The emergence of artificial intelligence (AI) in Nuclear Medicine, and the promise of synthetic intelligence, heralds an era of disruptive technology with the potential to re-invigorate the ecosystem of Nuclear Medicine and reengineer the landscape in which Nuclear Medicine is practiced. While AI is not new in Nuclear Medicine, more recent developments and applications of machine learning and deep learning create refreshed interest in the ethical issues associated with AI implementation in Nuclear Medicine. Insight into the architecture, operation and implementation of AI in Nuclear Medicine is beyond the scope of this discussion and has been reported elsewhere (1–3). Nonetheless,it is important to provide key definitions.AI is a general term used to describe algorithms designed for recognition, problem solving and reasoning generally associated with human intelligence, essentially imitating some aspects of intelligent human behaviour (1, 3). An artificial neural network (ANN) is a subset of AI, and in medical imaging, an ANN is an image analysis algorithm composed of layers of connected nodes that simulate the neuronal connections of the human brain (2, 3). ANNs are designed to analyse data and recognise trends or patterns that inform predictions (e.g. classification of disease). A convolutional neural network (CNN) is a type of ANN used for deep learning that employs a convolutional process to extract features from the image itself, while an ANN typically has feature data as the input (1, 2).Machine learning (ML) is a subtype of AI that employs ML algorithms through data analysis without being explicitly programmed. ML tends to be associated with solving problems of logic after learning from human-defined teaching cases typical of an ANN (1, 2). Deep learning (DL) is then a sub-type of ML that adds a number of processing layers (depth) to detect complex features in an image typical of a CNN (1, 2).Synthetic intelligence (SI) provides authentic higher-order reasoning using technology like quantum logic (2), while AI simply imitates human thought. In healthcare delivery in general, AI has two different types of presence in the patient care experience: virtual and physical. Virtual applications are often thought of as software-type algorithms that integrate into the patient care episode often for the purposes of decision-making.Conversely, physical presence is often in the form of a material tangible solution such as a robot or present machine which can interact directly with the patient (4). In Nuclear Medicine, AI solutions are most commonly virtual. Nonetheless, as the field continues to evolve, it will be critical to consider ethical challenges associated with physically present AI solutions.AI in Nuclear Medicine and Molecular Imaging ushers in an exciting era with reengineered and reimagined clinical and research capabilities. AI has the potential to improve workflow and productivity, reduce costs, increase accuracy and empower research and discovery. With this comes a duty of care to patients to ensure AI-augmented diagnosis or treatment provides the best outcomes for patients. Perhaps the most contentious topic to navigate in AI application to Nuclear Medicine is the ethical questions that arise when using human data to develop human-targeted applications.These ethical considerations relate to three distinct areas: the data used, the algorithms applied and the manner in which they are applied to practice. A white paper from the French radiology community (5) and a joint statement of European and North America societies (6) also identify these three areas in the dynamic between ethical and social issues for AI in medical imaging (Fig. 1).
|Number of pages||5|
|Journal||European Journal of Nuclear Medicine and Molecular Imaging|
|Early online date||11 Jan 2020|
|Publication status||Published - Apr 2020|