Prior to the AIMed Radiology virtual event held in association with the American College of Radiology (ACR) on November 5, AIMed is excited to speak with Dr. Geraldine McGinty, ACR President on the challenges of deploying artificial intelligence (AI) in radiology; who should pay and be liable for the technology, and her thoughts for the next decade.
AIMed: When did ACR begin to realize that AI is probably going to do something different to medicine, particularly Radiology?
Dr. McGinty: ACR did a strategic plan in 2014; at a time when we were still feeling the aftermath of the 2008 financial crash. The plan involved a lot of rigorous work but when we refreshed it three years later, we realized we did not include anything about AI. All of a sudden, we felt like we were behind the curve.
We noticed due to the expanded computer processing power, more patient data from sources such as the electronic health records (EHRs) and accelerated demands for imaging, after a slowdown in the immediate post-crash period that AI was now an obvious factor in our profession’s future and that we needed to influence that future. In order to ensure that influence is in the service of our patients, we created our Data Science Institute in 2017.
AIMed: Radiologists are probably the techiest of all clinicians and they are proud of it. However, not all medical subspecialties are equally enthusiastic towards AI. In your opinion, do you think radiologists are the techiest of all? If so, how do you think radiologists can influence their colleagues in other medical subspecialties and make them equally excited about AI?
Dr. McGinty: I think you are right; most radiologists are excited about the possibilities associated with AI probably because Radiology is a specialty at the interface of technology and humanity. Rather than resisting change, we tend to be excited by innovation.
We have been disappointed in the past when certain tools and technologies have not lived up to their promises and I think we will be more demanding in future. We won’t just be looking to new technology to improve diagnostic accuracy and outcomes although that’s a given. We’ll also be looking for tools that promote health equity, cost effectiveness, and sustainability.
Radiologists are not scared of AI but we are appropriately cautious about adopting tools without evidence that they actually make a difference. Our ability to rigorously evaluate new tools and ensure that we are thoughtful about their implementation can definitely help other specialties.
AIMed: As a healthcare leader as well as a radiologist, what are some of the challenges you often come across in the use of AI in Radiology?
Dr. McGinty: Most of us now function in large health systems where demands on IT, governance, and infrastructure are significant. It becomes extremely difficult to introduce new tools because there’s limited bandwidth or appetite for risk. We also lack clarity on the way many of these algorithms actually work which can make us cautious.
We’ll need to know that the tool was trained on a diverse enough dataset, that it will function in our organization in the way it was designed and that we can rely on it to augment our practice with confidence. And then, the uncertainty about how we will get reimbursed is a definitely also a barrier even though I won’t say it is number one.
AIMed: As an expert on payment models for imaging, how do you think the payment models for AI should be?
Dr. McGinty: I have thought that AI would most likely be incorporated into what’s called the “practice expense” or part of a reimbursement that reflects the cost of equipment and supplies. A colleague of mine said in a Tweet recently that AI needs to be paid by the person who benefit most but the problem is developers, physicians and patients are all likely to benefit from AI.
At the beginning of the month, CMS (the Centers for Medicare & Medicaid Services) approved the first AI augmented medical care reimbursement on a deep learning model which identifies signs of a stroke in brain CT scans. There isn’t a lot of transparency on how the pricing decision was made. It appears to be a short-term decision and compared to the professional reimbursement for the radiologist seems lopsided. While it’s an important milestone, I’m not sure it has really taught us a lot about what the future looks like.
AIMed: Do you think it is right to assume those who get paid for creating or using AI should also be the ones who held responsible when AI is not working as intended?
Dr. McGinty: I think that’s an excellent question. Who should be held responsible. In any value chain, that includes a physician and a patient. I would argue that the physician has to assume the ultimate responsibility for the care that the patient needs and the ultimate ethical responsibility for the safety of the same patient. We can extend the argument to product liability for sure but I see physicians as having the control to ensure the tools are chosen and used in a thoughtful way.
AIMed: Have you ever imagine in an ideal world; how will AI be helping radiologists?
Dr. McGinty: Of course. I would love to think that in the future I am walking into my office in the morning and looking at cases with the highest potential for abnormalities when I am fresh. I would love to see risk models for disease built from an integration of imaging, pathology and genomics data and offer my patients personalized screening. I would love to create reports that are not just word documents but a product which captures the incredible power of the images and are intelligible not just to the referring physician but also to the patient.
I see many opportunities to improve care particularly on the administrative side where in the US, those costs are inordinately high. From automated protocoling for imaging to converting physician targeted reports and patient centered summaries; there are many low hanging fruits. Finding and capitalizing on those opportunities will require relationship building and ensuring a diversity of perspectives, especially if we want to improve outcomes in specific communities where we have existing disparities.
As much as I want to convey a sense of real excitement about the possibilities that AI can give us, I want to stress the importance of conversations between all the stakeholders, especially if we want to really make a change. If one group of people tries to advance AI alone, we are unlikely to realize its full potential.
AIMed: How do you think radiology and AI will turn out to be in the next decade?
Dr. McGinty: I am very confident that 10 years from now, I will be using AI whenever I’m interpreting a mammogram. I hope that I’ll be able to deliver more personalized care especially to populations who are not optimally served by the current screening guidelines such as Black women under 40.
*This article was originally published on AIMed Blog on 15 September 2020.