The Reading Room

Leveraging the algorithm: How imaging leaders can benefit from early AI adoption

by Lea Halim and Ty Aderhold

Despite a limited number of current applications, Artificial Intelligence (AI) has taken the radiology world by storm. Every week, there's another news article on AI's potential to change radiology. What forces are driving this focus?

Learn the latest imaging market trends for 2018

The focus on AI stems in part from a fear of AI replacing radiologists—this sort of concern exists in other professions, medical or otherwise. However, for some imaging leaders, the focus on artificial intelligence is less on how to avoid it, and more on how to leverage it successfully as an early adopter.

Radiologist replacement? Not so fast

Before we dive into early applications of AI in radiology, let's quickly touch on a couple of reasons as to why it'll be a long time before AI replaces radiologists:

  • AI's "what's wrong with this picture?" problem. While AI technology can regularly identify objects in pictures, it's still a long way off from being able to place objects in a picture in context with one another to identify what's "wrong." At least in the near future, AI algorithms will continue to struggle with broad questions or applications. So, it'll be  a while time before AI can conduct a complete read of a radiology image.
  • Major regulatory barriers. Deep learning algorithms—the type of AI most commonly used in radiology—usually operate as a "black box." When these algorithms make a prediction, researchers have little understanding of what led the algorithm to make that prediction. This lack of information presents a major hurdle to FDA approval. Until more deep learning algorithms can document the steps taken to reach their conclusions, AI remains far removed from making independent clinical decisions.

Should radiologists embrace or fear machine learning?

Early applications for AI in radiology

So, if AI won't be independently reading radiology images any time soon, how can imaging leaders leverage developments in AI? The answer, in large part, lies in repetitive, non-complex tasks that can be accomplished by AI to improve quality and efficiency. Here are a few examples of current use cases for AI in radiology:

1. 'Intelligent' speech recognition

2. Worklist management and exam escalation

3. Quantitative measurement solutions

4. Clinical information briefings

The future of AI in radiology

In the near future, early adopters of AI will likely see significant efficiency and quality gains that will separate them from competitors. Within five to 10 years, AI will most likely handle all rote daily tasks done by radiologists, allowing them more time for patient care and findings review.

In the longer term, AI has the chance to transform the role of the radiologist.  As more algorithms are developed to make clinical decisions, such as the analysis and characterization of a nodule, radiologists may experience more fundamental changes in how they read and interpret images.

However, even if AI algorithms do become sophisticated enough to make complex reads, there will always be an important role for radiologists to play in understanding, interpreting, and reviewing the reads this futuristic AI software makes.

Article based in part on conversations and presentations from the 2017 Economics of Diagnostic Imaging Conference held in Arlington, Va. 


Get the imaging leader's cheat sheet

Imaging is impacted by almost every health system change—even when not targeted directly. Stay up-to-date with our cheat cheet.

Download Now

Next, get our artificial intelligence cheat sheet

Recent advances across several industries and task types have focused new attention on the field of AI. Download this cheat sheet to learn more about what AI holds in store for health care.

Get the Cheat Sheet