4 Reasons Why AI Won't Replace Radiologists
March 27, 2018 -- Are you worried about artificial intelligence (AI) eliminating your job? Don't be. Even with the rapid development of AI, the great majority of radiologists will likely continue to have jobs in the decades to come, according to an article published online March 27 in the Harvard Business Review.
There are four reasons why radiologists won't disappear from the labor force, according to Dr. Keith Dreyer, PhD, of Harvard Medical School and Thomas Davenport, PhD, from Babson College.
1. Radiologists do more than interpret images.
They also consult with other physicians on diagnosis and treatment, treat diseases, perform image-guided medical interventions, relate findings from images to other medical records and test results, and discuss procedures and results with patients, along with many other activities. In the unlikely event that AI took over image reading and interpretation, most radiologists could redirect their focus to these other essential activities, according to Dreyer and Davenport.
2. Clinical processes for AI-based image work are a long way from being ready.
Different imaging vendors and deep-learning algorithms are focused on different aspects of the use cases they address, making it very difficult to embed these algorithms into current clinical practice. The American College of Radiology's Data Science Institute is working to define the inputs and outputs for deep-learning software vendors, as well as create a comprehensive collection of use cases to define the clinical process, image requirements, and explanation of outputs. This process will take many years, though, and further expand the role of radiologists in the AI world, according to the authors.
3. Deep-learning algorithms for image recognition must be trained on labeled data -- images from patients who have received a definitive diagnosis.
Deep-learning algorithms have achieved high levels of success in other types of image recognition tasks when they have been trained on millions of labeled images, such as cat photos on the Internet. However, there isn't an aggregated repository of radiology images; these studies are owned by vendors, hospitals and physicians, and imaging facilities and patients. It will be challenging and time-consuming to collect and label these cases to accumulate a critical mass for AI training, according to the authors.
4. Changes in medical regulation and health insurance are required for automated image analysis to take off.
A number of key issues need to be worked out with AI, such as who is responsible if a machine misdiagnoses a case or how healthcare payers will reimburse for an AI diagnosis. AI radiology machines may need to become substantially better than radiologists to drive the required regulatory and reimbursement changes, according to the authors.
"Radiologists, like the lawyers, financial planners, accountants, and other professionals who are seeing some job tasks be performed by smart machines, will find changes to, rather than replacement of, their current jobs," Dreyer and Davenport wrote.
That doesn't mean radiologists won't need to adapt or learn new skills and work processes. Those who refuse to work with AI may find their jobs threatened, according to the authors.
"There are substantial medical and productivity benefits to be gained from integrating AI with radiological practice," they wrote. "The productivity improvements may even mean that radiologists can spend more time doing what many of them find most fulfilling: consulting with other physicians about diagnoses and treatment strategies. If the predicted improvements in deep-learning image analysis are realized, then providers, patients, and payors will gravitate toward the radiologists who have figured out how to work effectively alongside AI."