To become leaders in AI, radiologists must address a variety of challenges

Artificial intelligence (AI) is one of the biggest topics in healthcare today, and the authors of a recent analysis published in the Journal of the American College of Radiology wrote at length about radiology’s role in the development and implementation of these state-of-the-art technologies.

“The radiology community has played a leading role in propelling medicine into its digital age and now has the opportunity to become a leader in exploring medical applications of AI,” wrote lead author James H. Thrall, MD, department of radiology at Massachusetts General Hospital in Boston, and colleagues. “The tens of millions of radiology reports and billions of images now archived in digital form exemplify the concept of ‘big data’ and constitute the required substrate for AI research.”

Thrall and colleagues covered considerable ground, including the various challenges specialists face as they work to use AI to their advantage. “None of the challenges alone will be a showstopper but all may slow progress and need to be addressed,” the authors wrote.

The analysis separated AI-related challenges into three primary categories:

1. Circumstantial challenges

Concerns that AI will someday replace radiologists are nothing new, but they still persist. While some experts think such fear is hyperbole, Thrall et al. point out others still think radiology’s time “as a thriving specialty” could be coming to a close. What’s more likely, the authors add, is AI will be used to improve the quality and efficiency of radiology instead of taking over the industry completely.

Radiologists must also address the fact that researching AI requires significant manpower and funding. “This can be addressed by recruiting scientists with backgrounds in AI into radiology and through the educational programs already being undertaken by radiology professional societies,” the authors wrote. “Historically, once an area is recognized as important, capable people quickly populate it, so this is not likely to be a long-term issue. Practicing radiologists will need to learn about AI but will not need to become experts in AI research or design of AI programs to beneficially use AI-based results.”

And as far as funds are concerned, the industry’s best hope is leaning on “forward-looking departments and institutions” in these early stages of development and implementation.

2. Intrinsic Challenges

Uncertainty about processing speeds is just one of the many intrinsic challenges associated with AI and radiology; you can’t do much with AI working with limited resources. However, Thrall and colleagues believe this will take care of itself over time.

“At present, computing systems fast enough to supply results in a clinically relevant time frame for emergency or urgent diagnoses are not generally available in medical institutions,” the authors wrote. “However, this is not likely to be a practical problem going forward because of rapid development of lower-cost graphics processing unit–based computing systems and easy access to cloud computing solutions.”

Another natural challenge is determining how to validate results. AI programs need a “source of truth” to make the proper diagnosis, but what kind of data should be used?

The authors offered clarity on this issue. “The source of truth can come from patient outcomes or results of other ‘gold standard’ testing methods apart from the imaging method being studied, but the source of truth used must be rigorous and should be explicitly stated for each AI program that is developed and used clinically,” they wrote.

3. General limitations of AI

Other challenges are related to the basic ups and downs one might expect when working with AI. It requires massive datasets, for instance, and providers may struggle at times to collect enough data to truly make a difference. Not having enough data can result in inaccuracies, but so can “overfitting” that data—it’s all about finding the right balance.

The authors added that specialists may also run into issues because imaging data includes so many variables.

“The biggest limitation for AI in imaging may be inherent limitations in defining normal versus abnormal in continuously variable biologic data,” the authors wrote. “Ranges for normal are set as a certain number of standard deviations from the mean of a supposedly ‘normal’ population. This means for any test or measurement, a given percentage of truly normal people will have ‘abnormal’ results. Investigators in AI will face this conundrum where nominal criteria for normal versus abnormal can be difficult to define when, for example, setting limits for organ sizes.”

Even though all of these challenges exist, the authors sound confident radiologists will seize the day and make the most of this opportunity to lead. “It is not yet clear what the full or final role of AI methods will be in imaging or their impact on radiologists,” they wrote. “What is clear is that AI provides a promising new set of tools for interrogating image data that should be explored with vigor.”