The ethical and legal issues that will arise from the implementation of AIS in the radiological practice can be classified into two categories:
- Those related to the development of AI algorithms.
- Those related to the use of AIS in the radiological practice.
1. Issues related to the development of AI algorithms
The development of AI algorithms requires a significant quantity of medical data, especially those based on Deep Learning (Fig. 1). This kind of data usually contains sensitive information belonging to the patient’s private life; hence the use and dissemination of this kind of data bring new ethical and legal problems.
1.1 Ethical issues
Currently, all the evidence indicates that the development of AIS will improve the medical practice. However, it entails the use and dissemination of a large amount of medical data. Is it ethical to use patients data to develop AIS?, is an informed consent the best legal basis to do it?
There are two more aspects that makes these distinctions even harder. Firstly, the beneficiaries of these systems will probably be patients different from those whose data will be used. Secondly, informed consent delays the research, and it is a big barrier, so other legal basis to use the data are being considered. General Data Protection Regulation (GPDR) allows the use of data for biomedical purposes without informed consent as long as the purpose is of public interest and an Ethics Committee has certified that. So the question is, what can ethically be used for the purpose of public interest?
To solve that issue, we propose to make an analysis based on the four basic principles of medical ethics (Fig. 2): beneficence, non-maleficence, autonomy, and justice.
The beneficence principle compels the development of AIS and thus to use the data because it is supposed that AIS will improve the medical practice, and biomedical community have do whatever they are able to improve the healthcare, as long as the action respects non-maleficence – with a responsible use of data patients should not suffer any damage – and justice – according to equality and human rights. But, what happens with autonomy?
The autonomy principle defends the patients's right to deny the use of their data. This principle requires to one to obtain an informed consent before using the data. However, the beneficence principle defends the opposite. It means there is a conflict between these two principles. Non-maleficence and justice are respected in both cases.
For these situations of conflict, the principle of double effect can be useful. It says that when an action will have two consequences (one good and one bad), the action will be ethical if the benefits are bigger than damages, and the purpose of the agent is to get a benefit.
According to that, the use of medical data for a research purpose would be ethical, since the beneficence of society outweights the autonomy of a single patient.
1.2 Legal issues
The use of personal data is regulated by GPDR-2016/679, whose aim is to give European citizens control to over their data. Although it was firstly designed for business, it is currently the legal framework for biomedical research too.
According to article 6, personal data may not be processed unless there is at least one legal basis to do it, and there are two possible basis for biomedical research: to obtain an informed consent (regulate in articles 4, 7, and 8) or to perform a task considered as public interest, usually by an Ethics Committee.
In our opinion, the second scenario is ill-defined, and biobanks of medical data could be a great opportunity to create a more solid ethical and legal framework (Fig. 3). In the last years, several international associations have called for the creation of biobanks, such as the European Union (EU), European Society of Radiology (ESR), and Radiological Society of North America (RSNA).
However, biobanks also bring legal challenges. The first challenge is data ownership. It is important to clarify who is the owner of the medical data since the owner is who can give data to the biobank, and to clarify who will be the owner when data are in the biobank.
Some possible solutions have already been proposed. The most popular is a system similar than the one used with patents, a kind of embargo period after which the images will be open access.
Secondly, the other big challenge is the capitalisation of data, or the use of medical data for private and economic purposes. The arrival of biobanks could bring a proper environment to convert clinical data into a business source. “Data is the oil of the digital era” (The Economist, 2017).
2. Issues related to the use of AIS in the radiological practice
The other important challenge concerns the use of AIS in the radiological practice. Artificial Intelligence software is designed to help radiologists in making decisions, so AI programs will have an influence on the patient’s outcome. What kind of ethical issues will arise from using a non-moral system in radiology? And, who is legally responsible for these decisions?
2.1 Ethical issues
The medical profession continuously deals with ethical conflicts because there are no universal rules for each situation, and the medical actions are performed on people. Doctors must proceed as moral agents when they are practising medicine. In ethics, a moral agent is someone who is entrusted to make decisions based on what is right or wrong. If AIS are making medical decisions, they have to behave as moral agents following an ethics code.
Therefore, the biomedical community has the challenge of creating a universal ethical code for AIS. Once again, this hypothetical code should be based on the four principles of medical ethics since it is the only way to achieve a global agreement of the entire biomedical community.
2.2 Legal issues
Several legal conflicts have not been solved yet, and we would like to introduce two of them.
Firstly, to determine who is responsible for an action performed by AIS. Civil liability is the legal requirement to compensate another party for causing them damage, and since a machine is not able to satisfy this premise, it cannot be responsible for the action performed. In addition, AIS are getting more and more autonomous, and the more autonomous systems are, the more diluted the responsibility becomes. Thus clearly this question is increasingly getting more complicated.
To this extent, some solutions have been proposed and discussed in the EU, such as to create a European Agency for Robotics and AI to join experts from the EU Member States, and to create what is called “electronic personality” (Fig. 4).
Finally, we would like to introduce another big concern that we have. It is about if the training sets used to develop the AIS must be open-access by law.
A training set is a group of examples used to fit the parameters of an AI-model. Usually, it consists in a group of patients with one pathology. AIS learn how to perform some task by fitting their parameters based on the characteristics of one population (training set). When they are used to perform the same task in a different population, the results can be completely different. Hence, training sets could be very important and useful information for radiologists and clinicians, and maybe they should be open-access by law, as the medical drugs information is.