3 things radiologists need to realize AI’s full potential

AI is expected to impact radiology more than perhaps any other medical specialty. In healthcare, though, nothing is a given.

The authors of a new analysis published in the Canadian Association of Radiologists Journal explored Canada’s place in the research and development of AI technologies—and what is necessary to ensure radiologists don’t waste this game-changing opportunity.

“With an educated and diverse population, and universal health care, Canada is well positioned to be a leader of medical imaging–related AI research and implementation,” wrote authors Christian B. van der Pol, MD, and Michael N. Patlas, MD, of McMaster University in Ontario, Canada. “The realization of the full potential of AI in radiology will however depend on advancements on several fronts.”

According to the authors, these are three crucial areas when it comes to making the most of AI’s limitless potential:

1. Integrated medical data repositories

“Well-labelled and large data sets of images and other clinical information are needed,” the authors wrote. “Although there are anonymized image data sets available to the public, organs and pathologies are generally not yet segmented and relevant clinical data are often not available.”

AI models, like human physicians, must consider a titanic amount of patient data when making a diagnosis. The more information made available, the more accurate an algorithm can be.

2. Protected patient privacy

Removing personal health information should be a top priority during every step of dataset development. “Customized solutions addressing the nuances of each centers’ data storage” will be needed for this to happen in a consistent, reliable way that works for everyone involved.

“It is critical for radiologists, data scientists, and others to work closely together with information technology departments to establish pathways for database creation,” the authors added.

3. Transparency and accountability

Healthcare providers are often skeptical of technologies they do not fully understand, a serious challenge facing the implementation of AI on a global level. One way to combat this issue is by working to make the AI decision-making process remaining as transparent as possible, allowing other researchers to replicate one AI model’s performance with their own algorithm.

The authors also noted that radiologists must understand that the benefits of using AI “clearly outweigh” any risks. In the specialty does not work to make this point clear, radiology could face “a risk of being sidelined” as AI continues to evolve.

“Our specialty has benefitted from prior technological advancements including the transition from film to digital imaging, PACS, and implementation of voice recognition software,” the team concluded. “Artificial intelligence provides another exciting opportunity: the future of radiology in Canada is bright.”