The utilization of artificial intelligence (AI) to assist with image interpretation may generate much of the headlines in radiology analytics, but it's workflow-enhancing tools -- some based on AI and some not -- that currently offer the most proven value for today's world of radiology.
To better manage costs and improve efficiency across all imaging departments in an enterprise, healthcare institutions are increasingly turning to analytics applications that provide business intelligence -- enabling better understanding of the radiology department's working processes at scale and driving operational improvements.
Some examples include enterprise-wide fleet management services across image modalities, staff tracking and training, radiation dose management, image quality analytics, and even analyzing patterns in missed patient radiology appointments, according to Steve Holloway of Signify Research. Other notable analytics trends where AI can help include the development of "smart" tools that can provide load balancing for radiology worklists in larger networks.
AI is also actively being deployed in triage applications -- identifying and moving high-profile cases to the top of the radiologist worklist. The ability to hasten a patient's intervention has a clear benefit in a value-based care environment, Holloway noted.
"Anything where there's a clear and actionable sort of improvement in speed of care and speed of diagnosis is a definite driver, and there are solutions out there today that are starting to be deployed for that," he said.
Business analytics will become valuable as radiology departments continue to seek efficiency improvements, according to Dr. Cheryl Petersilge of Vidagos Adivsors.
"As they say you can't improve what you can't measure," Petersilge said. "I believe predictive analytics will become more widespread as organizations realize the power they have to offer -- for example, in helping match staffing to needs."
Any application that increases revenue and reduces cost is of interest to healthcare institutions today, said Michael J. Cannavo of Image Management Consultants.
"Getting optimal value from each [full-time equivalent] is many cases the only driver being considered," he said. "Most facilities are just starting to get an understanding of how analytics can be used so they aren't nearly being optimized as often as they should."
AI bridges the gap
Although having access to data from a laboratory information system or digital pathology system is beneficial for radiologists, lack of access won't stop them from reading a study without it, Cannavo said. However, deep-learning technology can help bridge this gap by combining and analyzing data from several different clinical applications beyond radiology. With high accuracy and processing time of less than a second, these algorithms can provide analysis that can help radiologists make more informed decisions, he said.
"[AI] can take data from multiple clinical systems and make a decision based on the combined findings of all systems and the diagnostic images versus just the radiology images alone," Cannavo said. "This usually doesn't change the radiologist interpretation, but provides a much higher confidence level in the findings when AI and the radiologist both agree."
Vendor-agnostic AI analytic tools also have been developed, enabling these software applications to work with third-party PACS or imaging systems, Holloway said.
Compared to analytics that provide business intelligence, clinical analytics for image analysis is more challenging to develop and implement. Many unanswered questions remain, including how to develop AI algorithms without bias, who should pay for the software, and how to deliver the applications to customers, Petersilge noted.
"I see most [enterprise imaging] vendors developing a marketplace for distribution of AI because they want to be prepared, and I hear lots of people asking what is the standard of care regarding AI," she said. "To me, from a realistic perspective, we are very early on in this journey. There is no standard of care."
As a result, organizations need to proceed very slowly and truly understand what they are deploying, Petersilge advised.
"In the end, I do believe we will have the AI-enabled radiologist with AI performing many of the routine tasks, freeing up the radiologist to engage in more critical thinking activities as patient (not image)-focused members of the care team," she said. "I embrace this future vision."
Demonstrating cost justification remains an important issue for adoption of AI algorithms, and resolving this challenge will be critical for the long-term success of AI software developers, said consultant Herman Oosterwijk of OTech.
Can the cloud help?
The use of the cloud in enterprise imaging should help improve the flow and accessibility of data and is likely to be a net benefit for analytics -- especially for workflow-enhancing applications, according to Holloway.
"If you are going into an on-premises model across a large network, you generally have quite a few data silos," he said. "So getting those all to be patched together and the [analytics] application [implemented] is quite an undertaking with an on-premises model and lots of legacy applications. Generally, if you're shifting to a cloud model, you tend to do a little bit of work upfront there to move up to the cloud in a coordinated platform-based manner, and, therefore, you should be able to query that platform much more easily in terms of data access."
The cloud also provides access to more computer processing power. Many AI algorithms that have been developed by software developers aren't well integrated into PACS, so images are being sent to the AI on a cloud-based marketplace or platform. The AI results are then sent back to the PACS.
"It's a model that's being used a lot already," Holloway said. "What we expect to see over time, though, is the better algorithms being embedded more and more into the core diagnostic platforms."
Analytics will continue to be a growing market in radiology, but how fast the market grows will depend on what kind of return on investment is shown, Cannavo said.
Overall, analytics tools -- including AI -- will likely be integrated more into broader imaging IT platforms in the near future, Holloway said. In addition, there will likely be market consolidation among AI vendors, due to acquisitions and the exits of companies that don't get additional funding, struggle to prove a return on investment from their offerings, or fail to receive U.S. Food and Drug Administration (FDA) clearance.
"I think [we'll see] reduction in the number of solutions and richer and more mature solutions from those who survive," he said.
The number of radiology AI software approved by the FDA will continue to steadily increase, and deployments will see a slight growth as customers get comfortable with the technology, according to Holloway.
"Operational workflow tools should grow more rapidly from an actual deployment and payment perspective near term, but longer term, obviously, the AI piece has more of a potential as an overall market because it has a far greater impact on return on investment long term," he said.