Latest / AI Meets MedTech / AI Inside MRI: The End of the Service Call as We Know It | Philips
Transcript
- Anna Nesterova: AI is inside of your MRI machine. The question is, who else is watching? Welcome to limited series of AI Meets MedTech, dedicated to a conference which is called MedTech Meets AI, which will happen in Barcelona on October fourteenth and fifteenth. Over the next few episodes, I will be talking to some of the keynote speakers of the event. And today, my first guest is Binu Gangar Haran. He's a senior director of services and solutions at Philips, where he leads the MRI business unit. Binu will be speaking at the conference on Gen AI and MedTech and Field Service and risks and ethics associated with use of AI. Binu, welcome to AI Meets MedTech. Binu Gangadharan: Gl glad to be here, Anna. Anna Nesterova: I want to ask you a few questions about how your company is implementing AI in practice. Let's start with well, MRI business unit is the one that you're leading right now. let's start talking about what's been changing in the last five, six years since we've seen this boom of AI. Binu Gangadharan: Yeah, thanks for for the for the invite and and the in the session today, Anna. I would talk about MR service in general, but I will also talk and represent the industry service. and and predominantly when we talk about gen AI in field service, I mean it is human intelligence is the foundation, which I always say AI is a multiplier. guardrails don't stop us down or don't slow us down. They are they are what let it go for us further, right? And safely. which is probably the the key aspiration and and the topic that I'm going to talk about as well. Coming back to your question in terms of what has changed, honestly, a quite significant shift from purely a product-only business to solution-driven business. You know, we sh we we we we call it shift which which was led by the value drivers that were there in the industry or in health care in general, predominantly driven by the external factors. You talk about availability of staff, the the volumes of cases, you know, talk about image in in general. It's predominantly for lifestyle diseases, which you see now and and we all know how things are rising. While the capabilities are still limited and it needs space and adoption is lower because the staffing Is very less of trained professionals. So it's a perfect Anna Nesterova: Hm. Binu Gangadharan: storm. And to maximize that value, you would need better tech products, but at the same time, better value solutions. So the drive it towards solution build is changing in the industry if I if I come back to to to your concern and question that you asked. Anna Nesterova: Yeah. So what you're saying th that the role of the service engineers is changing right now, is that correct? Binu Gangadharan: Yeah. y you're right. So the role of service engineers are the engineers used to be detectives arriving in a cold case, up up until now. Now they are arriving with case practically half sold, if if I put Anna Nesterova: Mm-hmm. Binu Gangadharan: it put it in a in a very light way. But we are chasing what we call it as zero unplanned downtime. Big tech equipment, customers don't want systems to be going down, volume of patients are you know, catapult if they are not available for a longer period of time. But at the same time their revenues are impacted at as well. Access to care for a lot of people are dependent on them. When you get things get postponed, it it has an impact, a cascading dominoes impact. Right now, roughly one in five service events we act are on trigger not by alarms going off, but by our AI spotting a a pattern in telemetry or the log data before anything actually fails. Now the the that changes the job to your point. It it used to be drive out, diagnose on the site, maybe drive back with the part. you don't know what what you'd need during your first visit. So couple of visits are needed sometimes. now the engineer shows up already armed with likely cause the right part, slot schedule with a hospital around their clinical workflow, not against it. a lot of of what used to be that needed an on-site visit, a physical visit, a warm blood being there disruptively, now gets solved remotely, proactively, Anna Nesterova: Yes. Binu Gangadharan: and even maybe predictive in capabilities coming with IoT acceleration, for example, in future. and and and honest part, the job is shifting from search and diagnose towards judgment, verification, and the harder hands on work that AI can't do. we deliberately invest in training though. even periodic unassisted drills happen to the engineers so that instincts don't autopry, you know, they because the day AI is wrong is the day you need your engineer who still knows how to think without it. So that's Anna Nesterova: Yeah, that was my actually next question. Because if what if AI is wrong or something is not correct? I wonder what skills you think will be necessary for engineers and technicians in the next five years, let's say. Binu Gangadharan: I think they there, as I as I said earlier about guardrails and AI as multipliers, not as Anna Nesterova: Mm-hmm. Binu Gangadharan: it's the human intelligence is a foundation. So what is needed is predominantly how we use validated pathways. How to validate those pathways for AI to follow? How do we write scripts for us to validate for a new person coming in, for example? you also need a lot of capabilities on how these pathways are converging from multiple sources. Earlier you used to be looking driving the car, looking at the rear view mirror, for example, right? I know what has been done in the past and what needs to be done. But now things are converging based on a lot of unstructured data solved in the past, written in the history, or in the documentations of the products. as it is designed and best practices. So if you that convergence and the execution of that convergence faster is the paradigm shift that's happening. Yeah. Anna Nesterova: Yeah. If we talk about remote monitoring now, as you mentioned, now a lot of things are can be done remotely. We're also hearing more and more now about rogue AI and autonomous agents infiltrating government systems and critical infrastructure. Hospitals are Binu Gangadharan: Yep. Yeah. Anna Nesterova: already cautious, right, about connecting their devices and sharing data. And that anxiety, I imagine, is only growing with the speed of development of of AI and obviously all of those functions that can be done remotely. How do you earn customers trust when it comes to sharing data and doing a lot of things remotely now? Binu Gangadharan: Yeah. great question to to to to to say the least. I think that trust part is so relevant. as you said, I mean we've we we heard a lot of these agents going rogue recently, in Australia, for example. But in health tech, that's what the guardrails do, right? They are there not to auto execute, but to be validated as a recommendation to somebody. Now you don't earn trust with marketing claims that we can do. XYZ things. No med tech companies do that. But they earn by letting somebody else try to break that news off the capability first. you know, our you know, I I quote I quote our head of product security who put plainly cybersecurity i i you know was the realm where we validated a lot of these things in the beginning. you know, if you if you turn the pages back to 2015-2020. and it isn't device manufacturers problem alone, it takes hospital, the manufacturers, the regulators and independent researchers all in the loop. And concretely we work with things like FDA backed independent testers, the biohackers, village device labs who try to ensure that we run proactive vulnerability disclosure policy rather than sitting on a problem. Now, the backdrop that makes this existential is not optional. yeah yeah, like you mentioned and like we said, 89% of of healthcare providers have data breaches in less than two years of going live. But Gen AI layer specifically same instincts that the model only ever sees what is explicitly granted. Nothing leaves Anna Nesterova: Mm-hmm. Binu Gangadharan: our environment to train somebody else's system. Everything is a recommendation and it's logged and it needs human sign-off, right? Which means it's not only documented the way a EU AI Act, for example, which is now you know coming close in 2028, but also it is embedded into our own quality systems of that requires it. So trust here Anna Nesterova: Yeah. Binu Gangadharan: isn't a slogan, it's a stack of specific checkable guardrails, as I mentioned op earlier. These guardrails are not to slow you down. These guardrails help us to execute it faster, safely, and to avoid cost of non-quality which should have happened otherwise. So it's not taking over, it's guiding you in a guardrail with a human sign-off validated by somebody who does that execution work. And in MedTech it we have not gone that far where things auto-correct itself, you know, unless it is validated. Yeah. Anna Nesterova: Yeah. And what do you think about the new regulatory pressures that w med tech companies like Phillips will see in in the near future? Because I do think Binu Gangadharan: Absolutely. Anna Nesterova: these types of processes will require more regulations and testing. Do you foresee some changes in regulatory regulatory requirements? For companies like Phillips and for AI algorithms. Binu Gangadharan: Yeah, absolutely. I I I as as I said earlier, I think it is it is it is not something new to this industry. It is a part of it. Regulation isn't slowing us down, it's telling exactly how to build, which honestly makes the job easier, not harder, actually, if you look at it c on another side. Big one, as I said, about EU AI Act, which any any devices in the MDR or medical device regulation classification. Has to comply with, you know, components is automatically classified high risk the moment it touches the market. So you need to comply with it. Now, practically that means service copilot, for example, has to live inside the same quality system. We already run that on the device itself. Risk management, ISOs, for example, not a separate innovation project that sits in a sideline to comply with some new regulation, right? So every Gen AI features which we ship. in field service is grounded. It's answers from actual manuals, case histories, cites its sources and never auto-executes anything, right? So a person always signs off. We have run this playbook before, as I said in the opening, with with cybersecurity, proactive disclosure, third party testing. Same instincts here. So go get we get ahead of the regulator rather than being told what to do. So things don't change because somebody's doing it. We are building that guarded and helping that regulation, which is generally good for industry. So it's and it's not somebody competing with each other. I mean, it's about your own capabilities that needs to comply within the quality. So our internal quality normally of the MedTech companies are much more stringent than what externals force upon you. Anna Nesterova: Yeah. And of course Phillips has a very strong reputation in the industry for doing ethical ethically right stuff. Binu Gangadharan: I I and I said as I said, I yeah, I as I said I think it is it's like the car industry. You can't release things in the market without having certain safety guidelines mapped and executed and tested and certified. Same thing with it's not just with Phillips, with all everything that touches a human, everything that touches a a medical device regulatory, compliance becomes a foundation rather than a necessary. he will it will define me. Anna Nesterova: But all of that also requires a lot of investments of the company. So if you look at the Phillips R and D budget today compared to even five years ago, how much it it has shifted towards AI, software development, cybersecurity, et cetera, versus just hardware? Binu Gangadharan: Yeah, well great question. And I did started with the value that is driving us towards a solution rather than just a hardware, right? and these are all publicly released figures. So back in 2020 fifteen or so, if you look at it, it was a quarter of your of your business. Whereas if if I quote the twenty twenty-five capital days results or you know published numbers, they are up to forty five, fifty percent. of investment. Now these are hard investments done. It was a bet that hardware alone wasn't where healthcare value is heading. And more and more products are built around those capabilities now. Some of the products, whether it is as I said, if you have to build a product because you have a higher electronic capability or a semiconductor capability, but that also needs you to do faster things. Right. So software solution built on top of that. Well, used to take fifteen, twenty minutes of a scan, for example, it takes a minute less, less than a minute, even sometimes in certain cases. Now, why is that needed? It's needed because the loads of patients which are waiting were 15, now they're fifteen hundred. So it's driven both ways, right? And and so it's Anna Nesterova: Yeah. Binu Gangadharan: yes, deliberate move, deliberate investments, but most of the companies. But did the market force the timing? also yes, because of all the margin pressures that hospital goes through, the aging population demographics, staff shortages that we talked about. but all of these arrived much faster than what we most of the health tech predicted. that's why AI and software resents aren't a side investment anymore, they are the growth engine. Anna Nesterova: Which proves my point because that's actually the reason why I'm doing this podcast. I'm really showcasing how healthcare is changing, how AI is truly shaping it into something I wouldn't say different, but somewhat different from what it was even five, ten years ago. So what you share right now definitely is proving that. And without Binu Gangadharan: Yeah, and goes back to the first statement, right, Anna to interrupt you. It's a multiplier for a for a lot of business and health tech is not not in isolation. it it Anna Nesterova: Correct. Binu Gangadharan: is it is definitely challenging human intelligence, which is a foundation even in service, but it is a multiplier in this industry. Anna Nesterova: True. And without giving away your keynote, can we touch a little bit on risks and ethical questions when it comes to gen AI in MedTech? Binu Gangadharan: Yeah, glad you didn't break the surprise that I'm gonna bring there. But but indeed, no the so I I talked about the risk ethics and potential mitigation, right? So risks about garbage in, garbage out. We used to use that word in analytics when the big data cybersecurity world was chiming with big data world. now it's AI. garbage in, garbage out practically means we use a lot of references from data, from solved problems, history. As I said, what is available can be validated. there we can lead with bad information clouding the solutioning, which means hallucination, you know, f forecasting mitigations that we need to put in place. horizontal scanning that we need to do ensuring that the right element is teaching the right agents and the models, not the wrong things. So how do you filter those things as out to ensure that A it is written better in the future? We bake it in when we design a new product and we solve the problem which is there with older systems. Anna Nesterova: Mm-hmm. So what is the biggest risk that you still can identify in field service as of now? Binu Gangadharan: I touched upon being you know, like drills we talk about are the field drills that we talked about. It it not just happen in in war rooms or war games as well. It happens everywhere. We need to train people to be capable for few of future, for the future, but also to make sure that you can roll back and do the basics right. Right. no matter how much what capability I have with the tool, I'm an engineer. to begin with sometimes. So I need to know how to troubleshoot. I need to r know how to write A B C D or tighten that piece of software or or tighten that piece of hardware and how to drill it down to know what exactly is going wrong. And that skill set, that capability needs a different pair of eyes and different pair of training gloves, if I may. And that's Anna Nesterova: Mm-hmm. Binu Gangadharan: changing a little bit. That's probably where we are more to very more cautious. How do we validate what can be signed off? How do we validate the pathways that are being used by tens and thousands of engineers in the field by when it is designed by experts? In the past it as I said, you know, a detective goes into the cold case, it's fine. you know, you know you what you're you can do, but now you're doing it with maybe thousands of of of kilometers away from. So sometimes that needs a different pair of eyes and ears. Anna Nesterova: And also the requirements are higher now. I think the expectations from the company and from the engineers are higher from hospitals too, right? And patients, I guess, because they say, well, now you have all the AI tools. You have to be able to do it faster, right? So I think expectations are different. And you've been talking about is more like over reliance That's the risks we may see if engineers can Binu Gangadharan: Absolutely. Anna Nesterova: over rely on AI and We forget about the basics, as you said. I think that's one of the main risks of of Gen A. Binu Gangadharan: Absolutely. Yeah. you're right, over reliance of it, let me let me switch start no you know you can have a driverless car, but you should know the fundamentals of driving. You may have an autopilot flying, but you need to know. The skills of a glider to to become an autopilot's pilot. So it's the same thing here as well. You can have capabilities, but your fundamentals should not be shaking. Yes, you need new skills, but your fundamentals can't be shaking here. So you're Anna Nesterova: Yeah, awesome. Binu Gangadharan: talking about health tech and you're talking about about people impact, then you have three layers of fallback contingencies and how to go through that. is relevant even in field service. But at the same time, your point. It's also because the external environment that they are in. Number of cases, number of you know issues that come don't allow systems to be down for a longer period of time. You need precision like an airlines industry was a few years back. You can't have an engine fuel while On the flight. Same thing. a scanners cannot be down, a machine scan cannot be down, zero unplanned downtime is a key which if you are setting ourselves as an expectation, it's not that they are asking or the customers asking. We set ourselves that bar high. Anna Nesterova: Yeah. Well I have one my last question for you, Benu. So what are your expectations from the conference that is coming up in a couple of weeks? And maybe what that you would like your listeners and attendees to to leave the room with after you keynote? Binu Gangadharan: So I think two major elements, the developments, capability, of course, understanding with people what are new things and what's the glitz and glamour of of this capability. Where is it heading towards? What are people doing? You know, getting to hear and listen a lot. But at the same time, I would I would like to ground us on the efficacy of what we use in terms of guardrails, in terms of compliance, in terms of fallbacks, in in terms of value extraction rather than competition driven commercial excellence, if I may use that word. It is a need, not a want. Anna Nesterova: Yeah. Thank you so much, Binu. Thanks for being at AI Mits MedTech. Binu Gangadharan: Well pleasure talking to you, Anna. Thanks a lot.