How AI Can Help Radiology Groups Address Reimbursement Changes

RBMA, ACR and other groups are actively lobbying Congress to mitigate the impact of these changes, but with bipartisan focus on this issue after years of pushing off significant changes, it is hard to see how some sort of reform will not be enacted. In the near term, lobbying efforts will hopefully prevail, but these are likely to only delay ongoing reimbursement pressures at best. With healthcare spend at roughly 17% of GDP, every player in healthcare will have to accept lower reimbursements over time.

With this in mind, radiology groups will be forced to seek innovative solutions to maintain current levels of revenue, or, dare we say it, to grow.  If groups are unable to maintain revenue, they will need to cut costs.  Radiologist salaries typically compose 70-90% of practice costs, with IT, billing, and G&A costs making up the balance. Some cost efficiencies can potentially be gained here, but aside from cutting personnel or lowering salaries, most adjustments will not be significant enough to offset an 11% reduction in reimbursements.

Instead, increasing productivity becomes the most viable strategy.  However, radiologists are already quite efficient, and rather overworked. Study volume is often so high that fatigue and burnout are very common concerns amongst radiology practices. According to research from Mayo Clinic, radiologists already need to review one image every 3 to 4 seconds to meet workload demands, and per recent survey data, 88.4% of US radiologists are at or over capacity, feeling overworked and overextended. This leaves radiology groups in a difficult position - how can they simultaneously increase radiologist productivity, while also helping radiologists avoid burnout? The answer lies in the bellwether that consistently impacts our field before most others: advances in technology.

As a profession, we’ve been at the forefront of adopting new technology to refine our workflow: PACS, voice recognition, smart worklists, etc. By streamlining our workflow, we offload more of the repetitive activities that tax our time and attention, freeing up our ability to focus on expert image interpretation and diagnosis. Fortunately for us, the latest technology - artificial intelligence - is maturing at just the right time to allow us to automate more tedious aspects of our workflow in ways that weren’t previously possible.

When the public thinks of AI in the context of radiology, they tend to imagine algorithms analyzing imaging studies from beginning to end, and somehow replacing radiologists. AI is a number of years away from having a practical end-to-end impact on image analysis and diagnosis in real world clinical practice, due to a wide range of technical and market challenges. But even when it does mature to this stage, AI will augment what radiologists do, not replace them. Just as other professions, such as law, financial planning, and accounting are seeing some repetitive tasks performed by smart machines, radiologists will find AI gradually enhancing their current roles.

There are a few imaging applications of AI that are already proving useful for very specific needs; for example, some pulmonary nodule detection algorithms are improving accuracy in new lung cancer diagnosis.  Other imaging applications work outside the radiologist’s workflow; one promising app was recently granted a Medicare New Technology Add-on Payment for improving stroke care, by shortening the turnaround time on stroke screening for neurologists and neurosurgeons.  This often bypasses the radiologist altogether, though it can be very useful for stroke centers.

For radiology groups, what matters much more is how we can improve radiology workflow right now with AI. For example, there are a number of AI solutions that focus on triage or worklist prioritization through identification of critical imaging findings. Other AI solutions focus on simplifying parts of the workflow before image interpretation, such as image acquisition, streamlining access to relevant patient history, or aiding in report generation. In fact, the ACR has a fantastic list for the many diverse use cases of AI in streamlining radiology workflows.  

One such AI product my company has built reads the Findings section of a radiologist’s report as it is being dictated and composes a customized Impression section using the individual radiologist’s dictation style, that can then be edited and signed off. In addition, it pulls in all significant findings, highlights pertinent negatives, and answers the main clinical question. In doing this, it helps reduce radiologists’ mental fatigue, increases efficiency, and improves accuracy.

As new products are developed, an important factor in AI success is ease of deployment. For example, AI products should ideally integrate seamlessly with radiologists’ preferred PACS, voice recognition, worklist and RIS, and operate without any additional clicks, hotkeys, or new windows. After all, if the central aim of incorporating AI into the workflow is to increase productivity while also curbing burnout, it is of no use if it’s not readily adopted by radiologists. A solution that does not require heavy change management and ongoing training is much preferred for radiologist adoption. 

The practice of radiology has long been at the forefront of technological innovation, but we haven’t seen as much in the way of major advances since the advent of PACS and voice recognition. Artificial intelligence is the next step in a long line of new technologies that have been championed by innovative radiologists, and especially when it comes to improving radiology workflow, it’s a technology we can start applying now. The most savvy radiology groups are already doing this, but more of us can easily adopt this technology to address the challenges of looming reimbursement cuts without overtaxing radiologists.

Originally published in the January edition of RBMA Bulletin