George D. Magoulas and Nancy Pouloudi
The concern about the ethical implications of the use of artificial intelligence techniques in medicine is ongoing. On the one hand, the use of artificial intelligence increasingly provides opportunities to facilitate and enhance the work of medical experts and ultimately to improve the efficiency and quality of medical care. On the other hand, the debate about the appropriate level and the role of intelligent decision support has become more complex, as technical, organisational and social issues become intertwined. In this paper we use a research project that applies neural network-based methodologies as an opportunity to study the ethical issues that may rise from the application of artificial intelligence techniques in a medical context.
Advances in neurocomputing have opened the way for the establishment of decision support systems which are able to learn complex associations by example. It is acknowledged that the appropriate use of the neural network-based methodologies in medical problem solving could be very effective to improve the efficiency and the quality of medical care. The growing number of projects that employ neural network based methodologies in medical care makes necessary to examine the level and the role that neural networks will play in the development of automatic diagnostic systems.
This paper considers the use of neural networks in image processing and knowledge representation in image interpretation systems from an ethical perspective. The diagnostic attributes of such systems are based on the result of the feature extraction procedure from digital images, in conjunction with previously available medical knowledge, in order to expand this knowledge. A key issue in the use of neural network methods in a medical application of this kind is that it is unclear how decisions are reached. Most neural networks suffer from the opaqueness of their learned associations. In medical applications, this black box nature may make clinicians reluctant to utilise a neural network application, no matter how great the claims made for its performance. Thus, there is need to enhance neural networks with rule extraction capabilities. In addition, it is necessary to examine how productivity can be increased and how quality can be assured. This examination addresses the specification of the problem, the development of appropriate representations for the network input and output information and the preparation of the training, testing and validation data.
Using the acquisition, processing, storage, dissemination and use stages of an information life cycle as a basis for the different stages for the development and use of neural network medical imaging applications, this paper elaborates on the broader ethical implications of the use of neural applications in medicine. These include issues of interpretation, coordination between the technology and the human expert, validation of results and professional responsibilities. Although we concentrate on the use of neural network-based methodologies, the ethical issues discussed in this paper are relevant to a broader spectrum of artificial intelligence and information technology applications in health care. To illustrate this broader set of topics, the paper concludes with suggestions for future research in the area of medical informatics that can support ethical practice.