In the case of devices and products used for healthcare purposes, there is a huge risk that people's health might be adversely affected by an inadequately tested and controlled product. The legislator has therefore developed a detailed set of procedures that
- 30/31 -
guide applicant manufacturers through a series of stages. Under the MDR,[12] for example, a clinical trial is a documented series of events that produces a meaningful and measurable clinical outcome through the evaluation of evidence, under the personal responsibility of the investigator, which is otherwise the responsibility of the sponsoring company as bankroller and manager. The trial subject (i.e., participating patient) gives his/her prior informed consent in writing to any information that is relevant to his/her decision and that confirms that his/her consent is voluntary and freely expressed to participate in the trial. The AI Code, which is not yet in force, should include instructions for the use of AI for certain high-risk AI systems, which should also mean guidance, lessons, and training.[13] The point is providing information: transparent use for the person for whom the system is performing an operation. In other words, the information should be given to the person concerned (data subject).
A Jogkódex-előfizetéséhez tartozó felhasználónévvel és jelszóval is be tud jelentkezni.
Az ORAC Kiadó előfizetéses folyóiratainak „valós idejű” (a nyomtatott lapszámok megjelenésével egyidejű) eléréséhez kérjen ajánlatot a Szakcikk Adatbázis Plusz-ra!
Visszaugrás
