Innovations driven by artificial intelligence and advances in precision medicine are rapidly changing many facets of the healthcare industry, including radiology and imaging. Few people understand these developments better than Homer Pien, PhD.
Dr. Pien is the chief scientific officer of diagnosis and treatment for Philips. He took the time to speak with Becker's about the biggest trends that will impact imaging and radiology departments in 2019.
Editor's Note: Responses have been lightly edited for length and clarity
Question: With the shift from volume- to value-based care, what changes can we expect in the radiology department?
Dr. Homer Pien: We can expect to see a couple different things. It's probably worth noting that the shift from volume to value is actually not a universal trend around the world, but we are seeing fairly slow and steady progress towards value-based medicine within the U.S. So, when we look at specialties like diagnostic radiology, which doesn't directly dictate patient treatment, quantifying value is far more challenging. So, we anticipate radiologists will need to more robustly justify their contribution to patient care.
Now, one way that's happening is that we're seeing radiology being pulled closer and closer to various service lines such as oncology and cardiology. And we're seeing these multi-disciplinary relationships become much more important in radiology departments. I believe direct contribution to patient care is a trend we're going to see more and more as value-based healthcare becomes more important.
Q: Artificial intelligence holds much promise. What are some of the early wins you see for next year?
HP: Artificial intelligence (AI) represents a dramatic opportunity to improve healthcare in general, radiology specifically. Many point solutions utilizing AI exist today but we must be judicious about applying it where it is most useful and recognize where there are unrealistic expectations.
At Philips, we broadly categorize artificial intelligence into operational workflow AI and clinical AI and there may be some overlap between these two. For example, you could develop a clinical triage algorithm that would also include workflow efficiency.
Clinical AI generally requires a class two approval from the Food and Drug Administration, and monetizing these algorithms typically requires insurance reimbursement. So, both the regulatory clearance and insurance reimbursement will require a significant amount of clinical evidence. For this reason, the adoption of the clinical applications has been a little bit slower than those AI solutions that focus on operational and workflow efficiency.
This is why I see operational AI as the "low hanging fruit" because those algorithms which enable much more efficient operations and workflow do not require the class two approval process and can directly and immediately contribute to the reduction of operating expenses of hospitals and radiology departments. For those reasons, I believe we'll see artificial intelligence applied in operational areas more readily in 2019. The next leap forward will be to integrate AI offerings into seamless and complete disease-centric solutions.
Q: What about the clinical domain? Where will AI have its biggest impact there?
HP: Broadly speaking, AI solves two types of challenging clinical problems. One type is those problems that require huge amounts of data. The prototypical examples for that will be lung CT, which involves 250, 300, 325 slices of data the radiologist needs to scroll through. That's a huge amount of data, so using AI to help identify areas of focus, I believe will be a very promising area for AI. Similarly, CT colonography, also involves hundreds of slices of data because we're going through such a large portion of the anatomy. So, that's a domain in which AI could really make some substantial contributions. So that's one area AI can have a big impact in the clinical domain.
The second area would be instances in which very subtle features or nuances can have a dramatic impact on outcomes. These are things that some specialty trained radiologists may be good at identifying but general radiologists may not be. A perfect example of this is identifying intracranial hemorrhage. When you have a bleed in the brain, especially if it's what we consider to be an arterial bleed which is a high-pressure bleed, even a small bleed can be fatal for the patient if it's misdiagnosed. A department that may not have the subspecialty radiologists and is primarily staffed by general radiologists, they would be much more challenged to find those lesions because they are so difficult to see. That's another example of a different kind of case in which AI could actually make a significant clinical contribution.
Q: What role will virtual radiology play in imaging services?
HP: Hospitals are under huge amounts of financial pressure, which is going to drive increased adoption of alternative delivery models for radiology including the virtualization of radiology. The virtualization of radiology services will provide scalability to accommodate increasing patient volumes, and the agility to flex capacity according to demand during nights, holidays or other unpredictable events.
Virtual radiology also helps us address the issue of radiology shortages. At times, we'll buy subspecialty reading services as needed because not every single hospital or every single imaging center is going to staff a neuroradiologist on a 24/7 basis. So, if you have an emergency room and you do have a CT scanner and a neuroradiology patient comes through, what do you do? Virtualization for imaging services really becomes critical in those cases.
Additionally, depending on the virtualization model, such service may also be able to decrease the variability and disparity we see in these large hub and spoke health systems. Being able to distribute images and tap into the expertise you may have at the hub through virtualization will help ensure optimal operational performance, reduce staff workload and ensure equal quality of care across the health system network.
Q: What progress will we see in 2019 towards precision medicine?
HP: The north star for how radiology will evolve is the enablement of precision medicine, a single patient view that will improve outcomes at lower cost. At its very core, precision medicine is really about the integration of large quantities of diagnostic data. Essentially, what you're trying to do with precision medicine is render an earlier and more definitive diagnosis of a patient for better outcomes.
Precision diagnostics is the first step to precision medicine: the right diagnosis at the right time leads to the right therapy. To achieve precise therapies guided by imaging, I believe AI could help select the right imaging study at the right time, and ideally it'll lead to the right therapeutic intervention as quickly as possible.
In 2019, we do expect to see continuing steady progress toward the integration of radiology. In particular, what we're talking about is really the integration of imaging data with histopathology and molecular pathology and electronic health records. There are two parts here. One is really on the integration of the reports, which is predominately a natural language processing problem and the other part is much more about scientific research. This integration is occurring more so at the report level right now but we'll continue to evolve the science at the raw data level and the image level. This is what we refer to as "radio-genomics" when we integrate radiology and pathology EHR data at the raw data level. That's obviously much more challenging than doing it at the report level, but it also holds a great amount of promise for precision medicine.
Want More? Click here to learn how design services can be leveraged to transform healthcare delivery.