S1E1 – Qure.ai – Rohit Ghosh | Journey of an Indian Healthtech Company going Global | Covid-19 Test

Rohit: The second part of the challenge was about generalizability of AI and the biggest thing that we realized is that for AI systems to become robust Most important things to train AI on huge amount of data I think the fact that there’s so much variation in Chest X Rays, right? Starting from how different machines are producing it all the way to how the lighting is, and what is the exposure So there are a bunch of things that can completely vary how the chect X rays look and frankly as mentioned earlier right as someone coming from a non medicine background, to understand all of those variations firstly was very difficult Laisha: Welcome to the CLL podcast series, where we talk with leading companies and startups disrupting the world with AI and ML This is the first episode of the QURE.ai podcast series This week, we take a closer look at health tech startup QURE.ai Which went from using AI to make diagnostic imaging easier and more affordable to focusing on coronavirus testing globally Qure.ai develops deep learning solutions that automatically read and interpret medical images, like x-rays CT scans and MRIs The company’s product is an AI based radiology diagnostic aid and the flagship applications focused on chest x-ray, abnormality detection, and brain CT scan analysis for emergency care This Mumbai, based healthcare startup recently raised $16 million in funding led by Sequoia, India and supported by mass mutual ventures, Southeast Asia For the audience, the same investor Also invested in technology, backed companies, such as BYJU’s, Zomato, GoJek , CRED and many more Now as the world battles a pandemic QURE.ai is ready to do its bit to help curb the further spread of coronavirus QURE.ai is also the winner of the AI game changer award at the sixth NASSCOM big data and analytics summit 2018 Today we have with us, one of the founding members and chief of staff of Qure.ai, Rohit Ghosh Today’s podcast, we’ll walk you through how Qure.ai is tapping deep tech We’ll hear from Rohit why, what and how behind what a tech startup is building We also get an overview of how Qure.ai desrupted the health care products and dealt with the initial setbacks whilst launching their first flagship product QXR, along with its first time experience from Rohit about recent funding, they secured And to top it all off, we will touch upon the many ups and downs QURE.ai Faced on the journey of being recognized as the most accurate algorithm for detecting tuberculosis on chest X Ray So let’s dive deep into the origins of Qure.ai Let’s welcome rohit founding member of Qure.ai And chief of staff from the travel enthusiast to a TEDx speaker he Is not only a Google developer expert for ML, but also a semi radiologist Like he calls himself Rohit its amazing to see how you’ve excelled in your career and have transitioned from being an AI scientist involved in building R &D products in the computer vision area to being the chief of staff at Qure.ai Today Can you tell me what your day looks like and how do you give yourself breaks between hectic days, especially during this time of working from home? Rohit: Absolutely Thanks, Laisha for firstly inviting me onto the podcast Absolutely great to be here Day on day I think it’s a fairly difficult question Because, as you mentioned, the journey has been fairly widespread across last four years what they look like to what it looks right now, probably before Covid was very different In fact, to be honest, probably before COVID I was not even as tending as most days in office actually A lot of it was actually in client applications but after COVID, I think it’s been a fairly lot more relaxing for me and that they could take some time out, spend some time at home and do it, in terms of taking breaks I think my best sort of way to relaxe myself while working from home primarily to listen to music and read books So I couldn’t do that while I was doing all this hectic travels and all of that today I do have that opportunity if I’m hooked onto a book, I really love reading a bunch of books So I take time out between those meetings and sit nicely and can read that up So that’s, I think is my biggest break that I can avail myself as of today Laisha: I think I can totally relate to you.as I’m also working from home these days, but yeah, let’s now move to talking about how Qure.ai started We know that Qure.ai was found In March of 2016, by Prashant Warrier, Pooja Rao and an amazing founding team of AI scientists with clinical scientific and regulatory knowledge, Qure.ai thought of different uses of AI and deep learning

AI powered choice for kids that could grow up with a kid AI based fashion, and then the team wanted to apply AI into healthcare So what was the thought process that went into using AI for healthcare and starting with solving one of the great problems in the world? The inflated demand for radiologists, Rohit: within all of us, there was a strong, Intensity to make an impactful, innovation I think impact was one of the major thing that sort of drove all of us and still drives all of us And I think in terms of the ideas, the fact that healthcare systems specifically in countries like India and in other countries is hugely broken and there’s There’s really big scope of innovation that will completely change the paradigm of healthcare At least back in 2016 It definitely seemed like so from that perspective, we definitely wanted to choose healthcare because, and the idea we actually set out also the target and what is still the motto of the company is to make healthcare accessible and affordable for everyone So I think that was primary the reason why healthcare zero down upon yeah And I think within healthcare, the idea was more, when we started, it was not really as clear as to what exactly would be the journey we would be looking at, within healthcare, we started off on a different track and I think over a period of time, we realized that mean in terms of impact, the biggest problem that currently most places have Was around the lack of radiologists and that’s the problem that seemed definitely like the first frontier for AI to conquer So that’s how we went about that whole business of looking at AI primarily for radiology Laisha: Okay And then you slowly transitioned into solving the TB problem as well in India and across the world Rohit: Yeah, I think TB was fairly down the line I think it was probably somewhere around April or may around that 2016 we sort of, decided to go ahead and look at radiology and within radiology there are a bunch of modalities, body parts that we were looking at early on and, somewhere around, I think October of 2016, we started looking at what we can do with chest X rays, given that the chest X-rays are normally the most highly reported voluminous scans, and there’s not a lot of radiologists who want to read x-rays so plain films are the most neglected domain across entire radiology, and that’s where we started looking at and I think once we started looking at Chest Xrays the questions then came along was what within chest x-rays we want to look at and focus that so from that perspective, TB definitely seemed like something, which was it actually hit home, right? Because, in India we have the it’s India, it’s called the TB Capital of the world And the few things that we are probably, not really proud of And so there definitely chest x-ray is critical? core part of the diagnosis process And we figured that the AI systems currently existing systems at that point of time were not adequate and capable enough to take care of great in standing his reads And there was a huge value that we could see right away, so that’s, I think somewhere around it ended up 2016 start of 2017 We started looking at what we can do for TB Laisha: so for the audience, I would say India with 2.8 million cases, annually is, Being counted as the TB capital of India, as you mentioned, and also Qure.ai is solving one of the major health problems of TB in India, but, as an AI scientist and as a founding member, you’d agree that it’s difficult taking a product from a research phase to a market So the early life of the product, started with a glitch The first time your team deployed QXR at a radiology center in India, around two and a half years ago There’s all a drop in accuracy from what they had observed in training and test datasets to what the doctors say, no doubt, training, deep learning models, especially in healthcare Is only one part of building successfully AI product and bringing it to healthcare Practitioners is a formidable and an interesting challenge in itself So how did the team handle the problem of generalizing the AI algorithms in place and achieving more than 90% accuracy in detecting 15 of the most common chest x-ray abnormalities today? Rohit: Yeah, I think that was actually the incident that you mentioned So it’s still vivid Remember that? So back in 2017 and we saw this huge jump drop in accuracies of when we deployed with them So there were a lot of learnings from that particular incident One of the Most important ones, if I would call it was the fact that interpretability is something that we need to do, not just for the sake of better user experience for radiologist, but also something that in generally

Help us understand what AI is doing So that was, I think, one of the most critical things that we understood during that particular incident So that was, I think that was one of the major learning Second learning was obviously in terms of testing, And how you evaluate its AI algorithms and you know how to make it more generalizable And generalizablility as a concept was something we started dabbling with back then And I think both those pieces about AI interpretability of AI, as well as generalized AI was something that we mostly saw firsthand, how it can completely affect deployments and the performances So from that point onwards, we started working, I think, in both those aspects Interpretability obviously we went ahead and we have published some of the research work that we did in building interpretable systems for AI, specifically for medical imaging Because again, a lot of interpretability work at that point of time was primarily around what was done on natural images There was not really anything that was out there, done for medical images So it was a huge task to take that You stop learning from natural images convert it for medical images and put that out So that was one part of the problem using interpretability, how we went about it The second part of the challenge was about generalizability of AI and the biggest thing that we realized is that for AI systems to become robust Most important things to train AI on huge amount of data I think the fact that there’s so much variation in Chest X Rays, right? Starting from how different machines are producing it all the way to how the lighting is, and what is the exposure So there are a bunch of things that can completely vary how the chect X rays look and frankly as mentioned earlier right as someone coming from a non medicine background, to understand all of those variations firstly was very difficult And once you understand some of those things, then I think it was more of how to make AI systems robust to all of that And I think huge amount of data and also a lot of Other research techniques that we can incorporate it from naturally made and other domains, not only images, other domains of machine learning is, but then being incorporated in a spot of power Processes that I think made the whole system a lot more robust and yeah, I think, but that’s a very important thing you mentioned I think today when we talked to a lot of people and people in building up their AI system, this is an important message that I always want to relay That building part is a very interesting and obviously, adventurous part of the journey, but also apart from that to make it real is a Fairly longer shot And to do that, it’s also requested not more than that than not yeah, not more dedication and effort to make it through it Laisha: totally agree with you Rohit on that So I’ve also seen that Qure.ai Has very, has been very active in publishing white papers and putting out their research and data sets in the world And I would like to know your opinion on the gap That’s there between the academic research papers and building a commercially viable software Rohit: absolutely I think of firstly, I would break them So when you say white papers, so these are mostly all the work that we have been putting out in terms of peer reviewed publications and within them, I think, there’s obviously a worldwide We have obviously strive to make as much of publications out in the open and be as transparent about that entire process I think the bigger challenge that is there is putting that into commercial use case And the major reason why I see that as a challenge is firstly, in terms of say for example data, right? Or things that we all put out for peer review there’s nothing really that while, myself also as a reviewer for a bunch of journals And the thing with that process of peer reviewing some of this work is that you are primarily trying to save the AI, training And, some of the way it has the inferencing has worked Is that sort of correct? Is that working as per expectations, but what you’re not checking for example, is that what is the amount of data that has been trained if the data is enough to come to the conclusion Did the data include all sets of different varieties that it would probably the AI systems would seem real, so it doesn’t include all of that And it’s very easy for us researchers to for example, quote some performance, say on a hundred or 200 images and go out with that And when that is extremely, obviously there’s a huge challenge let’s say something which is, say detecting cancer from early cancer from some bit of medical images, that’s definitely, doing it even on a hundred, 200 images of fairly common developer, but in reality to put something out like that would be a dangerous proposition because you do not really know how they really work in real life

And all of that So I think today there’s not much of a publication, which actually is on a huge amount of data sets or most importantly I don’t think there are a lot of publications out there, which actually talk about how AI systems are actually benefiting a patient It’s obviously a lot of it is around accuracies or efficiencies, all of that so obviously some of those things are definitely lacking when you look from a commercial perspective, Laisha: So Rohit talking about data Qure.ai started with a lesser subset of around 50,000 X-rays, and then finally went onto curating a database of about 2.5 million x-rays so creating a data set of such high volume is not easy So what were the initial challenges in terms of data collections and what were your sources of data? If you could share with us and how did the team handle all of this in the beginning? Rohit: absolutely It’s an interesting story actually when we started out, we were probably not even at 50,000 we had, then we actually started off, it was completely just scraping images from that site manually looking at them and doing that on a very as we’ve got hacky basis, and while that was good to get started up, we definitely realized, even way before our first deployment and even not maybe for thinking about generalizability, we realised that we definitely need a huge amount of data to make some of these systems as really robust And that’s the point of time we started looking at what we can do to make the systems more robust, get more data into the system And, we tried working out with different, partners, I would say primary healthcare providers, or, some kind of radiology outsourcing platforms and different partners and universities as well, and tried to work with them to understand what sort of problems they want to solve around chest X- rays on not only chest X-rays other parts of imaging, but primarily focus on X-rays chest Xrays and we tried to see what we can do for them And that’s how we got started with that process of, working with different partners and, engaging with them to provide a solution at the solution in exchange of the data that they were providing And we obviously have to make our processes working ensuring, whether the Patient consents and everything very quietly for all of that So that was a long journey and it was definitely Not really long it was definitely worthwhile as the, it was tough, but it was definitely worth while to work with all our partners and, engage with them A lot of them on had reservations around sharing data So that also was perfectly fine for us at that point but yeah, I think overall, some of the partners definitely agreed to work with us and that’s how we have the huge amount of data that we have today So I think that obviously is one of the, what a period of time that happened And once that was all of the data was inside the system, I think it made the AI a hundred, to 1000 times better than what it was back in October of 2016 Laisha: Okay Definitely data plays a really big role in making algorithms robust and actually putting it out in the market And that’s why Qure.ai Have come a long way And after the successful launch of chest X-ray interpretation tool in 2017, it bagged many awards globally, Top 30 promising technology startups in India, the net exploRe award for innovative healthcare AI from over 2000 globally technology initiatives NVIDIA social innovation award and winning the digital pathology challenge at MICC AI So how do you think recognition on these global platforms has helped Qure.ai to create a far reaching impact for healthcare and technology in health? Rohit: absolutely laisha now, actually when we went through that list I was also like, okay, that’s a lot of recognitions than we have, but, but yes, we have, thanks,to what we have been able to build over the last two, three years, we could actually get a good amount of recognition quite easily and obviously add to that, the publications So I think that, for the first time there was some Indian company that actually went out and, Tried to win on these kind of challenges or you know getting this kind of global recognition So that definitely mattered to us and it definitely helped in the perspective of how Qure is I think looked at it’s not necessarily looked at as an Indian AI company trying to just do innovations within the scope of this country, but it’s primarily, I think that global theme of what we w can do when a kind of quality that we can provide sitting right out of India that Testament and then validation was definitely something that came as part of this kind of recognitions on international platforms and the international media coverage is that we got along with those So I think our aspirations were definitely there from day one to look at how, what we can do with Qure and make it to a global organization but yeah just having that aspiration, it was not enough to be able to go out and

get some of this global recognitions definitely established Qure from there onwards and thats something yeah, that’s going to look at the entire global healthcare system and think about problems that we can solve sitting from India So that definitely helps I think that global recognition definitely helps create that concept of a company, which is not substandard or something, just because thats based out of India Laisha: That’s amazing Rohit And it’s always inspiring to hear the backstory of how you became what you are today Moving forward The next Eureka moment for the team was when you deployed the QER head CT solution to the trauma and stroke cases at the hospital in India around April, 2018 And head CT scan solution was also showcased at GTC 2017 and SIIM 2017, amazingly in March 29, Qure.ai’s traumatic brain injury and stroke triage tool also received CE certification So I’m intrigued to understand what kind of medical research went into creating an easily accessible diagnostic tool And how long did it actually take it to bring out to people? Rohit: I think head CT was something we started working probably Three four months later on when we started working on chest X-Ray our first sort of submission international So there’s this conference called RSNA, which is radiology’s, probably largest conference across the award So RSNA in 2017 was the first time we, had, sent something over for head CT and then got approved there So that kind of give us a confidence that this definitely is a good problem to solve and probably there are lot of things that we can do around with it But I think in terms of solving that problem, it was genuinely a lot more difficult than Chest Xrays says which is primarily because of the fact that the images are 3D as opposed to X-Rays which are 2D And not only that, I think that complexities that I’come in with 3D images, like dimensionality and some of those problems are also fairly critical for us to solve before we could even get something that was robust out there for example, I remember this story and this probably I have told to lots of people I think when we were working on head CTs, one of our biggest challenge was how we detect fractures from CT scans And this was very difficult for us because I mean fractures are basically this thin line, right? And it’s very difficult to confuse that from other things that are very similar, looking on a head CT scan, and it’s almost similar to looking for a needle in a haystack and how we went about solving their problems is very interesting We actually talked to a lot of radiologists after almost one month 2 months of hard trying and solve the problem We couldn’t get away around, so we started talking to radiologists to understand how they go about it, because it seemed like something that human beings would also have a very difficult time doing And then we talked to radiologists, we understood how they do it, and they do actually use some other techniques, which is basically just trying to see where the trauma is And from there, they try and detect patches And that kind of also opened up a new way of thinking for us, the fact that, the way radiologists actually go about solving some of their challenges, how diagnosing some of the things is very interesting and very definitely worth learning From that perspective So obviously started doing some of that already during 2016 and we had started, but, how to actually go on a much deeper level and use some of that for solving this problem was very interesting and that completely Change in the way we were looking at problems to solve and, and it’s problem The idea is that AI is obviously something that is helpful If you give it a lot of data, it would definitely give it a diagnosis but what you can do to make that AI system, understand better, like when you teach children you give a hint and you can also give thse kind of hints to the AI systems to make them learn better and more robustly So that’s some of the things we picked up, how radiologists do that decision mimic all of that decision making into AI systems So I think, yeah, that head CT process doing that was definitely a much tougher AI challenge if I would call it But once we did it, I think, then obviously the kind of approvals and the validations we got, including our publication in Lancet this was in October of 2018, the hits would be algorithm and the results that we had caught with it got published in the Lancet I think the most prestigious medical journal and the interesting part was there was no AI medical imaging publication in the Lancet before that And this was lancets first ever AI in medical imaging publication in October

So yeah, that was a huge validation of the fact that what we are on the right path and then the CE approvals and everything started coming in, which obviously, and then my bunch of different users, as you mentioned, the first one was in Kerela and then on the journey has been onwards and upwards Laisha: Amazing! Talking about the CE approvals and FDA approvals Let’s fast forward to May, 2018 The QXR became the first AI based chest,x-ray interpretation tool to receive the CE certification Tell us about, more about how Qure.ai expanded to other countries and dealt with all of these data compliance laws to CE and FDA approvals Rohit: So in terms of expansion, I think, there are two parts to that, right? One is whether there’s actual use case to solve And the second part is primarily around the regulatory approval tasks and they do that So when we started, as I mentioned, some of the global recombination that we had bought, obviously set up the stage for us to look at something broader than what we were doing and to do that, obviously we had to prove ourselves, what? We have quality in a international platform to see definitely is one of the most stringent regulatory authorities responsible for entire Europe and across the world It’s a fairly recognized, global regulatory authority So that definitely made sense for us to approach CE for certification and, getting the product approval and obviously with the quality of research that we had done and the performances that we had achieved, it was fairly easy for us to get there And once we did that, Then the idea was to finally see which places have need for something like this, just because we have an AI solution which can look at chest x-ray and give automatic, diagnosis does not necessarily mean that The world is in everywhere There’s a lack of radiologist that needs to be solved So from that perspective, we realize that a lot of the things that we’re doing on TB, obviously extrapolates to a lot of places in Southeast Asia and other parts of the world as well, where they’re struggling with tuberculosis as a problem at that point of time, it was surprising to me also to know that almost a hundred odd countries across the world actually have that as a very significant problem And part of there Next I got tenure and votes actually include complete TB eradication So that was definitely there Apart from that, with the chest extra, we found there was a huge use case around a high volume centers, right? High volume radiology places where they do a lot of extra scans and they don’t have enough number of radiologists to look at that so that’s when we devise our algorithm, which can completely separate the normal chest x-ray reduces a huge workload, and that had its own use case in other parts of the world, specifically parts of the world, which are in developed countries, but TB not necessarily is a big problem, but the bigger challenge is primarily they don’t have as much of radiologist apart of Europe Also in remote parts of these developed countries where we don’t have enough number of radiologists reporting So I think those, I think that the primary ways in which we looked at expanding back in 2018, obviously over the period of last few years, the more use cases that the extra software can do, and that has enable access to even in different other geographies or segments But primarily back then it was, I think those two use cases were there that sort of led the way Laisha: I should say that it’s has been quite a journey of many firsts and lot of ups and downs And, it’s commendable, how Qure.ai Is working towards solving the problems that a large scale, where there’s one radiologist for one lot people, especially in India So what is the driving factor in providing the solutions at such affordable rates? Like I read that it’s around $1.50 for radiology interpretation Rohit: so for us, I think the, if you’re asking for the drive, the idea was obviously embedded within Qure back in 2016 when we started, which is to make healthcare more accessible and affordable so if we had to look at base image we could completely change or infact atleast the major healthcare delivery systems The, obviously one of the aspect is how you can make it more cost efficient And the other aspect is around obviously quality And I think both of them are fairly important when talking about healthcare delivery systems, you obviously want to make them affordable as well as at a good quality It’s not necessarily with compromising the quality So I think keeping those two things in mind we always wanted to look at ways we can bring down the costs off health care delivery and yeah Keeping specifically in countries like India, Where, it’s fairly distributed There’s a lot of requirement for it, but there is not a very clear path to you know, getting paid for some of those things So from that perspective, I think we did a fairly good job of building a solution that can do that at a fairly, at least definitely much, cost effective and then the existing modes of solution

So that’s something, for example, in tuberculosis, I talk specifically, tuberculosis, the biggest challenge is time, really? before Qure aI solutions were existing, the easiest way to test people was something called a microbiological test Osteo values bomb And that’s one of the reasons why you cannot as a TB program or someone running TB program You do not, you cannot scale that to a lot of people The biggest challenges you have to do with probably what a very few limited people in a small geography, And with the AI And it says what we could do was finally really built in a screening system, which could primarily look at patients It could identify whether they actually are likely to be, even then only those patients are suspected of TB at the ones who get tested, using the microbiological tests and.this Process if you’re using a hundred or 200 experiments, let’s say to test a hundred machines to test a hundred people, So certainly using a hundred machines, you would be able to test out 500, 2000 people, just and leveraging the existing infrastructure on extremes So I think those are the ways which we have Been able to drive innovation It’s just not about the software itself i think its Primarily more about the fact that we will leverage . Existing infrastructure, like extra x-rays right x-ray is almost there across all hospitals, different kinds of vans, all of those So just may exist using some of those standard things We have been able to also lower down the whole healthcare delivery cost of healthcare delivery in some of those geographies Laisha: Awesome So talking about scaling up and taking things forward earlier this year, Qure.ai, secured funding of $16 million, big congratulations to the whole team And recently they have also been recognized as most accurate algorithm for detecting tuberculosis on chest X-rays for the stop TB partnership How do you feel about this and how has such my students shaped the QXR sat as a product and help Qure.Ai create a name for it globally Rohit: I think you’ve mentioned two things So the first one is funding So thanks for the wishes and yeah, that’s obviously a milestone in the journey and it’s a validation on the kind of work that we have been doing, but yeah Talking about the publications, I think, yeah, that’s a phenomenal publication because as I was mentioning earlier, And you had asked this question about what is the gap between what had been done academically and what has been done commercially? The biggest challenges has mentioning them not is that there’s not a single data set, for example, to compare say different AI solutions your AI solutions are not completing on aggravate, so there’s no way to know, if one AI solution says 99% accuracy vs someone says, use the 91 or 92 There’s not really a way to understand which one is actually marketed because they’ve been doing on their own data And there’s nothing really to understand how to go about that process and a clinical fusion was trying to understand which system to look at And this was a second dimension This is stopped in partnership back in October of 2019 They had done a similar study It included three AI companies, and they have looked at patients from 2000 patients from the Nepal and Cameron, and they did this similar kind of study where they compare the AI solution And what they found was that 2,000 is probably a very small number to actually a certain, which one is good So what they did was in this second round of studies, if I would call it this to look at 23,000 patients, and not only just three AI companies, they looked at five AI companies and They did this over a fairly good month or 2, I think almost two to two and a half months of effort They’ve put in a huge kudos to that entire team to do this and have an in depth, independent evaluation exercise for the betterment of the entire community So once it is that spread out, I think it was fairly, very strong validation of the quality of research that we have been doing so far The fact that on a 23,000 datasets among five AI companies was the best performing AI solution.and by a margin That definitely boosted And then not only the moral of the team in terms of what we had been believing in and, being very focused, I think, as challenged with Qure and the way we have been looking at this, as we have been very focused on say a few things, we have not really dived into bunch of, yeah So for example, their companies which for AI called chest CT, abdomen CT, 15 different things, that’s it but for us, we have been very focused on chest X-rays and head CT and just because we have been very focused on just doing those two things exceedingly well, So that strategy we’re actually right in going ahead with that strategy, was something we got a validation I think that this kind of results come in And obviously I think in overall sense, I think the very fact that all this five AI companies performed better than human leaders also was a very strong point

I think, that actually helped people understand that Okay There’s a possibility that Yeah, Ai could actually perform better than human readers in certain kinds of, tasks and it probably might make sense to use AI, at least for the screening kind of a thing So that knowledge, I think, was most important to be Brought out to the rest of the world and I am genuinely big congratulations and kudos to the stop TB and for bringing that whole message out, about AI and what different options you have within AI and how it performs on similar Yeah So Laisha: I would agree that, getting approval and getting recognition in health tech it’s a long journey I’d like to shift our focus to the current pandemic that we all are dealing with the COVID-19 pandemic So Qure has released a COVID-19 progression monitoring tool and pandemic response platform And basically what it does is to sort patients into those that need immediate attention and to those that can live under home quarantine It’s keeping used to monitor disease progression in six countries, including Italy, Sandra fell university hospital in Milan So I want to understand what went into tailoring the QXR product to cope up with this current problem of COVID-19 so fast Rohit: primarily all the great work was done by the R&D team to make this radiant I still remember the meeting in march 2020, just few days before the lockdown Thing happened in India and we were really discussing about what we can do for Covid and, the most, In all of these answer was obviously what we can do with the QXR product, because that already had, see, the different findings, radiological findings of covid were already, something we will see approved for So it was more about understanding, what exactly how those manifestations look actually for what are the characteristics of them? for example, if it’s an opacity it’s unilateral or bilateral, and different kind of medical characteristics of those kinds of things So it was more about understanding that piece So there was thankfully again, a big time So then that research radiology community out there put out a, lot of research efforts still then based on the patients in China, what are the kinds of findings that they’re seeing from this patients? So that made our job a lot more easier because we could finally understand very easily going through some of this literature it says is that, this is the kind of findings that we already have the capability to detect, and this is how the findings normally look so keeping that in mind, I think we jumped right there with the action to see what we can do with that kind of data, and then involved with some of the partners to help us get some access to Covid specific data, which made the best thing and the training of the algorithm, them a lot more stronger And I think just around the same point of time, we, because of the fact that we were doing already something similar with chest x-rays, we already had a lot of interests that were coming and Sandra fellows, for example, I think the first site we went live in back in know somewhere in early April And I don’t remember exactly But yeah And then onwards it’s been, yeah, you obviously, with that, particular deployment, we also learned a lot of different things about how patients are using it, how physicians are going to actually think decision based on that And once that story became clear, then we could expand into other geographies and other locations across the world, trying to help them The challenges, obviously, AI is build that before that day, I was really, something was just, people are looking more with the IFN suspicion What happened with covid and the pandemic was the healthcare systems across the globe, got severely burdened with the kind of, Workload that was coming in And there were a lot of centers I know where I had talked to some of this, radiologists And they were spending almost a weekend that entire days from morning seven to almost midnight in the hospital wards, trying to help people, even some of them even did not have there protective kits along with them So it was a real big challenge I still remember talking to physicians across the globe was actually a Massacre of sorts And at that point of time, AI could definitely, suddenly act as someone who’s almost part of the team, right? Except that it’s not a human being, but it’s a machine which is able to work for them and help them reduce their workload this up pandemic Also changed their perspective of what AI can do and people looking at it from a suspicion, to looking at it with more of a support tool So that a shift in perspective also, I think broadly happened during this first day during this, that pandemic and after I think we bend it like that, it leads then obviously there are a lot of other places, including India and UK, and he just bought in as well

we went late So this, I think created the foundation stone the stories that started coming out of some of these locations, how physicians using it, paved the way for more physicians globally Laisha: Great So talking about challenges, Qure.AI also tied hands with the municipal corporation of greater Mumbai to start using chest X rays to detect coronavirus cases What was some of the challenges that you face with the density of population like that? Or from mumbai Rohit: Yeah so ironically the entireteam, at least before covid was, is based out of Mumbai and I’m still in, based in Mumbai And the fact that while Bombay soon, it became very clear within the first one month or so that there’s going to be a Bombay is going to be a hotbed for some time And the fact that we could Step up to that particular need in the city that you are situated in was a big matter of big pride for all of us And when we started with Bombay municipal corporation, I think, the first aim was primarily to look at some hotspot sites across the city where we can deploy this, where this can help in doing that, screening for people and the immediate challenge in Bombay was that I think a lot of people were assymptomatic carriers Not really, I think it was all set knowledge by the municipal, Corporation then there’s almost 70, 75% of the people assymptomatic which meant that there are a lot of people who were completely showing no signs, till a certain point of time, and then suddenly within a matter of three, four days, they would suddenly become critical and then there will be, need to be admitted And there’s obviously a challenge around getting the hospital beds So using the extra solution what we could do was specifically pick up this kind of people when showing no symptoms, but had already some kind of chest infection pick up those patients and bring them, triage them for RTPCR so that was obviously the biggest use case that set up came out Of Bombay And once it did this deployment in some of the hospitals say, it became really something that made a lot of sense for everyone And then, what obviously was asked for us from us the next step was that can we deploy some of these things in mobile vans? The reason being, a lot of population in Mumbai is primaryly densely located And, for this kind of people, you asking them to come all the way to hospital to get tested was, something that did not that sounded a bit probably challenging in terms of obviously the entire transport and everything locked down in the city So what we did was we put this extra assistance AI in vans and the vans would go out and test people, in, specifically in that containment zones So the homes with the highest covid activity and then people who were likely to have COVID or was showing any kind of chest inflammation those patients were sent for RTPCR test right then and there So this, I think the way we did it in, Bombay is fairly different from what we did in other parts of the world and that’s primarily because of the kind of city Bombay is and challenges that are there in, of doing this in hospitals centers But, this definitely, prove to be a very fruitful measure of how we tackled covid I think it’s been phenomenal I think the way they have tested out people have their number of people they’ve tested and also how they manage those patients that have been tested It’s sending them to a isolation, home isolation that has been phenomenal Laisha: I think it’s amazing how QURE.ai is contributing to flattening the curve, not just in India, but globally and kudos to your team for that coming for to how Qure.AI has been a part of many firsts in getting CE approvals for QXR in coupon Firstly, congratulations on receiving the 4in1 FDA clearance for QVR head CT, scanning and aI solution for the audience Can you tell how Getting this approval is a big thing in the health tech field Rohit: absolutely Firstly, thanks for that Yes It’s been a fairly first pearl in Indian AI company and we are proub to be leading that way FDA is obviously the most stringent, in terms of regulatory authorities across the globe and, FDA stamp of approval Definitely makes not only.It’s all it’s meant primarily for usage and approval within USA but because of the, various process and the Varied, I think it makes it a completely acceptable standard across the entire globe and fairly respected standard If I would call it so to get an FDA approval for any kind of drugs or any kind of medical device is a fairly big deal that way just because of the Global approval, that you get the same product and the fact that FDA goes about that process very strictly and stringently also mentioned the means that anyone who’s going through that process and getting an approval and definitely has done their due course of diligence and their work has been fairly, reviewed by everyone involved

So from that perspective, FDA definitely makes a lot of if I would call it in terms of that approval in that sector of Ai but outside that, I think the other big, not big thing, but I think the other thing is that, since anyone across the globe can do that, and anyone can apply for, you have to work being an Indian company and to be able to go there and, do this completely remotely, actually that was also fairly a challenge, I would say to do some of this interactions and, completely from sitting out of India So that definitely also was something that we were not expecting, but then that fact that we did it and we could do it successfully, definitely helped us as well Laisha: Amazing So what our audience again Qure.ai At is present in 28 countries today and has impacted more than 600,000 lives And its technology is being used in what many countries? For the screening of tuberculosis That’s not all recently nodded MedTech has joined forces with Qure.ai To provide radiology practices with superior artificial intelligence in the Nordics and Baltic regions like Sweden, Denmark, and Norway So with all of this going on, what’s next for Qure.ai? Rohit: I think a big part of our focus would be still , in terms of product Let’s probably speak of that first, as I mentioned, we have been extremely focused, in terms of what we have been building on chest x-rays, and that there’s a reason we could build, and kind of solutions that we have been able to.build And I think from a product perspective, we obviously want to keep looking at this Particular modalities and what we can do to make the solution that you take to the next notch, do more findings And we can detect some of these things apart from that obviously, as I mentioned earlier also that we definitely don’t want to limit ourselves to a particular geography or a particular kind of market So from that perspective, obviously, as you mentioned, Nordic MedTech is there and there are a lot of our partners across the globe with whom we have been working to sorta expand our access to other markets The goal, obviously longterm goal is, as I mentioned with Fairly probably as part of our mission is to make healthcare more accessible and affordable and keeping that as a, in mind, we definitely want to look at problems for which, for example, healthcare is not problems, but I would say it’s the patients are, people who are delivering the healthcare services, what are the problems they’re facing in terms of when it’s not accessible or affordable and how we can use the AI to make some of that possible I think from that perspective, we are definitely looking to do probably different products that we can look into and, expanding what we are already having and as well as expanding to other geographies, get the regulatory approvals for all of those geographies That’s definitely part of our one year plan That Laisha: sounds amazing And it has been quite that involved and interesting A discussion so far So if we wanted to have a little fun with you in a segment that we call rapid fire So let’s start with the first question Walk from home vs in Rohit: office absolutely work from home I don’t do any to extend answer, just one word is fine? Laisha: Yup You can explain the answer Rohit: reading books and you know what better not to waste 2hours in traveling So definitely work from home Awesome Laisha: Are you a morning person or a night person? Rohit: I was night person, but with COVID and work from home, I’m a morning person now Laisha: Okay Success vs fame Rohit: depends what we actually call success theme definitely not is important, but it depends from person to person Laisha: Okay So next one covid 19 Boon or a bane Rohit: boon for definitely, both from a professional perspective Definitely what Qure.ai Has been able to do And personal perspective as a work from home thing has really worked out for me So absolutely a Laisha: boon right Engineering versus medical Rohit: Oh, this is stuff, engineering still but, medical you go back to my 10th and I had to do this all over again I would still not do it I love it I really loved reading about it Interesting aspects, but I guess the I’m better off doing the, the computer science pieces So Laisha: you’re happy being the,semi-radiologist It’s just true Rohit: Yeah I guess that’s what I have to settle for in this life Laisha: awesome Mathematics vs statistics, Rohit: maths anyday Statisticsis fairly an expanded version of math I don’t really look at it very differently from there, but, inherently, since childhood I learned, I have trying to read a lot more on maths

I did just It’s a lot more fancier for me Laisha: Okay So we are almost at the end of the podcast and we wanted to know you a little personally So why the name Qure.ai? Rohit: Oh, no, this is funny Actually, I get asked this a lot of Qure actually the reason we book event with that is we wanted to have cure, with the Q This is how you C right in the original cure and we didn’t get a domain for it That’s why we settled for cure with the Q it just sounded yes, sweet subtle and also Catches your attention So that was about it Laisha: Okay So you’ve been to many countries across the globe What is the one place you want to travel once this pandemic is over? Rohit: I’m genuinely I don’t want to travel I use, I used to hate all that Java schedule that I had, but if I had to actually look back and, one only one place that I really enjoy I think that was in Azerbaijan, which is Baku that something I really love, definitely given a chance I would want to go back there Laisha: Okay So living in Mumbai, what’s your take on Mumbaikers lifestyle Rohit: I love it I hail from Kolkata, which is fairly laid back and which is, not as active if I would call it and it has got a warmth of its own, but Mumbai for me is a city of dreams Like I moved here almost 10 years back, exactly, almost 10 years back And, it’s been a fancy journey from, not only the college, but also after college So for me, it’s a city of dreams it’s a place of constantly working towards your dreams and achieving them And this city definitely rewards you for working hard So for me, that’s for me what mumbai really is Laisha: so with this, we come to the end of episode 1 of the Qure.ai Podcast with Rohit Thank you for sharing your journey, bringing affordable, accurate, and faster medical diagnosis Worldwide The pandemic may have put the brakes in some plans, but the team is nevertheless looking to expand Qure.ai’s Product portfolio globally Thank you Rohit: Thanks Laisha: a lot Rohit: for inviting me on the podcast Laisha and Kunal thanks a lot for letting me speak about the journey and, awesomess that we have Like something pasted on the chat box If you could just take a minute to just read that, it’ll be helpful for us to end it Nicely Thanks for calling me for the podcast It was a pleasure to be here and was definitely fun Make sure to subscribe to the Co- learning lounge, YouTube channel and join their community Stay tuned for upcoming episodes in the series We are going to talk a lot about what other aspects of Qure are and they’re definitely doing fantastic work building tech communities across the globe So please definitely support them