How Envision Healthcare Is Incorporating AI into Radiology Workflow

National medical group Envision Healthcare has one of the largest radiology practices in the country, providing services in multiple states. Among his responsibilities as senior vice president, interventional and diagnostic radiologist Roi Bittane, M.D., focuses on strategy and the integration of new technologies. He recently spoke with Healthcare Innovation about how Envision is incorporating AI tools into practice, including the co-development of a new emergency department communications tool. 

Besides radiology, Nashville-based Envision also provides services in emergency and hospitalist medicine, anesthesiology and neonatology. It reported 19 million patient encounters in 2023.

Healthcare Innovation: When promising innovations come onto the market, how does Envision work with partner hospitals to decide which things to deploy and how to integrate them into workflows and make sure they work with whatever PACS system or EHR that the hospital system is using?

Bittane: It really depends on the health system. Some health systems do their own analyses and make decisions that are in the best interest of their wider health system, but in most cases it falls somewhere in the middle. We’re part of a team that takes into consideration the hospital system’s clinical needs, and the radiology service needs as it relates to workflow. It all funnels down to how we provide better patient care, faster patient care. We serve, to some degree, as consultants or members of a team, and that’s the more common situation. We have very large health systems that see us as their go-to people when they evaluate new technologies and new workflows. 

HCI: But as far as making the investment, though, it’s the health system itself doing it? Envision isn’t buying the solutions itself? 

Bittane: Well, that’s one pathway. The other pathway is that Envision does maintain its own PACS infrastructure that we use with many of our clients. Within that infrastructure, we have both classic algorithm-based AI and also some communications-based tools that I would say are unique to Envision and also part of an AI picture. In that respect, we decide what to deploy.

HCI: What are some of the most significant impacts you’re seeing from integrating AI tools? Is it faster turnaround time or something else?

Bittane: I would say the most significant influence is on quality of care and accuracy. I do think that AI algorithms do increase the ability to make findings faster. But that’s very difficult to quantify, to be completely honest. What is easier to quantify, or to evaluate, is the accuracy. And I do believe that for the specific algorithms that you implement, you avoid various mistakes and various fallouts if you’re using AI. To me, that translates to higher quality care. 

HCI: Can it help the radiologist prioritize in an emergency setting which studies need to be read first?

Bittane: Absolutely. We do that within our system. When the algorithm identifies a positive finding, it does indeed raise a flag. It pushes cases to the top of the list. It facilitates us reading them faster and providing a result to the ED faster, which translates to better quality.

HCI: Why are we seeing more progress in algorithm development in radiology than in other fields of medicine? 

Bittane: I think it is due to the fact that the radiology infrastructure is digital. At the end of the day, we’re looking at digital images. They can be sent to different locations easily, sent for analysis easily, etc. That’s the reason you’re seeing a lot of activity there. Where things will become more interesting is in workflow orchestration, which can be relevant to other fields, not only radiology. You see patients in what order? What’s the workflow? All of these things are also based on how patients actually flow through a certain service. So that’s coming down the pike as well. 

HCI: Envision has been working with the company called Aidoc. Is there something about their services or solutions that stand out?

Bittane: I would say a couple of things stand out. One is that the founders have a wealth of experience in AI at the ground level. These are not theoreticians. These are folks who actually have years of experience. They know exactly what they’re doing. They’ve worked in various national level organizations that study AI and implement it in multiple different use cases. 

The company has several hundred folks on the back end, who are all PhDs in physics and engineering, etc., and have real-world experience. 

The other piece is they actually have quite a few algorithms that are FDA-approved. We’re working with them on various unique tools, like our ED communication app. So it’s been a very fruitful relationship, mostly because of the will to work together. 

HCI: Could you describe the work on that ED communications tool? 

Bittane: There’s a very interesting thing happening in radiology now. Many of the radiologists have a preference for working outside of the hospital. We’re seeing a major shift toward teleradiology. That is creating a gap between the folks in the hospital and the radiologists who are sitting in their home or in an office. Actually, it creates quite a few challenges, especially for folks who are used to the classic workflow. If a radiologist is not located on site, it becomes more difficult to discuss a case, so there’s a communication gap. 

With Aidoc we have developed something that’s based on the Amazon service model in order to bridge that communication gap. In the Amazon service model, you go to a website and order your products, You can track where the product is in the fulfillment cycle, and when it’s ready, it’s delivered directly to your door. If you have a problem, at the push of a button, you can reach out to Amazon and receive some type of resolution to the problem. 

We try to emulate that to some degree with radiology. The app that we developed allows folks, once they order a study through whichever on-site system they choose, to track the study and see where it is in the treatment cycle. Has it been read? Has the report been sent back? That saves the ED physicians a major headache and a lot of time tracking what’s happening with their patient’s case. 

The second piece is once the study has been read and the report is ready, they immediately get a message with the report, as well as the images sent directly to their phone. So there’s really no delay. Today, in some cases, you see physicians going into their EHR and clicking refresh, looking for the case. This alleviates all of that. The moment your study is read, it’s sent directly to you. That’s emulating that Amazon model.

Another component involves resolving issues. For instance, the report might say we found the stone on the right side on the CT, but the ED doc says the patient has left-side pain and it is not fitting the clinical picture. At the push of a button, they can communicate with us and reach the radiologist and really resolve that issue more or less immediately, Again, closing that communications gap is becoming very significant due to the movement of radiologists off site.

HCI: With the AI tools, is there the potential for an increase in the number of incidental findings discovered? And do health systems generally have good ways to track and follow up on those with patients? 

Bittane: There is a lot of information traveling back and forth between not only radiologists and ED physicians or surgeons, but between all physicians. Keeping that information within a framework, and making sure it’s tracked appropriately and the patient gets the information is definitely a challenge. 

What I’m seeing is that the larger systems are implementing specific programs to track specific types of lesions and ensure that at the end of a treatment encounter, whether it’s in the ED or in the hospital, we actually go back to the patient and make sure that they are aware that there’s a finding that they need to follow up on. I think health systems have a lot of room to improve in that respect. There are many companies — and Aidoc is one of them — developing various types of tracking tools that are now integrated more and more, either directly into PAC systems or into EHRs. So it’s being addressed. Can I tell you that I think it’s in a perfect state? Absolutely not. We have a lot more room to improve.

HCI: Anything else you would like to mention? 

Bittane: I think we’re on the verge of significant changes in radiology. Besides the shift of diagnostic radiology from the on-site workspace to a remote workspace, the second big shift relates to the integration of AI, and how much of AI is actually going to be integrated and what the influence will be. Right now we’re seeing the initial steps, because we have a limited number of algorithms. But as this develops, we’re going to start to see hundreds or  algorithms, there is a question of how that’s going to influence radiology and the health systems and patients.