Imad Nijim, chief information officer of MEDNAX Radiology Solutions and Virtual Radiologic (vRad), has been in medical imaging and informatics for more than 18 years. He’s seen a lot during that time, but nothing quite as groundbreaking or exciting as the artificial intelligence (AI) currently being developed by researchers all over the world. Nijim spoke with Radiology Business about MEDNAX Radiology Solutions’ plans for AI, what he sees in the industry today, and the company’s big plans for RSNA 2018.
You were appointed Chief Information Officer of MEDNAX Radiology Solutions in April 2018. What was your initial impression of the company and its AI capabilities?
Imad Nijim: One of our most tremendous assets is the rich and geographically diverse dataset. MEDNAX Radiology Solutions is connected to over 50,000 modalities, serving 2,100 facilities with more than 800 radiologists—the majority of whom use a common imaging platform. vRad already had a robust AI program when I came on. I was impressed by the disciplined approach to building the necessary natural language processing (NLP) tools for patient care, the statistical modeling needed to evaluate the impact of using AI for triage purposes, and the technology know-how to pull it all together. I had no doubts the vRad team was well down the right path and am excited to expand and accelerate our AI program under MEDNAX Radiology Solutions.
There are a lot of conflicting opinions out there about how AI should be used in radiology. What is MEDNAX Radiology Solutions’ philosophy toward AI?
We’re very practical. Our philosophy is that we want to develop models that have an immediate and measurable impact, and it seems like the rest of the industry is starting to reach that same conclusion. At RSNA 2018, I expect to hear a lot of people say, “OK, the hype about AI is behind us. Now what?” That was our thinking from day one. Instead of chasing AI because of a trend, or because people think they can monetize data, we invested in solutions that we build and deploy practically for our practice. For example, we currently have models that can detect pulmonary embolism in a chest CT. If this happens, and the study was not ordered as an emergent study, our reading platform elevates the priority of the study, reducing what could’ve been a 25-minute turnaround time to a 10-minute turnaround time. Another practical example is our chest x-ray model. Of more than one million chest x-ray studies we read each year, 50% are missing or have an incorrect AP/PA vs. Lateral View Type Classification. Unchecked, this would have negative impact on hanging protocols and billing. Our AI algorithm addresses these issues invisibly behind the scenes with 99.8% accuracy.
What are some of the things you are most excited about when it comes to developing AI at MEDNAX Radiology Solutions?
The partnership between our medical leadership and technical leadership is incredibly strong, and that’s a key recipe for success. I’ve seen a lot of smart start-ups come around that know the data, but they lack the medical leadership. I’ve also seen a lot of smart medical minds that lack the technical expertise. We’re excited to have these two groups working so closely together daily to build the right models for the radiologists.
Our growing ecosystem of partnerships is also very exciting. We have developed our own AI models internally, of course, but we have also developed a substantial ecosystem of partners working with us to achieve shared goals. This ecosystem really built momentum in 2018. We will be expanding our partnerships in 2019, working on AI model development, third party model validation, and data curation.
Data curation? Can you elaborate?
Data curation includes analyzing reports at scale using vRad’s natural language processing capabilities. Data curation produces automated labels, radiologist image annotations, automated anonymization of data, and securely-extracted clinical content. We are in a unique position to drive innovation and I am excited about taking everything we’re learning to that next step.
What does MEDNAX Radiology Solutions have in development for the next 2-3 years when it comes to AI? What’s the future look like?
With our data collection and annotation well-understood and our triage models in development, the next step is to use AI to improve our already stellar radiologist quality program. We are working closely with our medical quality leadership and our partners to take this very important next step together.
On the technology side, most of our AI development is focused on a single model. But that will begin to change in the next few years. You see headlines now about AI models that can detect lesions or assess fractures. Those are all great, but our plan is to combine numerous AI models, in conjunction with our NLP and our data analytics, to construct a pipeline approach to AI and workflow that will improve the overall quality of patient care. We want to push our entire industry forward in big steps.
RSNA 2018 is coming up later this month in Chicago. What does MEDNAX Radiology Solutions have planned for the big show?
In addition to daily reading platform demos and showing radiologists the many personally and financially rewarding reasons to consider joining MEDNAX Radiology Solutions and vRad, we’re excited to show imaging leaders that we’re one of the most innovative radiology service providers out there when it comes to AI.
At RSNA 2018 we will introduce the MEDNAX Radiology AI (MDRAI) Incubator, a collaborative program focusing on data sharing for model development, data curation and image annotation, and model validation. Our vision is to bring together our radiologists, our rich and diverse clinical dataset, and our partners into an ecosystem to build future tools for radiology. We’re at the front of the pack. We can really show attendees how AI will work with radiologists to optimize patient care. It should be a lot of fun.
Editor’s Note: vRad will be exhibiting at RSNA 2018 at their main booth #7938 in the North Hall and a recruiting booth #4273 in the South Hall. A networking reception is also planned for Monday, Nov. 26, at 6:30 p.m. Complete RSNA event information is available at www.vrad.com/rsna