LAUSR

For the radiologist facing the start of a new era in artificial intelligence (AI), it is reasonable to ask where we fit. The role of the radiologist as a medical imaging specialist should remain unchanged; however, new knowledge and training will be required for radiologists to successfully integrate AI into their practice. As the inaugural AI Fellow at the Royal Adelaide Hospital (RAH), my aim with this short commentary is to share my experiences and thoughts about acquiring and developing this knowledge, as well as drawing attention to some important considerations for radiologists working in the AI space. I completed my radiology training via the Western Australian Radiology Training Program and have always had a latent interest in Artificial Intelligence. My initial searches for information on pursuing advanced training in AI found very few formal informatics fellowships, essentially available only in a handful of sites in North America. However, the discovery of the possibility of pursuing an AI fellowship in Australia was a very attractive opportunity and I began my year at the RAH in June 2020. The structure of the fellowship is a two-day research and three-day clinical split, covering the full gamut of knowledge required to evaluate clinical AI systems from the perspective of a practicing radiologist, and has included concepts such as machine learning, Python programming, statistics, AI safety, and commercial AI integration. Some of these areas are highlighted below in relation to broader considerations for the clinical radiologist in AI. Radiology sits at an advantageous interface between clinical medicine and technology, making it an attractive target for AI development. The application of modern AI to medical imaging has produced a wide variety of clinical AI tools, performing tasks such as alerting clinicians of important clinical findings, imaging analysis and interpretation, improvement of imaging algorithms and image acquisition/protocolling. A key issue for radiologists remains that whilst the majority of these applications directly impact upon the clinical radiology workflow, they remain tools that have been developed largely by non-radiologists. This combined with the rapidly expanding marketplace for these applications and the novel nature of these products creates important challenges for radiologists and their interaction with AI in the clinical setting, not the least of which are the ethical concerns, medico-legal liability, and the workforce implications of working alongside AI. As experts in the clinical effects of medical imaging technology, radiologists should ultimately remain a gatekeeper and key stakeholder in the decision to integrate AI into clinical practice, and this concept has been formalised at a policy level in the ‘Standards of Practice for Artificial Intelligence’ released by the Royal Australian and New Zealand College of Radiologists. A key concept of this document is the proposed role of a Chief Radiologist Information Officer (CRIO), which is described as a radiologist who is responsible for AI integration at a practice level with key responsibilities