AI is a little acronym for one of the biggest changes taking place in imaging. Over the last year, it is fair to say that AI has presented as a mix of unprecedented buzz, hyperbole, and finally – something that could actually bring meaningful value to providers.

And although headlines like “Will AI Replace Radiologists?” have not shown any signs of slowing down, industry insiders have taken a firm position that this is really more about augmenting human intelligence than replacing it with something that stands alone and makes decisions.

“We are challenged to communicate a clear message of the continued bright future for radiologists amidst the hype around AI replacing radiologists,” Dr. Geraldine McGinty, chair of the American College of Radiology (ACR) Board of Chancellors told HCB News. “In general, radiologists are innovators by nature, and our community is engaged and ready to use these new tools to improve quality of care and outcomes.”

Indeed, AI and its underlying machine learning will be a key focus at this year’s 2018 RSNA meeting, with expanded learning and research opportunities for stakeholders to sift through the promotional language and see what’s really happening. RSNA’s leadership believe that imaging is now entering its “fourth generation”, characterized by big imaging AI investments that will soon make their presence felt in regular imaging facilities throughout the country.

In June, RSNA even announced the creation of a new journal titled Radiology Artificial Intelligence.

Not that AI doesn’t pose challenges. Dr. Kurt Shoppe, the chair of the ACR Reimbursement Committee, told HCB News in July that there are some details to work out related to radiologist payment policies and who will cover the overhead costs of AI implementations. While those specifics remain somewhat uncertain, Shoppe echoed a sentiment that has become familiar in the industry: that AI would help radiologists reduce time on mundane, repetitive tasks and allow more time to spend with patients and hospital teams.

Dr. Matt Lungren, associate director for artificial intelligence and imaging at Stanford Medical Center, said that one of the biggest challenges with creating powerful AI systems is having the abundance of data required to train the model.

“They’re very hungry for data,” Lungren told HCB News, adding that privacy laws and siloed medical centers can hamper access to those large quantities in imaging where other industries might not run into those challenges, causing a “critical bottleneck in the advancement of the field.”

He also stressed that real-life AI application could be "brittle," meaning it may be able to provide an impressive demonstration of an application that works well under specific OEM and or data system conditions, but that doesn’t mean that performance automatically transfers to any individual hospital or clinical OEM platforms.

Fortunately, work is being done to address these challenges. For example, the National Institutes of Health is making its largest data sets available to scientists around the world.

He reminds his imaging colleagues to evaluate emerging AI platforms critically. If, to take an example, an AI application claims to lower radiation dose exposure, that metric is measurable. This cautious process is how he and his Stanford colleagues are deploying AI solutions – and much of their work will be detailed before RSNA 2018.

According to Lungren, Stanford researchers are reviewing several imaging AI applications, including related to lowering dose, image reconstruction, and lower contrast, and will be sharing some of their work at RSNA.

"We're having a lot of fun seeing how (AI) models, and humans interact,” he said. “But our models may work awesome at Stanford, but is it going to work at Duke or Cornell? We don't know; we have to find out."

Leadership that reflects the times
AI isn’t just a technology issue, it’s also a leadership issue, according to Dr. McGinty.

“The rapid acceleration of research and innovation around data science has presented both challenges and opportunities for imaging,” she said. “Through the development of standardized use cases – the problems that are most worth solving and most amenable to an AI solution – we are influencing this rapidly evolving industry to ensure that we fully leverage this opportunity.”

The ACR has established the Radiology Data Science Institute (ACRDSI) to work with all players in the imaging field – from radiologists to industry leaders to government agencies – in the interest of establishing a unified approach to harnessing these emerging tools. ACRDSI is led by a team of AI savvy radiologists and informatics and IT experts.

The need for groups like the ACRDSI may be indicative of how much uncertainty still exists in the emerging AI landscape. One trouble spot is in a lack of training for AI in medical schools.

Some experts argue there is too much reliance on computer scientists to bridge the machine learning, or AI, gap in medicine. In fact, a prospective review of the existing literature published in the journal NPJ Digital Medicine earlier this year found exactly that.

“By the time medical students become researchers or fellows, it’s probably too late,” Dr. Vijaya B. Kolachalama, Ph.D., lead author and assistant professor of medicine at Boston University School of Medicine (BUSM) told HCB News.

And what will this computer science emphasis mean for the radiologists who are utilizing it? Dr. Richard Gunderman, the John A. Campbell Professor of Radiology at Indiana University, is interested in the ethics of radiology and wonders if imaging leadership is doing enough to deter students from the temptations of billing fraud.

With PACS and EHRs, radiologists don’t get the face time they used to with referring physicians – or even each other. Some radiologists even now work out of their homes. With social isolation more and more built into imaging, leadership needs to keep eyes, ears, and heart on the human factor.

Although many experts hope that AI will help bring radiologists back into the hospital community and out from behind their screens, new levels of automation could also exacerbate the risk for social isolation.

“Our love for technology can become too big,” Gunderman told HCB News. “Our appreciation for interaction can atrophy. While we’re mining the bleeding edge of AI, leaders must remember to advocate for a professional identity and a robust community of professionals. We must actively advocate that imaging is not just an economic activity, but a patient care endeavor.”

An update on value-based payment models
In the past year, imaging thought leaders began defining imaging value under the evolving Medicare Access and CHIP Reauthorization Act of 2015 (MACRA) and the Merit-based Incentive Payment System (MIPS) compensation models.

While the details of MACRA and MIPS are starting to emerge, many radiologists are unsure what is involved in participating. The temptation might be to continue for as long as possible in the fee-for-service environment, but experts argue that’s not a viable long-term strategy.

“When you think about it, imaging is expensive,” said Nelly Ganesan, senior healthcare consultant with Avalere Health. “The reimbursement they are submitting to CMS is pretty high. Even the four percent negative payment adjustment to an imaging center is going to be significant.”

According to Ganesan, the number of imaging measures in the MIPS program provides an opportunity for radiologists to do well in the value-based model. She expects that most radiologists will participate in the MIPS model, rather than in alternative care models (ACOs).

“I do think the opportunity and ability for radiologists to receive the upward payments adjustments is pretty likely, given the ease of participation for them,” said Ganesan.

At last year’s RSNA meeting, a panel of physicians urged radiologists to get onboard with emerging with MACRA, MIPS and ACOs. Dr. Jim Whitfill, chief medical officer, Innovation Care Partners, and clinical associate professor, University of Arizona College of Medicine, spoke about the role of ACOs in meeting the new value-based compensation goals.

We checked in with Whitfill, who is also the new chair of the Society of Imaging Informatics in Medicine (SIIM), to find out how the ACO model is working. He reports that his practice, Scottsdale Health Partners, is scheduled to receive a $19.5 million payment from CMS for achieving savings under the ACO model. Their success represented the third-best result in terms of savings percentage in the U.S. this year, and the sixth best, based on total dollars saved in the delivery of primary care.

"Whereas many health systems need to have one foot in the fee-for-service canoe and one foot in the value-based canoe, we, at least in the clinically integrated network, are in the value-based canoe – that’s all we do,” said Whitfill. “We are owned by a larger health system that has to do both, but we’re a wholly separate subsidiary.”

Despite their success, Whitfill cautions that he believes the wholesale pivot from fee-for-service to value-based reimbursement is still several years away. But, Whitfill gives a lot of credit to the radiologists in their network of 1,800 physicians who are responsive in supporting value-based principles.

This might include, for example, responding quickly with an exam and reading on a Friday afternoon to keep a patient from being admitted to the hospital over the weekend.

Informatics and the road to higher-quality healthcare
Whitfill, who describes himself as an informatician, said there is still a lot of groundwork that needs to be laid concerning the underlying informatics platforms, but acknowledges that radiology has been on the leading edge. In many ways, AI and informatics are related subjects and bring to mind similar concerns.

“What we’ve seen is that these tools are absolutely going to be disruptive and innovative, but they are not going to replace human beings, they’re just going to change what we do,” said Whitfill. “It will probably be true that the humans who use machine learning will win out over humans who don’t use machine learning.”

He and his network and practice have eight different value-based contracts covering 100,000 lives. They are active in implementing advanced informatics engineering to large healthcare databases. This includes a new project to create machine learning/natural language for lab and test data across 62 electronic health records, three different labs, and multiple radiology practices.

“What we believe is that if we can find the hidden patterns in all that siloed data, we’ll be able to predict with better accuracy who is going to get sick, and more importantly, who we can help before they get sick,” Whitfill said. "We think for a value-based organization like ours it will be a huge win for the patient and a huge win for the practice economically."

Informatics, he said, has been a vital part of their success, and he has been working for a while to roll out these tools in their practice and network. But the reality is that most of their success has come from physician and clinician engagement. The value of informatics is that it provides independent data to focus on process improvement – but the rest is up to people.

"Having our clinicians be the leaders of this organization and charting the path we need to take has been critical in getting the results we have," he added. "The biggest thing that informatics does in value-based care is if you put ten clinicians around a table, they'll give you ten different areas where there is an opportunity to fix the waste they see in their own practice.”

Making imaging more patient-centric
Patients have complaints about imaging. Whether the exams make them anxious or they don’t feel like they’re getting enough feedback during a scan, or they simply don’t understand their reports – there is a lot of room for improvement in the patient experience in radiology.

A number of these issues could be solved by greater collaboration between radiologists and patients; more face time to talk about what’s happening and what it means. And while an improvement to these relationships might be foretold by the growth of AI and sophisticated informatics, the truth is that many radiologists at the front lines today simply don’t have time.

During last year’s RSNA meeting, two radiologists published a study in the journal Radiology where they highlighted this challenge. HCB News checked in with the lead author to see how patient centricity had progressed since then.

“In the past year radiologists have continued to improve their patient focus,” Dr. Jennifer Kemp, vice president of Diversified Radiology of Colorado (DROC), told HCB News. “More and more radiologists are sharing stories on ACR Imaging 3.0 as well as ACR Engage Open Forum about what they are doing to become more patient-centered. Also, both the RSNA and ACR published a Patient-Centered Resident Curriculum this year.”

While reimbursements are increasingly linked to quality metrics tied to patient satisfaction, Kemp stressed that these behaviors also yield benefits to physicians as well, with studies showing physician/patient relationships at least somewhat mitigate burnout.

At DROC, Kemp and her colleagues have introduced a program in which patients are given direct results of lung cancer screening exams via conference call. The process consists of a review of images with the radiologist on a remote workstation minutes after their screening exam. The radiology group has also started a pilot program for patients who seem particularly anxious. Every morning for two hours, the CT and ultrasound technologists offer immediate scan interpretation to such patients.

“The result is that patients are grateful. Radiologists feel more rewarded and part of the care team, thus, less burnout – and technologists are engaged and empowered,” said Kemp. “There are challenges. Time is always the number one challenge to being patient-centric. So we offer these programs during times of the day that we have the slowest caseload.”