Radiologists point to need for ethical guidelines on AI

According to stakeholders, radiologists remain ultimately responsible for patient care and will need to acquire new skills to do their best for patients in the new AI ecosystem.

AI is complex and carries potential pitfalls and inherent biases, so the ethical application of AI in radiology, among other specialties in healthcare, will require stakeholders to consider how they are using data, as well as how they develop and operate decision-making tools.

That’s according to a statement from several of the world’s leading radiology and imaging informatics groups, including the American College of Radiology (ACR), the European Society of Radiology (ESR) and the Society for Imaging Informatics in Medicine (SIIM).

The problem, the statement summarizes, is that “widespread use of AI-based intelligent and autonomous systems in radiology can increase the risk of systemic errors with high consequence and highlights complex ethical and societal issues. Currently, there is little experience using AI for patient care in diverse clinical settings. Extensive research is needed to understand how to best deploy AI in clinical practice.”

The authors noted that AI has the potential to radically improve care delivery, among other things, but the technology also comes with a number of potential risks, such as the possibility of grave medical errors and the amplification of inherent biases.

“Because of the international nature of AI research, rapid pace of technology development and cross-border deployment of AI software, an ethical framework for AI in radiology was much needed,” said An Tang, MD, MSc, FRCPC, chair of CAR's AI Working Group and co-author. 

“This multi-society statement highlights ethical issues and discusses how to detect and manage them in a manner that is beneficial to patients.”

According to the statement, the ethics of data collection, management, and analytics are essential to the use of AI in radiology, but when an AI model is implemented, radiologists should know how they will document and notify patients about how they will use patient data and recognize what biases may exist in the data used to train algorithms. They should also reflect on the steps they have taken to mitigate these biases, and how users should consider any remaining biases.

“Developments in artificial intelligence represent one of the most exciting, and most challenging, changes in how radiology services will be delivered to patients in the near future,” said Dr. Adrian Brady, Chairperson of the ESR Quality, Safety and Standards Committee and co-author. 

“The potential for patient benefit from AI implementation is great, but there are also significant risks of unexpected or unplanned harmful effects of these changes. It's incumbent on professionals working in this area to ensure that patient and public benefit and safety are paramount.”