Sources of Value in Radiology AI

Dec 17, 2019

Christina Nelson, Communication Specialist

 In 2016, AI vendors made their debut at the Radiological Society of North America’s 103rd Scientific Assembly and Annual Meeting. As the years have passed, the impact AI has had at RSNA has rapidly grown, with the number of AI vendors exhibiting at RSNA reaching over 100 this year. With more vendors than ever demonstrating AI solutions in McCormick Place, the question of if AI will have a future in radiology has shifted into what value AI will bring to patients and providers.

Assessing radiology AI

The potential of AI to bring valuable benefits to medical imaging, including reducing radiologists workload and increasing productivity, has been asserted over the past three years. RSNA 2019 reminded vendors to critically examine the value that AI can bring to patients and providers. Ben Panter, CEO of Blackford Analysis, has summarized three areas to evaluate the value of AI in radiology. During his talk, he invited the audience to assess the following three areas:

  1. Increasing radiology efficiency
  2. Adding value for referrers
  3. Replacing or enhancing an existing procedure

Each of these areas highlights the true value machine learning, computer vision, and natural language processing can bring to radiologists when developed with a thorough consideration of how solutions will fit into the clinical environment.

Improving radiologists’ productivity

Diagnostic radiology practices are under pressure to deliver high quality results for increased volumes and often, with lower payments. This creates demand for operational efficiencies from technology. Optimizing efficiency and improving turnaround times can offer benefits to patients in time-critical situations.

For example, our 510(k)-pending triage software can detect intracranial hemorrhage (ICH) on non-contrast head CT images in just a few seconds to help triage and prioritize reading of images for patients with ICH. Other companies have triage software for ICH and other urgent conditions. Some large radiology groups are investing in NLP solutions to improve productivity of report creation and improve compliance with payer requirements for clinical documentation

Adding value to referrers

Some of the most positive feedback we received at RSNA was on our pre-clinical demonstration of an advanced 3D visualization and structured reporting system for coronary artery disease. Collaboratively designed with cardiologists, it will allow cardiologists to show patients their results in a more engaging way than the typical textual report from a CTA exam. With the transition to value-based care, patient engagement and subsequent adherence to therapeutic recommendations could make a significant difference in their outcomes. Creating more engaging experiences for patients, we believe, may lead to improvements in patient satisfaction scores and in their clinical outcomes. We would like to test those hypotheses with clinical collaboration partners.

Identifying new opportunities

Several vendors differentiated from their competitors at RSNA by using AI to perform tasks that have not been previously performed by radiologists alone. As an example, Subtle Medical is developing a suite of software solutions to improve medical imaging during the acquisition phase of the workflow. Their technologies allow hospitals and imaging centers to enhance images from noisy scans and to reduce the time it takes to capture a diagnostic image. This enabled greater economic benefits to imaging centers who can get more utilization from their equipment, and brings value to patients as they spend less time in the machine.

In our experience working with healthcare organizations and medical technology manufacturers to develop customized solutions, we have heard our clients share that they are looking for technologies that solve specified use cases that are especially valuable to their organization. The current generation of FDA-cleared AI devices show progress, and it may take more development before exceptionally valuable applications are available. We suspect that the most valuable use cases that machine learning can be applied to in healthcare have yet to be developed. Exciting new developments are possible when AI computer scientists collaborate with clinical partners.

Moving past the “peak of inflated expectations”

The Gartner Group Hype Cycle for AI in 2019 shows deep learning as just having passed the “peak of inflated expectations.” We are entering a crucial period where conversations between AI scientists and healthcare providers will drive successful innovation. As Dr. Xiaoxiao Liu, our Director of Products and Strategic Development, shared in her talk in the AI Theater at RSNA:

“True innovation starts in meaningful collaborations.”