"The healthcare delivery model is changing from hospital being within four walls to getting distributed and virtualized”

Q. Tell us a little about your background and how you got to digital healthcare.

Amit: I grew up in India and did my Master’s in Electronics and Communication from the Indian Institute of Science (IISc). Then, I started working in the fields of embedded systems, communication and networking before moving to GE Healthcare. When I joined GE Healthcare in 2016, I knew nothing about this space. But I was convinced because the whole digital transformation that they wanted to drive and the business was truly exciting. I discussed this with the entire leadership team and decided to take the plunge. Initially I was stumped to get a call for healthcare and then, I learned that the position would be in Milwaukee. Now, I’d been to the US many times before, yet I had no idea where Milwaukee was. When I opened Google Maps, I saw this tiny dot near Chicago and I concluded it was a suburb but, Milwaukee isn’t as near to Chicago.

So, I made two major transitions — one was from networking with Cisco into healthcare – and the second, from Bangalore to Milwaukee. The latter was more challenging but that’s how I ended up with my first role in GE as the CTO for the imaging business.

My focus has always been on digital transformation. And I’ll talk a little later about the Edison Platform, which we conceptualized here. Our CEO, Kieren Murphy at the time, told me that what GE is doing now is really relevant across healthcare. That’s when I picked up the Chief Digital Officer role for the company.

Q. There are CDOs being brought in from outside the industry in healthcare and it’s a good move since it brings a whole different perspective. What was your first impression when you started here? How has that changed over the past 5 years in GE, working with data aggregation, analytics, and advanced analytics?

Amit: Initially, I was taken aback by a couple of things. When you look at technologies like CT or MR or even anesthesia machines, bedside monitors, ICU equipment, there is deep and highly complex tech involved. It’s also a place where there is a fair bit of science – physics, electronics, and electromagnetics — that converges.

With GE Healthcare, I noticed that people who’ve been with GE for a long time really understand the space and the depths very well and that’s very important given how critical the work we do is, from a patient’s perspective.

When I joined GE Healthcare, everything seemed a little slow and there was too much process — a lot of attention towards testing, safety and compliance, privacy etc. When you come from a different world, it’s burdensome. So, it wasn’t a great experience to begin with ­­– the whole regulatory framework – but I started appreciating it over time. I realized why some of these are absolutely crucial. Over the 4-5 years that I’ve been here, I’d say, these are the strong pillars on which the industry stands.

Q. As a CDO, how are you defining digital at GE healthcare? What does that role mean to you?

Amit: If you look at healthcare products, the equipment generally is a combination of hardware and software. When some of this equipment was first made 40-70 years ago, it was always a combination, so software is not new to the healthcare industry.

There’s been a very strong realization that the data this equipment produces or other patient-centric data available within the health system, is a very important asset that can be used to significantly better the healthcare delivery models and outcomes for the patients. Everything revolves around what you do with data. That’s how I look at it from a digital perspective.

So how can we use this data and information produced by the equipment or what we capture for a given patient or cohorts of patients who have undergone similar journeys so that we can gain insights into what’s really going on? How can we learn about the disease status for a patient, a population of patients? How were they diagnosed, what treatment had they undergone and what was the outcome of that?

To learn all that, we need artificial intelligence. But our ability to take this data and information, convert that into insights, use applied techniques like AI to really learn from everything that has happened to better diagnose patients or create better treatments or more targeted treatments – that’s what precision health is all about. That’s how I look at it from a digital perspective.

Q. It’s a very comprehensive vision. Can you tell us a bit about Edison? How is it different from the other GE platforms? Which one’s more exciting?

Amit: I’ll probably answer the second question first. In that context, I’ll also talk about Edison. When you look at the platform, and this is a question that I’ve often got, there are two aspects to it.

One, the basic compute, memory and the rest of the OS, infrastructure etc. All platforms have that but that doesn’t define a platform. Its only what the platform is built on. When you look at the platform, its persona of the platform is defined by the domain it deals with. In this particular case, healthcare predicts that with industrial automation, when you look at that persona, it tends to be very different from industry-to-industry. So, when you look at healthcare, there are standards for data, for storing images, for communicating or connecting to different systems. There’s DICOM and HL7 and now FHIR, the nature of data is different. An image is a pretty big set of data. The way you process that information is different and so are the latency and timing requirements. When you look at all that, you also observe the real persona of this platform. And that’s what differentiates one platform from the other.

With regard to the healthcare platform, there’s a very distinct set of characteristics compared to industrial automation platforms. There may be intersections, for example, even on a healthcare platform you may actually look at the device, represent it and be able to manage it so that goes into the IOT and the industrial automation aspect of the platform. But specifically, when you look at what you’re trying to drive from the platform perspective, I’ll say that the Edison platform is targeted towards clinical workflows and applications.

That persona is very important, and that’s the basic difference between Edison and Predix. When we started that journey, the objectives were simple — GE Healthcare has about four million imaging devices in the field so roughly about and 350,000 or maybe 400,000 imaging devices and the rest are lifecare solutions, ultrasound machines. There are many devices and all produce a tremendous amount of information — we do two billion scans on our devices every year. So, you can imagine the amount of data that we produce.

One of the first things we had to do was, connect all these devices so that all the information could be ingested into a common place. Then, combine that information with the rest of the information available within the health system. I tend to look at it as a vertical axis and a horizontal axis for data. So, there’s the deep data from the device, which is vertical, and it’s combined with for example, horizontal information from within the health system or outside it — now more and more wearable devices, social media information, population health etc. Or there are EHRs and EMRs with patient information, also past imaging information in PACS and that’s horizontal. The Edison platform plays at the intersection of that vertical data and the horizontal information. So, this can all be combined.

One of our first objectives was to start aggregating information so that we could get a comprehensive 360-degree view of everything that’s happening (with a patient) along with all data that is available for them. So, we look at it as a longitudinal patient record. The platform provides supporting tools and services to process that information and create a workflow so that the insights gained either by analyzing the data or running algorithms on it, can be a part of the clinical workflow. It’s a very important aspect that we deal with, as far as healthcare is concerned, because at the end of the day, the solution generated has to fit in the clinician’s workflow, seamlessly. There are things that must be done on the platform and tools that have to be created from that workflow integration perspective. In a nutshell then, the platform is connected to devices and has the ability to ingest information from them, combine that with the horizontal information about the patient, use the processing tools, workflow tools and get insights, which may be converted into clinical workflows and through them to an outcome. That’s what the platform is and that’s how we break it down.

Q. Given GE has a certain heritage in the market with its medical technologies and diagnostic devices, there’s a lot of image data. How have you taken this and converted the aggregated data into advanced analytical insight? What are the challenges here?

Amit: One of the things we realized is when you look at the richness of the data, there is a tremendous amount of information already available, today. But 95% of that information is not used, and that’s a huge opportunity in healthcare because, as you make more use of the information that’s already there, you can get deeper insights and optimize the workflows.

Or you can be very specific and targeted about identification of the disease states. It can also be very targeted about therapies that would be effective for a specific biomarker or a specific disease state. What you start realizing then, is that, you can actually apply some of these insights at various levels across the enterprise, starting with the device.

A couple of examples just to illustrate the point here will be — On the CT and MRI — these are devices which are extremely sophisticated cameras — they’re really taking the picture of the patient and clearly offering insights into the patient, in general. The way the technology works is, when you scan a patient, you get a very weak signal from them. The job is to take that signal and convert it into an image so that the radiologist or physician can then check the image and offer a diagnosis.

In general, to create a better image, you need better hardware, physics, magnetics, more complexity in software etc. In the past, producing a better image would need next-generation hardware, software, all targeted towards processing that signal so that we get a better image because, the better the image, the more insights the clinician can get and the better the diagnosis, However, the image has a lot of noise. What we’ve done in the CT and MR is embed the algorithm to teach the clinicians how to differentiate between the signal and noise at the very raw level of the signal.

Q. What does that mean?

Amit: When you scan a patient, you might actually project “X” ray and then, on the other side, there’s a detector that transforms that projection of the “X” ray. Those transformations are very complex or in an MRI, you’re basically projecting a magnetic field on the patient and the body reacts – that’s to say, the hydrogen atoms react and you get a reflected signal back. It’s a very weak signal, and this is at a crude level.

So, a very weak signal means that there is a lot of noise around it. To convert this into an image requires a lot of transformations and substantially complex image-processing software. With AI, you teach the algorithm to “see” this pattern, which has this signal and this noise and the way to filter the noise. Since the signal is much better, the image will be enhanced quality-wise, too. So now, AI can be used to really create a much better signal before the software undertakes the processing.

We have been able to embed AI into the deepest levels of the product. So, for example, in the MRI machines, when a brain scan/neuro scan is to be done, one of the tricky elements entails the technician having to set the scan plane correctly. It’s a complex procedure that can take a while because it must get a 360-degree view. If it’s not done correctly, the image can be impacted. Using AI algorithms to do automatic scan selection correctly saves time, reduces variability from one technician to another, and lowers errors significantly while improving the quality of the image. So, it results in a significant amount of productivity as well as accuracy. That’s another example.

But then, we’re going all the way. So, we look at workflows and decide how to position the patient on the table. We check for use of the camera feed for analyses and then, do an automatic patient positioning. It’s a very good example of a workflow and we’ve done that, too. Then, the image is taken to the radiology department and interpreted there. When you interpret an image, you have to segment it, quantify or even take the measurements. And subsequently, you get into actual diagnostics.

If you see radiologists and how they work, they spend quite a bit of time segmenting the image, taking the measurements and it’s mostly repetitive, time-consuming and error prone. But you can use the algorithms to actually take a lot of the mundaneness out so this. It improves accuracy, focus, and enables faster diagnosis. So, we have done that. With AI, you can segment the image in 3-D, quantify it, take measurements and for some of the anatomies – the segmentation measurements can take hours depending on the complexity of the anatomies — you can reduce that time to literally 30 seconds. The last step is obviously clinical diagnostic processes and we’ve gone there to help with diagnosis using AI and that’s been the assist tool to the radiologists.

Q. Imaging, the world of radiology is a very high value, high impact-opportunity areas. GE Healthcare has assigned several multi-year contracts with health systems to help them with this. How do these relationships work? Since these are at-risk contracts with some accountability attached, how do you use data to build the infrastructure internally within GE healthcare? How do you demonstrate to your customers that your software delivers?

Amit: When it comes to data and the AI world, there’s no one company that is going to actually do it on all alone. There’s a vast ecosystem out there, a much bigger ecosystem outside of GE Healthcare than what we can do, inside. So, one of the first things we did on the Edison platform is that we opened it up. We publish the APIs, we encourage and run a number of accelerators. We work with start-up companies in India, China, and Europe. We’re working with significant number of companies here in the US too, to actually get their algorithms and their applications integrated on the platform so that we can take them to the providers and integrate them.

In many cases, we do a significant amount of integration work because, the value of the algorithm is enhanced significantly if it’s fitted well within the workflow. If it isn’t within the workflow, it becomes unusable even if it is a great algorithm. So, that’s what we’ve done.

Now through the pandemic, healthcare has witnessed a big change – something that would normally have taken 5-10 years — and that is, the basic delivery model of healthcare has changed from care being limited to inside the four walls of the hospital to it getting distributed and virtualized. And the cloud technologies have enabled this distribution and virtualization of healthcare delivery.

We work with multiple cloud vendors — AWS, Microsoft — and have a multi-cloud strategy. We are open to working with other cloud vendors, too. With AWS we have, a deeper sort of integration currently, and some similar work is being undertaken in some other application domains with Microsoft.

This entails taking some of the clinical workflows and applications that we are building to the cloud so that we can help the providers distribute care across geographies and take it outside of the hospital to the patient. When you do that, you must ensure data privacy, and HIPAA compliance. We encrypt information addressed in motion and are looking at technologies like confidential compute so that even the last mile between the storage and the compute infrastructure, is secure.

Q. How are you demonstrating value? How are your clients seeing the value?

Amit: Essentially, this is an area in which you can’t create a black box solution and deploy it. You need to work hand-in-hand with customers, providers, hospital systems and at times, with payers to really integrate things in a way that is visible to all, and everyone benefits from the accuracy, productivity, positional standpoints. Early engagement in terms of problem-solving is key for us.

Once we get into that dialogue, we need to really set a target. Stroke, for example, is an area where we want to really improve outcomes for patients. The stroke care pathway in general merits is a very detailed conversation about what the workflow is, how the clinicians do their work on a day-to-day basis, what the patient journey really looks like etc. Once you map that and then, get into very specific solutions around the pain points that we are trying to solve, that’s taken care of, on paper.

Post-analysis, you think an element can be reduced by 30% and another, optimized by 40%, reduce patient wait time can be lowered by X%. But you still need to create evidence around it. We’ve done two things — put in sufficient amount of telemetry to everything that we do from a software perspective or from a platform standpoint so that we can capture what’s happening when these applications are deployed. And that goes a long way in creating the evidence about “the before” and “the after.” We’ve also worked hand-in-hand with the health systems to be able to capture that information because a lot of that information gets registered in their systems, and we can work with them to see how we can actually look at the evidence and maximize what we get.

We’ve predicted something in terms of optimizations or operational efficiencies or accuracy or patient outcomes. But the next step is extremely important, and that’s how we work with the customer community.

Q. Increasingly, there will be more data versus more volume and velocity of data. All this can be aggregated, analyzed to drive health care outcomes through advanced analytics and AI. But have we been hobbled by interoperability challenges or self-inflicted problems, such as, algorithmic bias, data insufficiencies, issues related to the acceptance of AI in clinical decision making? Is the vision on the right track or are we further way from the goal?

Amit: I think we’re accelerating. What’s happened through the pandemic is, a lot of things that would have taken many years, actually got implemented very quickly. So, there is in fact, an ever-growing need and a push to deploy more analytics and use data more effectively and faster than before. However, for AI to be effective, the variety of data is very important.

What people have learned is that it is not just the quantity of the information which is important, but the variability of the information across different geographies. Different genetic makeup of the patients is extremely important. And that’s where people struggle, from the AI perspective.

Secondly, a tight integration in existing clinical workflows is noticed because you might have a great algorithm in place, but if it is not integrated in the clinical workflow, it is almost unusable. People underestimate the power that is required to actually do a deeper integration into an existing clinical workflow. That can be a significant barrier if you are not accounting for it right upfront when you actually start designing the full care pathway. Those are the things that need to be taken care of so information is available to us much more effectively through AI and that will change the healthcare delivery model for good, going forward.

Q. The pandemic has forced us to think, in creative ways, about how we can overcome challenges for the immediate future. It’s also laid the foundations for how healthcare might improve with all the virtualization. If there’s one best practice that you would like to share — with your peers or start-up founders, what would that be?

Amit: My learning is that you have to combine the power of computing and analytics with the knowledge of the clinical space. I would very strongly encourage people to form those partnerships with the clinical world. This is a work that needs to happen hand-in-hand with the physicians and the clinicians, and the health systems. So, if you are a startup company working in the AI space, joining hands in a larger ecosystem where you can actually get the domain knowledge, clinical knowledge and then, combine it with all the good things that are happening from connectivity, communications, computing, AI, — you will surely enable the best outcome as far as patients are concerned.

Second, I’d say, we can get very excited about technology, but we always need to focus on the last mile of healthcare, which is — what is the outcome as far as the patient is concerned? It has to improve the patient outcome, decrease the cost of healthcare as delivered to the patient, and help early detection-early treatment while almost going into wellness. But we can lose sight of our goals quickly by becoming very enamored with technology. Focus on the last mile to ensure that every effort put in eventually goes into the patient outcome.

We hope you enjoyed this podcast. Subscribe to our podcast series at www.thebigunlock.com and write to us at info@thebigunlock.com

Disclaimer: This Q&A has been derived from the podcast transcript and has been edited for readability and clarity