Learn More About AI in Radiology — Why Now Is the Time
It’s hard to go a day without hearing something about AI in radiology. The news is mixed — and confusing. Some stories are about how AI outperformed radiologists in a particular task. Others are about how a particular AI solution worked well at one practice but not at another. Still others call attention to the fact that some “AI” solutions aren’t even using AI. So what are you, a young and early career radiologist, supposed to do?
My goal is to convince you that if you haven’t taken the time to learn much about AI or get involved in the AI efforts within your practice, now is the time to start. There are three important roles we YPS radiologists — and all radiologists — can (and should) play in the process.
1. Help AI developers and entrepreneurs understand the clinical workflow and the problems that need solving.
Have you ever seen an article about AI performing some medical imaging-related task and thought to yourself, “Why would a radiologist need to use that?” As with any new technology, it is important to develop solutions that have clinical relevance and the potential to improve patient care. As the domain experts in radiology and its subspecialties, we have the knowledge to guide AI development and help our industry colleagues avoid the problem of creating solutions for problems that do not exist. The ACR’s Data Science Institute™ (DS) has introduced the AI-LAB™ (ailab.acr.org) — home to a number of AI-related resources but also to the ongoing effort to develop use cases to guide vendors in choosing which problems to solve. The use cases are detailed, templated descriptions of radiology tasks that could benefit from AI, and include a description of the relevant data elements needed to tackle and solve each problem. Interested in developing a use case? Contact the ACR DSI and share your expertise.
2. Evaluate AI solutions for accuracy and workflow integration.
When new technology works well in a test environment, there is a natural push to integrate it into a product. But the test and production environments are often very different. The real world is complex and sometimes unpredictable, whereas parameters in the test lab are often tightly controlled. It is important for AI solutions to be vetted in the clinical environment where they will be used, and to seamlessly integrate into the radiology workflow. Consider how often you would use an AI tool that required you to log into a separate system to review its output, as opposed to one whose results appear within your EMR, radiology information system, or hanging protocol with the rest of the case you’re interpreting?
3. Determine where to invest in AI for a radiology practice.
AI has the potential to play a role at all steps of the care chain as described by Imaging 3.0®: before, during, and after the patient undergoes imaging. Although most of the AI development to date has been focused on detecting findings in images, there are applications beyond image interpretation discussed by Paras C. Lakhani, MD, and colleagues (www.jacr.org/article/S1546-1440(17)31287-5/abstract) that will likely penetrate the radiology workflow more successfully and sooner than the image-based AI solutions.
Nevertheless, as we consider the numerous image-based AI solutions currently available on the market, it is difficult to know whether they work, and how well they might work on our real-world data compared to the curated test data. To this end, AI-LAB™ also allows you to evaluate different AI tools without exposing your data outside your practice’s firewall. As you consider how to invest in solutions, this is another way to trial and compare different approaches side-by-side, and choose the one that works best for your patients.
While many of us have theories about how the AI-enabled future of radiology will look, the truth is that none of us really know. As Maciej A. Mazurowski, PhD, wrote, all we do know is that AI will disrupt radiology (www.jacr.org/article/S1546-1440(19)30064-X/abstract). It is up to us to ensure that it is an innovative disruption — one that helps us take better care of our patients, and one that promotes the wellness of our specialty as a whole and ourselves as individuals.
By Tessa Cook, MD, PhD, assistant professor of radiology at the Perelman School of Medicine at the University of Pennsylvania in Philadelphia