One of the big topics of conversation at this year’s imaging and technology conferences – HIMSS, SIIM, and most recently the Ai-Med Global Summit – was the slow pace of AI adoption. The largest buzz, however, centered around large language model generative AI, and the prospects of applying a ChatGPT-like algorithm to imaging applications. The excitement around imaging applications for generative AI was broadly positive and impossible to deny, particularly in its potential for transforming clinical reporting.

But despite this hype at some of the country’s buzziest health IT conferences, overall AI adoption in healthcare remains relatively low. The hype may represent an aspirational ‘north star’ for the near future, but right now, most healthcare and imaging organizations still haven’t been able to take action on the promise of diagnostic AI – either because they haven’t been sold on making the investment, they’ve been impeded by other factors like recovering from the impacts of COVID-19, or they’re simply taking a wait-and-see approach before committing to serious budgets.

Changing this status quo requires changing the conversation. The key to AI adoption in healthcare right now isn’t pie-in-the-sky ambitions that are years off; it’s not offloading clinical tasks to ChatGPT. It’s demonstrating the real-world value of AI in making life easier for clinicians, helping to improve patient care outcomes, and providing quality care at scale. And each small step can improve care, accuracy, and move us closer to truly personalized medicine.

Applying AI to radiology workflows may sound small-scale on paper, but it can significantly make easier a job that has gotten increasingly crushing the last few years – and lay the groundwork for the next wave of AI healthcare applications.

The cost of not knowing

Today, 80% of healthcare data is locked away in silos – making it inaccessible and depriving radiologists and physicians of valuable patient insights.

Radiologists handle nearly 1 billion exams and 100 billion images per year – with only 31,000 radiologists to read them all. That’s 3 million images per year for each radiologist. And as the volume of healthcare data, which is estimated to hit 44 zettabytes in 2025, continues to explode, this silo problem will only get worse.

So will the consequences of siloed imaging data: patients falling ill, disabled, or dying, from otherwise preventable diseases that could be caught with early detection. For example, early detection could help prevent up to 80% of sepsis deaths.

The promise of AI is broad and continuously expanding

Analytic AI can help extract information that the eye and brain may miss, by helping to search patient populations for specific diseases; more quickly and proactively identify life-threatening conditions; and improve the workflows and processes used for collecting, analyzing, and acting on patient data.

Today, AI is most frequently used in narrow ways – focused on administration applications, like staffing, managing patient flows, and anticipating capacity shortfalls. These are important applications that make things easier for clinicians, but they only scratch the surface of what AI can do. Right now, in the imaging analytics realm, most radiologists have only adopted one or two AI applications at most into their jobs. But over time, this can and must change – for the benefit of the clinicians and the patients. Radiologists won’t be able to reasonably keep up with the demands of their jobs without it.

Why radiology workflows can unlock the next step in AI

Applying AI within radiology workflows is the next big leap for artificial intelligence in our profession, using AI to:

  • Help find, measure, track and analyze lung nodules
  • Alert care teams about time-sensitive emergency conditions like brain bleeds, strokes, and pulmonary emboli
  • Enable equitable distribution of studies from the worklist, taking into consideration urgency, radiologist preference and specialty, study complexity, and radiologist capacity
  • Provide a second pair of eyes that helps radiologists catch something critical that we may miss
  • Go outside the scope of why a patient was getting an exam in the first place, looking for incidental findings in other areas or relevant, related instances from their medical histories.

This has enormous implications for patient care, but it can also provide a better experience for the radiologist as well: more dynamically balancing caseloads across radiologists, so no one is overburdened and the right exams are equitably distributed to the right radiologists.

AI may also be able to eliminate some of the tedium often associated with the radiologist’s job. Arbitrary tasks like selecting the right contrast agent for a patient, the right dosage, the correct level of radiation, and the optimal scanning protocol – these are all steps that AI can help guide, automate, and personalize care delivery, at scale, to truly drive improvements in patient outcomes and radiologist satisfaction. Going beyond that, AI may also be used to help efficiently identify and recommend the best imaging protocols for clinicians, based on past scans or outcomes.

Ultimately, embedding AI into radiology workflows results in a workspace where the AI can tell the radiologist what to read next; highlight something they may have missed (an occurrence that happens 3-5% of the time for the average radiologist); and prioritize the patient exams in the worklist that can’t wait. An AI-guided worklist is one that takes into account the radiologist’s preferences, provides them with cases that are equitably distributed, and take into account the clinician’s specialty, the complexity of the study, and the urgency of the case.

And for those urgent cases, radiologists can be assured that they’re getting to the images fast enough so that if, for example, there’s an abnormality on the CT scan, there isn’t so much time that passes that the patient ends up going home and has to be called back in for another appointment. Instead, the AI helps confirm that abnormality that same day, while the patient is still in the doctor’s office, so they can get started on treatment immediately if necessary.

This is not some far-off futuristic scenario; all of that is possible with AI-driven workflows at our fingertips, and it’s something that’s possible right now, with the right investments.

Taking the jump

There are many reasonable concerns that have held back imaging organizations from more widely adopting AI, from wariness about privacy and security, to uncertainly about the ROI, to institutional inertia, and plain old fears about AI phasing out people’s jobs.

And while these concerns are fair to raise, the reality is that AI is not so much a threat to radiologists’ jobs, as radiologists’ abilities and progress are limited by the lack of AI. Radiologists have invaluable skills that AI can’t replicate – human judgment, empathy, and an intimate understanding of patient contexts. That’s the opportunity: to pair skills that only radiologists have with the value that AI can bring to the table.

Radiology has become an overwhelming, frustrating, disorganized job for the average radiologist – and as image volumes go up, these feelings will also grow, causing more attrition and burnout, and leading to even greater burdens for the radiologists who remain. Combined with that 3-5% miss rate attributed to basic human error, and the degree to which so many patients are falling ill or dying from preventable causes that could have been diagnosed earlier were it not for siloed data, and we’ve reached the threshold now where clinicians simply do themselves and their patients a disservice by not implementing AI in more applications.

Applying AI to radiology workflows is the logical and necessary next step – for both AI and radiology itself. It’s a combination that can reap enormous potential for patient care and the radiologist’s day-to-day experience, and it’s not such a sea change upgrade that it would be logistically or financially implausible for the organizations that have already begun to dip their toes into AI. All that’s needed is to take the leap.

Author: Marwan Sati – a distinguished engineer leading AI development in medical imaging and natural language processing for Merge by Merative.

This fascinating topic of AI within radiology, along with others will be discussed at the annual Ai-Med Global Summit, scheduled for May 29-31 2024 in Orlando. Book your place now! 

We believe in changing healthcare one connection at a time. If you are interested in the opinions in this piece, in connecting with the author, or the opportunity to submit an article, let us know. We love to help bring people together! [email protected]