In this podcast Mitchell Goldburgh, Global Solutions Leader at NTT DATA Services, and Dan Housman, CTO at Graticule discuss NTT’s Advocate AI service and how their solution can efficiently and effectively address the challenges of aggregating data from multiple healthcare sites into a data cooperative — accelerating insights utilizing previously untapped resources (for example, medical images, and associated reports).
Advocate AI’s rich data source and associated analytic tools enables Graticule to achieve the scale required to provide rapid feasibility analyses and help the industry address and answer complex healthcare questions today while opening up exciting possibilities for future innovation.
Hi everyone, this is Dan Housman. I’m here with the Novel Cohorts Podcast on behalf of the Graticule team. Today we have Mitchell Goldburgh from NTT DATA Services and we are interested in sharing a dialogue over some podcasts about ways we can create insights off of healthcare data using radiology and related information. We will probably do a couple of these to dive deep so we can keep your attention. Let’s start with introductions. Mitchell, please introduce yourself and explain what NTT DATA Services is and what you are doing in the Real-World Data business?
Thanks, Dan. NTT DATA Services is a digital business and IT service leader, part of a $109 billion corporation called NTT. We’re based out of Plano, Texas and help organizations accelerate and sustain value through learning about digital journeys. In this particular case, my team has a service called Advocate AI, which is a part of the healthcare provider organization and we teamed with our Life Sciences team and with Graticule to address the challenges of bringing data from multiple locations into a data cooperative so that people can process and digest that data to gain insights into medical imaging and associated reports in EMR data. And we’re pretty excited to be working with Graticule.
And this is Dan, to give folks a background on Graticule, we’re founded to help organizations run the types of studies they’re interested in with what we call advanced real-world data. That means we’re focused on the types of studies that require more than just claims information or basic structured EMR. We both do work that looks like a contract research organization where we’re doing studies in conjunction with health systems, as well as doing work to try and build out studies with complex datasets, like what’s available through NTT DATA and Advocate AI, sometimes even linking data together, and we’re excited to be working with the team at NTT DATA because there’s a really rich data source and some great tools on top of it that I think clients will really be interested in. So Mitchell, maybe you can tell me a bit about how this data gets to you? How big is this data set you’re working with? What makes your offering so powerful? Because I know we’re excited, but maybe you can tell us what got you excited and what you guys are working on?
Sure. That goes back to a legacy company that was acquired and brought into the NTT family. We started working with medical images and associated reports in hospitals, medical centers, oncology centers, way back in 2000. So, since that service inception we’ve processed over 450 million imaging studies across thousands of clinical sites. And it’s always been a hybrid cloud offering that allowed us to not only facilitate the management of the imaging, but also the exchange of the imaging data. In 2015, we began a journey to help our clients go beyond just enterprise imaging as a service, and launched the specific data cooperative service called Advocate AI, which allows us to bring not only NTT DATA clients but people that are interested in accelerating the adoption of Intelligent Automation and AI for the purposes of solving clinical real world problems and forming in this cooperative, bringing pharma, AI companies, data science, and real clinical data/evidence to make it actionable.
So, it’s a network that continues to grow. When we started with you, late last year, we had grown to have over 8 million studies catalogued and available for data mining from multiple locations, all in the US, but, you know, our vision is to go beyond the US as Graticule has done, and support at scale the automation of data curation.
Because the secret to advancing artificial intelligence is having access to the data, being able to bring together data at scale, and ensure that it’s de-identified but correlatable to not only the diagnostic imaging reports and the imaging services but other types of lab data, genomic data, pathology data, and bringing that all together to do the studies that Graticule is being asked to do. So we’re really excited.
Maybe I can talk a little bit about what gets us excited about working with your team. First of all, it’s hard to find big data, people have been talking about big data for a long time about, a place where you can do research with enough scale, where there’s enough content to be able to look at a specific problem.
And the fact that you have so many clients and you’ve done the work to negotiate agreements back to the providers that are willing to participate in this machine learning development, decision support development, any of the tools and algorithms we can put together so that we can find enough of whatever kind of imaging study, we need to build these analytics, it’s very helpful.
And so we are excited, as we’ve been working with your team, that we can find the kind of scale that’s necessary to answer hard questions. But that’s really just the beginning because I think for us that’s being able to answer static data questions, but there’s also capabilities and tools built within your infrastructure to make it a lot easier to do the kind of work we’re trying to do, and that’s within the tools of Advocate AI to be able to do that curation work so that we do not have to build lots of tools and focus on the studies we want to run, and be able to have, you know, a curation model of imaging studies and annotation tools, all integrated with the data itself, which isn’t something that is easy to find out there. I’d also say on top of that, you know, the grand vision of all this stuff, you know hey we do AI, think if you were to dig and ask people, has there been a big ROI to AI? It hasn’t emerged completely yet.
And, in order to get a benefit from being able to make different decisions to improve the workflow of clinical teams, we need a place where there’s scale integrated into the health system as well.
And so we’re really looking forward to some of the future phases, what we can do together to be able to take algorithms we can build and train and validate inside of retrospective data, and connect up with some of the health systems that Mitchell’s team at NTT DATA and Advocate AI have set up to, to be able to have an intervention which will impact patients.
And I think it’s going to be possible, because, we are connected to a group that’s really deeply integrated into the health systems, as well as some of the life sciences companies that are out there looking to do these kinds of things.
That’s a really good point, Dan and not, not to take you off your top track but, 70% to 80% of the time and, thus, cost is associated with curating data and preparing it for all the science that Graticule provides and our joint customers are providing. What we’re trying to do at scale is use intelligent automation to take all this dirty data and make it harmonious, so that it can be integrated and analysed and one of the interesting projects, and we’ll talk a little bit more about that, is understanding what we discovered that sometimes there’s data missing from the reports that’s extractable from the images themselves, which creates a stronger data set for some of your clients to explore their ideas.
Yeah and, I’d say, looking into the projects we’ve been working on together because we have done a number of these studies so far.
I’ll give an example, it’s often important to obtain measurements of different organ systems to see if a disease is progressing, and especially if it’s progressing rapidly which means that it’s more important to do an intervention, whether it’s, total kidney volume, it’s tumor growth, changes in cardiac tissue. And what we found is by working with a scale set of images we can find a whole lot of data that can be automatically pulled out of radiology reports, which have already put in those measurements of the volumetric dimensions of organs over time, which gives a strong indication of progression. But if we dive all the way into the images themselves often those measurements have been embedded in the image and there’s more measurements and more information directly in those images that aren’t stored in the text reports.
So, depending upon the level of value that can be gleaned from building out the Natural Language Processing for OCR, maybe even full AI to make the measurements directly on the image, we can really get to, at scale, understanding, in this case, progression, which is a pretty critical piece for understanding, who really needs to be treated or not treated with a drug. If the folks you’re looking for are the ones that are rapidly progressing or maybe looking at whether or not they’re benefiting from a treatment, to see if we’re stopping progression or reversing it.
So, it’s great to have all the capabilities and the tools, and I think we’re going to see that there’s a lot of value, just in measuring things in radiology that’s never been done before at scale.
That’s a great point and it’s oncology, it’s radiology, our vision is to go into dermatology and other areas but regarding the two points that you made, I just want to expand upon quickly.
One is the ROI of artificial intelligence doesn’t come from simply the detection and measurement, as you pointed out, it’s, taking action on those measurements to intervene to either early detect and therefore reduce the level of intervention or ensure through precision medicine that the intervention that you’re providing is effective, so that’s really where the ROI from AI comes from.
And I think you also mentioned, being that we have all sorts of imaging in our portfolio and it’s growing, right. If you look at the needs for de-identification at scale, that’s one of the tools that comes with the Advocate AI but also it’s identifying the measurements that may be in a screen capture for an ultrasound that we can extract that weren’t in the report, we’ve learned a lot in working through the projects that you’ve been bringing to the team, and we’re really excited to then help customers, especially with early adoption, marry opportunities where they participate in Advocate AI to link them up with perhaps some of your customers who are interested in piloting their ideas.
So, it’s really a community that Advocate AI fosters through the data cooperative not just a data repository.
And I’d say, on our side, now we’ve done some of these projects, we’re learning how to build tools to make the process much faster as well as defined tools that can, in a less directed way, instead of picking a certain organ to get the measurements for example, looking across all measurements in the whole system to see where there might be critical mass that could support known drugs, known diseases, that could be improving patient care, if we just developed out some of those measurements.
So, I think working together, we can rapidly evaluate the feasibility of using radiomics and radiology and get everyone to look at it as a tool. And there are so many things we can do going forwards to build off of what’s already been done so far, by doing things like linking data, working more closely with the health systems to do the translation and validation piece prospectively. We’re really at such an early stage and haven’t seen the full value, but I think we will in the next few years.
And we’re looking forward to working with you to uncover those opportunities and even through the sites that are participating in the questions they’re asking about what’s inside their data, it’s clear that not only can pharma and AI companies benefit from the data but the providers are benefiting as well.
In the end I think the big give back is to the providers and the patients, which is, if we do lots of research on the data, we’re doing it because there’s something out there that could give patients a better, better experience, better treatment, better options. And as we make these tools, either they can be made completely free, because we just publish papers and say just use this variable this way, or they can become commercial diagnostics or commercial tools that can be used, in the context of validated treatment patterns.
So, we’re excited to get to that we’re still pretty early in the journey on that, but I think, we’re in this business, at least at Graticule, and I think it’s also true from the Advocate AI team to have an impact on patient care. So, it’s pretty close to our mission. If we can get all the work we’re doing to the point of care.
Thanks Dan. Thanks for pulling this together and letting us introduce ourselves as a team, and I look forward to the next podcast to explore more, perhaps with some of our other partners.
Yeah, maybe we can dive into oncology or some specific field that has great interest these days and real-world data and, you know, that that’ll be a fun conversation if we can talk about some other topics that are a little deeper. But thanks so much and looking forward to speaking again soon.