AI Is Not the End of Radiology’s World

Artificial intelligence (AI) tools are not the Dooms Day machines for radiology that many naysayers would have you believe, said one industry expert.

In an opinion article published recently in Radiology: Artificial Intelligence, John Banja, Ph.D., professor and medical ethicist in the Center for Ethics at Emory University, addressed the often negative ballyhoo that has surrounded AI in radiology for several years – that the technology will soon replace the radiologist, putting you out of a job.

“What seems ethically imperative at present…is a steady and informed rebuttal of AI hype, especially as it is aimed at image-dependent technologies like radiology,” he said. “Today’s hospitals simply cannot function without radiologists, who are core to their diagnostic functions.”

It’s True – Interest in AI Is Growing

The past decade has seen an explosion of development around AI tools, Banja said, particularly in research. In 2008, there were only 100 publications in radiology literature that focused on AI tools specific to the industry. But, by 2018, that number had ballooned to approximately 700 papers.

In the same vein, investment in AI has also taken off – since 2010, the market has seen applications for 154,000 patents. Between in 2018 and 2019 alone, organizations that use AI expanded from 4 percent to 14 percent. And, according to the Worldwide Semiannual Artificial Intelligence Systems Spending Guide, the fastest AI growth between 2018 and 2022 is expected to be in healthcare with image recognition software being some of the most familiar models.

Related Content: Radiology Leaders Urge Proceeding with Caution with Autonomous AI

But, that still does not mean AI is in a position to upset the radiologist.

“It is a long way from observing the success of an AI model in a research setting to implementing it in routine clinical practice,” he explained. “And, it is a much longer way still to replacing human radiologic expertise.”

The More Likely Role for AI

Even though there is vast potential for AI in radiology, the technology still has some significant limitations. In most cases, the algorithms used in the industry have not been tested either on heterogeneous patient populations or across a variety of modalities. Consequently, the strength of their generalizability is in question.

This means the applications for AI tools are relatively narrow within radiology, lending the technology to be a radiologist’s “amiable apprentice” rather than an “awful adversary,” Banja said.

Additionally, AI’s current function is to identify potential problems and bring them to a radiologist’s attention. The next steps of decision-making, though, are outside its capabilities, keeping the radiologist’s position secure.

“Even if a plethora of models appeared that could detect and evaluate subtle lesions as well as an experienced radiologist,” he said, “those applications wouldn’t be able to determine whether to image or not, how to best provide protocol for an examination, teach, and communicate findings to clinicians and patients.”

A Development Hurdle

It is possible that one of the biggest obstacles to wider – and more complex – AI use is technology itself, Banja said. According to Wim Naude, a business professor from The Netherlands, the long-standing pattern of computer power doubling roughly every two year is largely winding down, potentially making it harder for AI to advance to the point where it can take over the wide variety of tasks that fall to the average individual radiologist.

Currently, the technology needed to achieve this level of multi-tasking does not functionally exist, Banja said. Reaching the goal would mean replacing AI’s silicon-based transistors with technologies that are still nascent, such as organic biochips or carbon nanotubules.

Overall, he concluded, there are many unanswered questions about how radiology, as a whole, will be able to seamlessly mesh AI solutions into current workflows and successfully navigate many of the other issues that currently face the industry.

“It remains anyone’s guess as to how AI applications will be affected by their integration with PACS, how liability trends or regulatory efforts will affect AI, whether reimbursement for AI will justify its use, how mergers and acquisitions will affect AI implementation, and how well AI models will accommodate ethical requirements related to informed consent, privacy, and patient access,” Banja said.

Once these issues are addressed, he said, it is possible that, rather than succumb to an AI take-over, radiology could ride the wave to a new height of professional excellence.