Artificial Intelligence in Teleradiology — An Interview with vRad (Virtual Radiologic), US’s Leading Teleradiology Firm
vRad (Virtual Radiologic) is the US’s leading teleradiology practice, delivering high-quality diagnostic imaging services to more than 2,100 facilities and radiology groups across the United States. By having 23 patents for telemedicine and radiology technologies under its belt, vRad is a leading innovator in the areas of artificial intelligence (AI), machine learning, imaging data analytics, and software to improve the quality of patient care, value for their clients, and the experience of their physicians.
In this interview, health-tech firm, Qure.ai’s Chief Commercial Officer, Chiranjiv Singh (CS) speaks with Imad Nijim (IN), Chief Information Officer at vRad, and Brian Baker (BB), Director of Software Engineering at vRad to learn about their experiences of working in teleradiology and how they leverage technological innovations as part of their care continuum.
CS: Could you please introduce yourself to the audience?
IN: My name is Imad Nijim. I am the CIO of vRad and have been with vRad for about 3 years. I have been in healthcare IT (Information Technology) my whole career since my first job in oncology systems. I am excited about our partnership, and I appreciate this opportunity to speak. Throughout my career I have seen waves of innovations come and go and AI is one of the most exciting things happening for a long time. BB: I’m Brian Baker. Thanks for having us on this podcast. I have been at vRad and in the medical space for about 10 years, and I have been part of the AI team since the beginning, which was about 5 years ago so it has been exciting to see the evolution of the space.
CS: Imad, as an insider how do you see the teleradiology adoption trending? Do you think it has reached its peak or is there still potential for growth?
IN: Teleradiology started out as a ‘nice-to-have’ about 20 years ago and vRad is celebrating its 20th anniversary in radiology this year. Over the last 2 decades, we have led the charge into what is teleradiology, and how to continue innovating in making it a big part of the care continuum.
As a percentage of total diagnostic imaging, teleradiology is growing. Even though our industry started as a convenient type of business, we are a part of the core patient care experience now. We focus on emergency medicine — and we’re really good at it but we have also seen a lot of growth in daytime imaging as well.
COVID-19 put the spotlight on telehealth, and teleradiology brings a lot of value in terms of physician safety and being able to do their jobs from home.
With respect to volumes, it has exceeded the pre-Covid times. There is a consistent growth in volume in the tele-space as well as the hospital space. All of us at vRad as well as non-vRad hospitals, radiologists, healthcare, IT informatics, executives — all of us are challenged by a capacity demand imbalance. There are just way more procedures than radiologists can handle, and that trend continues.
The trajectory of volume increases based on population, demographics, new techniques in imaging — any dimension you measure it at — the trajectory is growth on imaging. In contrast, the number of radiologists and the number of seats available in medical school are only slightly growing. Additionally, more than half the radiologists are above the age of fifty-five, which means there will be a significant number of retirements in the year to come. This demand for increased capacity is a problem that can be solved with technology. Our whole team at vRad is focusing on how we can make radiologists more efficient so we can tackle this imbalance.
CS: What do you see different in terms of the engineering work that both of you lead to enable your radiologists and your clinicians to perform with the increasing load and a parallel pressure on turnaround times?
IN: Turnaround time is a big part of the teleradiology business and a big part of our client satisfaction, and ultimately a big part of patient care. We keep a close eye on it and have built tools to make sure the right study is assigned to the right reporter at the right time so we can achieve that turnaround time.
We have built a stroke protocol and trauma protocol and ways for the clients to interact with us to better understand their status and urgency. We have built our platform from the ground up specifically focused on turnaround time.
BB: There are two important things to understand our approach.
For one, we are practical and also cautious. When we considered AI as a tool for our practice, there were 2 aspects that stood out.AI was something we wanted to start using as soon as possible and the first use case that came to light was prioritizing worklists. We’ve learned over the years that there are many rules and things for the prioritization process and AI is just a factor. Before we use any model, we plug it into the inference engine and figure out metrics to determine its sensitivity and specificity.
Secondly, we investigate the pathologies so we zoom into the things that can impact patients’ lives. So, pathologies like pulmonary embolism, inner cranium hemorrhagic strokes, pneumothorax, and aortic dissections matter a lot on time and thus have the highest impact. We go beyond the sensitivity and specificity and try to find how many patients got prioritized, and how many lives have been saved.
We also try to find out if it is fair to the other patients who did not get prioritized. We analyze all the false positives that are reported, and we shrink them as low as possible to find out the impact on other studies.
CS: What has been the response at vRad to the AI-related initiatives?
BB: We have been fortunate to have enthusiastic responses from our radiologists and other clinical users. We took a step back and coupled that with a healthy amount of skepticism. We focussed on the use case and how engineering is plugged in to solve it. We utilized this way of messaging to a radiologist where the original fears have faded away. Nowadays, clinicians have been asking us where AI is being used, the reality is that AI has been everywhere even though the physicians do not notice it. Part of the reason they haven’t noticed it is that we are very cautious, and we don’t want to frighten radiologists that AI is going to take their jobs or replace them completely. We want to encourage conversations that AI is going to support and help radiologists and not replace them.
We have integrated AI into the quality assurance space and piloting some quality assurance programs and when we catch something that doctors might have missed, they are thankful for it. To be honest the overall experience with clinical users has been very positive even though we were very cautious to start with.
CS: What has been the impact on your end customers? Are you measuring any metrics to see the impact on your end customers as well?
IN: We have had consistent turnaround times for the last year and AI has helped tremendously to achieve that. A lot of our clients do not know a lot about the technology behind the scenes, but they are always very impressed and excited with the AI working behind. We have a report that comes out which is perfectly tuned to the specific customers’ needs such as in style formatting, requirements for certain pathologies. The technology is disruptive as it is adding tremendous value, but it will not be replacing radiologists in the next 5 years. vRad was one of the innovators and we will run into challenges of every aspect from ordering to scheduling, to acquisition post processing work. A lot of things will be impacted by AI as there is a lot of innovations coming our way.
CS: Have you seen cases where adoption of AI tool has really impacted patients care? Cases like lead prioritization, saving a life or anything like that?
IN: We do have a handful of patient stories that have been greatly impacted by the technology. There was a patient in a rural hospital in Iowa who had an acute intracerebral hemorrhage (ICH) which was not a subtle case. What the AI did, in this case, was that while the study was waiting to be read by human eyes, the AI was able to alert the clinicians about the urgency of the case. So that concept of prioritization is what AI was able to do in this case.
The patient had several issues including a midline shift and subdural hematoma and the AI automatically dialed the number of the referring physician, and the patient was rushed to the operating room for surgery and all of this happened in less than 3 minutes. We had the communication, AI, inference engine, and the auto dialer reporting system all working in concert to get the patient to surger approximately 12 minutes sooner than they might have without AI prioritizing the study. A follow-up exam indicated that the stroke did not progress because of the early intervention.
CS: What are some of the stories or insights that took you by surprise while looking at AI solutions (specifically, to Brian Baker)?
BB: I am just trying to figure out how this new tool fits into existing tools and existing processes and there were some surprises along the way. Currently, my biggest set of surprises is the quality assurance space and the ability to use these models to overread radiologists. I think there is a lot of untapped potential that we can focus on. We see a lot of value in the background and prioritization processes. In quality assurance we are just adding to the overall processes that are happening rather than distracting the radiologist. vRad has a robust quality assurance program where we do finals, prelims and probably 30–40% of those prelims are getting overread. We also have a quality assurance portal that any client can go to submit a discrepancy. Depending on what study you read, it’s between 3 and 9 misses per thousand cases. So, to put it into perspective, we already have high quality, but we are always trying to be better. The concepts are being used to overread every single chest computed tomography angiogram (CTA), pulmonary embolism, ICH and that’s phenomenal. It’s been surprising to me how powerful this is going to be. We are already envisioning how can we make the quality assurance process more real-time and more efficient. We see new innovations in AI, not just as a business offering but also to enhance the workflow that can really help the clients. So, to sum up, I would say that the real surprises are the use cases or the micro use cases that come out of the ideas we have in implementation and expand from it to become a fully-featured use of AI and sometimes even enabling non-AI features.
CS: Any last thoughts, predictions on the post-COVID-19 world, or something that you are excited about in the future?
IN: I think that the quality assurance space is something to watch out for as that is what we are really focused on this year. We’ve been putting a lot of resources into enhancing and so hopefully we will get a better handle on the regulatory and FDA implications because that can bring a bunch of new innovations when it comes to patient safety.
BB: My insight from the past 18 months is that it’s a marathon and not a sprint. As we look to use AI in various use cases and building the engineering tools, it’s going to take a lot of time and these use cases are going to evolve and we’re going to learn AI as we go along. In the next five years, we’re going to continue learning as we have done till now.
CS: I completely agree that it is a marathon and not a sprint — and that’s exactly how we see it at Qure. It’s been a pleasure talking to both of you. Thank you for your time.