Last fall, radiologists from across the nation competed against one another for a chance to be recognized at the 2017 Radiological Society of North America conference in Chicago. Punching away at their keyboards night and day, their task was to create an algorithm to predict bone age via X-ray.

Competitions like this are becoming increasingly common as radiology stakeholders put vast amounts of time, money and labor into exploring the seemingly endless frontiers for enhancing imaging through software analytics. Ultimately, the common goal is to extract, analyze and integrate information from images into everyday practice, raising the bar on quality and timeliness of care through technology.

This emerging discipline, called imaging informatics, extends beyond interpreting scans to encompass all aspects of imaging workflow, and has been germinating ever since film transitioned to digital. Today, the expectations for imaging informatics to transform radiology are greater than ever before.

“It’s going to have applications throughout the delivery of the imaging services to patients and communication to clinicians,” Kevin McEnery, director of innovation imaging informatics at the University of Texas MD Anderson Cancer Center, told HCB News. “Over time, you’ll see applications of machine learning and other statistical-based processes to better understand patients. I think you’ll see examples of retrieving information for radiologists, informing decision support, and helping the radiologist to essentially prioritize work lists.”

McEnery, along with other experts, notes that with such innovations on the horizon, now is the time for providers, vendors and other health care players to look over, assess and redefine their own roles in order to effectively implement and nurture these capabilities when they arrive in their practices, enhancing workflow and subsequently, patient care.

‘Synthesizers of information’
The idea of using AI to interpret images and assist in diagnosis is getting closer to becoming a reality, and one specific area where software developers are already finding promise is in deciphering the genomics of a tumor.

“That sort of thing was almost unthinkable just a handful of years ago and now we have a very high accuracy for that tool," said Bradley Erickson, associate chair in the department of radiology at Mayo Clinic. "I think that really expands the value of imaging that’s being done today and increases the amount of information we can collect.”

For all this promise and potential, the advent of machine learning has many skeptics asking if this spells the end of humans in radiology. Or worse, a giant step toward a future that resembles the plot of The Matrix, where computer intelligence enslaves mankind.

“Am I scared it’s going to make the radiologist’s job go away? No,” Cheryl Petersilge, the medical director of integrated content for information technology division at Cleveland Clinic, told HCB News. “Do I think it will change the radiologist’s job? Of course. It’s going to for any imager. I think we might have to be in the role of educating our colleagues about appropriate uses of imaging. We should be doing this now. I think we’re going to be the synthesizers of information that the computer presents to us rather than the gatherer of information and then the synthesizer of information.”

Imaging technology experts believe that machine learning will act as an assistant, retrieving information that will aid radiologists in the assessment of a patient’s condition, but not making the final treatment or diagnostic determinations.

But in order for this to effectively happen, providers and practices must bring together players throughout the enterprise to form strategies ahead of time for when these technologies arrive. Such plans require input from nurses, radiologists, IT personnel and practice administrators, each of whom must become familiar with the ins and outs of their facility’s day-to-day activities and enterprise imaging workflow.

“How do they process the orders? How do they track information about that order?,” McEnery asked. “Any practice needs to understand their business in the context of the studies they’re doing and how those studies are interpreted. As the practice gets larger, there’s more complexity in that, because you have different expertise and radiologists and more subspecialty practice. You need to know how you get the study to the subspecialty person in the group.”

Aside from workflow considerations, there’s a much more fundamental factor that prevents some organizations from implementing certain imaging informatics tools: cost.

“It’s much easier for a luminary site or academic facility to invest in bleeding-edge technologies like AI than a small community hospital because they take not just considerable money to implement but a considerable amount of other resources too. These include IT and departmental support to radiologists who have extra bandwidth to evaluate the technology,” Michael Cannavo, an imaging IT consultant, told HCB News.

All purchases need to be cost-justifiable and, until the benefits of a particular system are well documented to show financial value, he says smaller hospitals are likely to keep their distance.

Erickson agrees that big hospitals may be more capable of experimenting with tools that let them automatically identify important components of images. “They may have advantages when they’re managing that data and the storage systems. The basic story is the same regardless of size, but as in many businesses, when you can do it at scale, you can often do it more effectively.”

Strategies and cost considerations for the enterprise
One of the key challenges with imaging informatics is figuring out who is responsible for taking initiative and leading the charge. Different hospitals have come up with different answers to this question, and the key is playing to your strengths.

“If IT views its role as mostly purchasing software and implementing it, and keeping the computers running, then doing something like AI may be challenging for them,” said Erickson. “In that case, the clinical departments are going to have to take on a stronger role in developing that expertise. If the IT department has an experimental arm then that may be where the expertise lies.”

Once a facility determines who is in charge of these operations, staff can begin training on how to interact with the new tools. This extends beyond just knowing how the technology works, and requires staff members to know what computers can or cannot do so that they are able to direct and assist the technology in conveying and representing information meaningfully.

One example involves free text versus structured text, with the use of AI algorithms able to decipher information more quickly from structured reports. Such reports can break down and require staff to use free text to assist with data computation, a feat that requires knowledge of how the algorithm works.

“A practice needs to understand where an off-the-shelf solution works for them and to what extent it needs to be customized versus how the practice itself needs to change,” said Petersilge. “At the end of the day, it’s about the patient receiving an examination with a report generated and how the institution makes that process efficient, safe and effective.”

Reaching this point requires a united front with all parties in agreement on how far they want to proceed with their enterprise imaging strategy, and a thorough understanding of what is financially feasible.

“Maybe it’s just sunsetting a small PACS that’s used by women’s health and incorporating that into the radiology department, which just got a new PACS system and you’re not going to be switching over to a VNA anytime soon,” said Petersilge. “Or you might be at the point where they need to replace the PACS and they want to start looking at a VNA strategy. It’s going to be a lot easier to move ahead quicker when you can put that big foundation in at the beginning.”

The future of AI and imaging informatics
The emergence of never-before-seen capabilities in imaging informatics and artificial intelligence are poised to change the way medical imaging exists within the continuum of care.

The emergence of smart work lists for prioritizing the most critical issues, as well as new guidance to ensure treatment and diagnosis are being done correctly, are poised to impact every facet of the hospital, along with new capabilities, such as 3-D viewing and printing, and innovations in older ones like RIS and PACS.

“Radiology departments are beginning to understand how crucial the electronic medical record can be to their practices,” said McEnery. “By integrating the entirety of the patient’s clinical presentation into their workflow, imaging departments are discovering new efficiencies that also provide improved satisfaction to the referring clinician and the patient.”

While the hype for AI may be bigger than ever, providers must understand that the impact of AI is still in its infancy with more research needed to better understand its capabilities and limitations.

“It remains to be seen if AI will be marketed as a stand-alone solution,” said Cannavo. “At this stage in the game, there are too few areas that AI has been approved for use on to make it financially viable.”

Still, the emergence of this groundbreaking technology underscores the need, more than ever, for imaging departments and hospitals to evaluate their workflow efficiency and what is needed to extract, synthesize and apply the most essential information to a patient’s care and treatment.