Radiology group unveils draft of AI guidelines for healthcare

While the use of AI in healthcare has many physicians and healthcare organizations excited about its potential, others have expressed doubt and called for the adoption of ethical standards before the technology is widely implemented.

The Royal Australia and New Zealand College of Radiologists (RANZCR) is answering that call after it recently unveiled a draft of ethical guidelines for the emerging use of machine learning and AI in medicine. In the drafted guidelines, the organization outlined eight ethical principles of safety for the research and deployment of AI tools and machine learning systems in medicine, with an emphasis on clinical radiology and radiation oncology.

The eight ethical principles are:

1. Safety—Patient safety should first be considered in the development, deployment and utilization of machine learning systems or AI tools.
2. Avoidance of bias—To avoid bias, systems should be trained on large amounts and a variety of data that’s representative of target patient populations.
3. Transparency and explainability—Developers of AI systems should consider designs that allow physicians to understand and explain how a decision was made.
4. Privacy and protection of data—Data used in AI research must be de-identified so a patient’s identity can’t be reconstructed.
5. Decision making on diagnosis and treatment—Systems should be used to enhance the decision-making process for diagnosis and treatments, and should still allow the doctor to have the final recommendation.
6. Liability for the decisions made—Healthcare organizations should provide information about the potential for shared liability when researching or implementing AI tools.
7. Application for human values—Physicians should apply humanitarian values to any circumstances AI tools are used.
8. Governance—Hospitals and healthcare organizations using AI technology should have accountable governance committees to oversee its use and compliance with ethical standards.