AI Governance: Plans in Place?

AI is here to stay in society. In medicine, new use cases for AI are developing at a rapid pace. Whether improving efficiency, image interpretation or report generation, AI has demonstrated early signs of playing an important role in radiology.

Successful integration of AI into radiology practice can and will affect patient care. As physicians, we should lead from the front and ensure the safe deployment of AI into our practices for our patients. As each practice enters agreements with AI vendors, many questions still need answers:

• How will your practice monitor AI effectiveness in clinical care, specific to your demographics?
• How does using AI impact quality care at your practice? How do you measure it?
• Are you using AI equitably and justly for your patients?
• How are you monitoring AI degradation over time?
• Who is liable for bad outcomes?
• If images are being manipulated by AI, how are the images affected? Is it based on physics? Is the raw data still available?
• Who owns the data?
• Does your hospital control medical imaging, or does your practice?
• Are other non-radiology specialties using medical imaging AI at your institution without your practice’s knowledge? Will they?

The use of AI is currently governed differently in each practice and hospital, if at all. It is important to know how each institution views AI, the stakeholders involved and their roles regarding patient care. Every practice should aim for the ability to distinguish AI applications that can be implemented earlier (e.g., some non-interpretative AI), and those that need time for rigorous evaluation. Depending on the political climate, relying on FDA regulation may not be enough.

Implementation of AI governance is imperative for excellent patient care. While a national solution has yet to be created, each institution should develop a whitepaper outlining the proper evaluation, maintenance and effectiveness of AI use. Depending on practice size and dynamics, consider creating committees to discuss regularly how AI is being used, if it is working and whether it still brings value. Monitoring for AI slippage and degradation overtime should be reevaluated regularly. As more and more AI solutions are created, ensure the solution truly fits each practice’s problem. Remember, everything looks like a nail to a hammer.

AI is exciting, innovative and could be the solution to many problems in medicine and particularly in radiology. We should embrace AI integration into radiology if it adds value. Be open to new ideas for practice transformation through AI but analyze carefully to provide excellent patient care.

Radiology AI governance and guidelines at the national and practice levels should be developed to maintain patient safety. To help address this, the ACR® launched the Recognized Center for Healthcare-AI (ARCH-AI), a set of criteria to guide radiology facilities in using AI safely. While some large institutions have joined, further adoption is needed.

The key to successful AI integration is physician leadership and thoughtful oversight to best take care of our patients who rely on us. Before taking your next AI vendor meeting, consider this question: Are AI governance plans in place?