Over the last few years, physicians, entrepreneurs, and engineers have heard constant chatter about AI replacing radiologists in the foreseeable future. The medical community surmised that the advent of AI would deter the best and the brightest from applying to radiology. However, attending the RFS program at ACR 2022 in Washington, D.C., helped solidify an inclination I’ve had for quite some time: Radiology residents want to work with AI. In fact, 2022 was perhaps the most competitive radiology match in recent history. And the competition for radiology spots will only continue to rise in the coming years. Clearly, the fear that AI will replace radiologists didn’t land in the world of medical students and radiology residents. I see two reasons for this:
- The radiology community did a terrific job at shedding light on the issue of AI and radiology early on. Leaders in academia published papers and gave talks about how AI could augment, not replace, the radiologist. The ACR provided education and insights into how AI is catapulting the field forward, with the radiologist at the helm.
- Medical students intrigued by radiology usually have a good understanding of how technology operates in the field of medicine. Whether its data science or engineering, a large portion of students entering the field of radiology have a comfortable relationship with how leveraging technology can augment patient care and quality.
During the RFS meeting, radiology residents attended a presentation by ACR Data Science Institute® (DSI) Chief Medical Officer Bibb Allen Jr., MD, FACR. By providing a high-level overview of the current state of AI in radiology, Allen offered residents insights into how they can use AI in their practices. Many residents, and perhaps attendings, who wish to begin exploring this domain struggle with where to begin. Radiologists are not meant to be coders, and initially, many may believe that to assist in the adoption of AI one needs a computer programming or engineering background. However, focusing on a few general principles may assist radiology residents in becoming more comfortable with innovations in the AI space.
Understanding Patient Outcomes and Clinical Viability
The implementation of technological innovation within healthcare must revolve around improving patient outcomes. That is the bedrock of innovation in medicine. When assessing the utility, value, and function of any AI algorithm, radiologists can be invaluable in determining the clinical pathway and management steps involved. Specifically, what steps in management protocol, diagnostics, or treatment are directly impacted through utilizing the software? It is not uncommon for non-clinicians to be making those decisions, even without a clinical background. As such, radiologists can impact the adoption and assessment of a new AI-based algorithm by developing a methodical calculus that demonstrates the efficacy and necessity — or the lack thereof — for any innovation in the AI arena.
The uniqueness of health economics as a field of scientific study rests in its ability to synthesize variables involving patient outcomes, reimbursement strategy, clinical research, and health policy. Granted, residents don’t go into medicine to learn about International Classification of Diseases codes and Diagnosis-Related Group analytics, but having a broad understanding of these pillars can assist in evaluating the legitimacy of an AI software in medicine. The ACR DSI’s free, on-demand learning hub offers a series on “Bringing AI to Practice,” including one on the AI value proposition. The ACR also offers sessions on AI-related economics and regulatory topics at the ACR Imaging Informatics Summit and in the e-learning hub. There are also other videos on regulatory guidance related to AI at acrdsi.org. An understanding of cost containment and reimbursements in the field of imaging may assist radiologists in their quest to understand the underlying value of AI in radiology.
Knowing Basic Algorithms
Although having a coding background is not a necessity for a radiologist to be involved in AI research, it is helpful to develop an understanding of how these algorithms work, especially as they pertain to radiologic imaging. The ACR DSI has put together a module with a general overview of these principles (learn more at acrdsi.org/DSI-Services). The goal isn’t to turn a radiologist into a programmer, but rather to help physicians understand how algorithms are built and what their utility is in imaging. Importantly, having this underlying understanding builds more trust in AI from the perspective of the radiologist, as they can now comprehend how and why an algorithm is capable of picking up on potentially missed pathologies.