BACK TO Business ResilienceZINNOV PODCAST | Business Resilience
Technology has emerged as a game-changer in the world’s fight against COVID-19. It has helped every industry build a significant amount of resilience, and the Healthcare industry is no exception. A key technology fortifying business against the Coronavirus war is Artificial Intelligence. But have we harnessed the full potential of AI? Do we have the right regulations in place to ensure data privacy and leverage it across the healthcare value chain? In this episode of the Zinnov Podcast, Amit Phadnis, GE Officer and Chief Digital Officer at GE Healthcare, shares his insights on how the Healthcare industry is leveraging AI and how AI and data will become remedy in the current scenario and how it can be further leveraged to enhance patient outcomes and experience.
Nitika: Hi everyone, welcome to another episode of the Zinnov podcast. I am Nitika Goel, CMO of Zinnov, and your host for today. Over the past few months, the entire healthcare industry has been put to test by the global pandemic. In these trying times, technology has played a key role in helping the industry build a significant amount of resilience. And one such technology is artificial intelligence or AI. To help us gain perspective on how healthcare outcomes have been enhanced with AI and ML, I have with me, Amit Phadnis, GE Officer, and Chief Digital Officer, at GE Healthcare. Amit leads GE’s digital and AI strategy business, and has more than 30 years of industry experience, and holds more than 25 US patents. Amit, thank you for being with us today.
Amit: Thanks, Nitika. Glad to be here. Thanks for inviting me to this podcast.
Nitika: Great. So, diving right in. AI is considered to be transformational for the healthcare industry. What has the role and impact of AI been during the COVID crisis?
Amit: That’s a great question. First of all, I would acknowledge, and say, yes, absolutely, if there is one technology that is transforming healthcare tremendously today and has the potential of transforming it even more going forward for the benefit of patients, it is AI. Now, the Covid crisis has been very interesting, to be frank with you. One would have expected AI/ML to play a much larger role during this particular crisis. But the pandemic hit us at a rate that was unprecedented in this particular scenario. AI in general needs data, and it needs ground truth on the data. So, AI cannot, by itself, generate knowledge at this, at least at this point. So, essentially, the way AI works is, when you have good data from patients in the past, and you have annotated ground truth on it, which means that you know exactly what the condition of the patient was, and the condition was recorded. You can apply those data sets to train a model. And then, once you have trained the model sufficiently, and validated it when you apply a new data set. In the case of the pandemic, unfortunately, when it hit, there were not many data sets that were available to train models, to begin with, right. So, it was not like COVID was detectable through AI right upfront; because, nobody had trained a model, and nobody had data around it. But, over the last two months, there were several companies across the globe, including GE Healthcare, that have successfully trained AI-based models for different modalities like x-ray, and CT. We are, at this point, looking at developing a cross-modality model between X-ray and CT and lab information together; because now we have data sets available. And what we are finding is, because of the availability of the data sets and our ability to build cross-modality models, the efficacy of these models in terms of its accuracy, is increasing significantly. And I think, there are quite a few models that are available today, which are being already used in the care pathway – in both detections, as well as, the treatment of the patients. And that will continue to improve going forward, as we as we go through this crisis over the next few months. That is where we are.
Nitika: Thank you. That is a really interesting perspective. And like you mentioned, we now have a deluge of data. We are also able to make certain inferences from that data. So how do you see the use cases? I know you did talk about it a little bit, but how do you see the use cases emerging in the near future?
Amit: Yes, actually, it’s a very important question. See, the use cases are very dependent on the particular care pathway that the country is adopting, and, when I say care pathway, what it essentially means is – If there is a suspected or confirmed COVID case, how is the country or the particular health system treating that patient? What are the steps that they follow? What do they do first? At what frequency do they repeat some procedures? and so on. So, the care pathway is very, very important. In some countries, they first do a test, then they do an X-ray, if required they do a CT to understand the lung condition in this particular case, and then they put the patient on a particular treatment path, and they repeat some of the scans, so on and so forth. In some countries, they are actually leading with CT and the pathological tests, right. So, the care pathways differ from one country to another country a little bit, and that determines how you use the AI models. Take the example of a cross-modality model. So, let’s say that we are developing a model using both X-ray and CT, just to keep it simple. What you typically do is, for the same patient, you need data sets for both X-ray as well as for CT. And you obviously need a ground truth to be annotated on both of those scans, and then train the model with both x-rays as well as CT – Just think about this as an X-ray superimposed on a CT or a CT superimposed on an X-ray image, and you’re training the model with that combined image. The advantage of this type of model is that, once you have trained that model, you may apply it in a country or in a situation where CT is not that easily accessible. Since you have trained the model with both X-ray and CT when you apply an X-ray image, it can be then used to say, for this particular patient, what would the CT artifacts look like on a CT image typically; so, it is as good as taking a CT image itself. So, you train the model, so that when I now apply as X-ray, you can predict what the CT parameters could be, which can come in extremely handy, in situations where CT- is not that easily accessible for patients, right. So, that’s one example of a care pathway. The second example is, you may just do an X-ray-based model, and an X-ray-based care pathway, and a CT-based care pathway. There are other combinations where we are combining the X-ray information with the lab information, and then combining those two parameters and generating the model, so that it can be then used in that particular care pathway. So, it’s very dependent on the way that the patient is treated.
Nitika: Interesting. So, AI is going to start playing a more critical role in the care pathways, right. So, will it be AI companies that are entering the healthcare space? Or is it going to be healthcare companies that are going to be focusing more on AI?
Amit: I think it can be both, right? It can definitely be both. I think what is important to the company is, does the company have access to the right data sets, depending on what type of model you’re trying to generate? Are those data sets properly annotated with ground truth, and what is the quality of that ground truth? And if you want to apply the AI model in different geographies, then you need to make sure that your model has been treated with a variety of data sets from different geographies; because, what we are seeing typically is that, depending on the genetic profile of the patients and a model developed in a particular data set in a particular country, does not work very well when applied to a totally new data set coming from a totally different country. So, the variety of the data set is extremely important, both from a model build-out perspective, as well as from a validation standpoint. And then finally, any AI is useful, if it is seamlessly integrated into a care pathway or in a clinical workflow. So, if a company does all of these things right, it doesn’t matter if it is an existing healthcare company which is trying to do this, or a new start-up, which is trying to get into healthcare. But some of these considerations are extremely important – Availability of data, Good ground truth on that data, and more importantly, variety of information or variety of data sets, and then, the ability to seamlessly apply the AI model in a given clinical workflow. So, all of those parameters would determine how successful this could be for any company.
Nitika: Got it. So, would you also say that the influence of AI and the way organizations are thinking about innovation and partnerships are changing in the current scenario?
Amit: Well, absolutely. As a result of COVID, social distancing norms, and because of the need to protect healthcare workers, most of the health systems are trying to do online consulting to the extent possible. What we are seeing consistently while talking to a number of our customers in GE Healthcare is that a lot of them have said that, now, over the last two months, they have been doing far more online consultations as compared to what they were doing before. Just to give a couple of examples, one of the large health systems here in the US said that they used to do just five to 10 online consultations every day, and that number has gone up for them to about 3000 online consultations in a single day. So, every day they are doing 3000 online consultations successfully. And what they’re realizing is that about 80% of the care pathway can be executed remotely without really bringing in the patient, and then 20% of the care pathway, might require bringing in the patient in a physical hospital setting. Similarly, 80% of the total number of patients can also be treated through online consultations. So, that is having a significant implication in the way that they are treating the patients. And, this has a great amount of replicability as far as AI is concerned, because when you’re doing an online consultation, a lot of initial triaging, regulatory clearances, both on the model as well as from a workflow perspective, can be done through AI; a lot of reading can be done through AI; segmentation, quantification of images, measurements, and all of that, can be done through AI. And that sort of releases the burden from a radiologist’s perspective or from a clinician’s perspective to a large extent. But we’ve seen significant trends in virtual hospitals and telemedicine, remote reads, remote execution of care pathways, in this pandemic. And the interesting thing is that most of them are now saying, well, there are a lot of benefits here. And while the pendulum has swung quite a bit in one direction, 3000/4000 online consults a day, post-COVID or not, people don’t think that this will go back to where it was before. So, you know, most of our customers are saying that they expect 60 to 65% of the care pathways to be still executed remotely, and in a virtual patch, going forward. So, this significant impact is changing the way care is delivered in the health systems.
Nitika: Great. That’s really interesting information. The other thing you talked about at some point is that regulations and data privacy have always been a key hurdle in the healthcare space. Will AI or technology play any role in solving this challenge you believe?
Amit: Well, actually, that’s an interesting question. Because, I think the regulations are necessary – You are dealing with patients/ human beings, and any mistake can have significant implications for patients. So, there are regulations for good reason and they are, really, important. There are a lot of regulatory changes that are happening because of the pandemic too. So for example, in the US, now there are quite a few reimbursements, which are possible on online consultations, or, for example, you could actually extend care outside of your state because you are now virtual, and so on. There are quite a few emergency use cases that have been approved, at least temporarily, by the FDA, in many care settings, essentially. Now, your question is, can AI help resolve some of the regulatory issues? And that I think, is a very interesting question. We haven’t looked at AI from that perspective. What FDA has been working on, along with several companies across the globe, is to get a perspective around how they need to look at regulations for AI, and how do you make sure that you appropriately frame the regulations for AI so that while we encourage innovation to go at a faster pace, we also protect both patients as well as the health systems, concerning how it gets applied, and from an outcome perspective. So, I think that’s where a lot of work has been done. Now, to your point, we could think about a scenario where you are actually using AI behind the scenes, to judge whether we are staying within a certain sort of regulations as defined by the FDA and if you could use that in the background, depending on the data that they get. But, that’s a really interesting question. I am not sure how this could potentially pan out from an application of AI for the regulation itself, rather than the other way around; the regulations that are going to be relevant, if we were to do an AI-based innovation in healthcare going forward.
Nitika: Got it. So, I’m glad we’ve given you another perspective to think about. Thank you so much for joining us. Just to summarize, you said good ground truth annotated from data, diversity of data sets, understanding workflows and care pathways in the case of the healthcare space, and obviously, also understanding the value of regulation and the regulations around AI, when we are looking at the healthcare value chain. Thank you once again for your time, and thank you to our audience for tuning in.
Amit: Thank you.