AI for Eye Care: A fireside chat on AI in clinical workflows

What this episode is about

Dr. Kim, Chief Medical Officer at Aravind Eye Care System, discusses the integration of AI in clinical workflows, particularly for diabetic retinopathy screening, highlighting the system's ability to provide world-class eye care to over 3,500 patients daily at a fraction of traditional costs. He emphasizes overcoming barriers of awareness, accessibility, and affordability through community outreach and innovative technologies, including the deployment of the AI-based ARDA system across multiple locations.

Key takeaways

• AI can democratize healthcare; Aravind screens 3,500 patients daily at a fraction of typical costs.
• Navigating awareness, accessibility, and affordability creates a healthcare ecosystem that respects patient dignity.
• AI boosts diabetic retinopathy screenings by 800%—it’s not just tech; it’s transformative healthcare.
• Rapid AI deployment is possible—lessons from ARDA show technology can quickly improve patient outcomes.
• Hindsight reveals: focusing on tech isn’t enough; the environment for adoption is crucial for success.

Who speaks in this episode?

• Dr. Kim (Guest) · 15 min speaking
• Yossi (Host) · 6 min speaking

What is discussed, and when

1. 0:00 – 5:00: diabetic retinopathy, Andy Kim, Artificial Intelligence, Aravind Eye Care System
2. 5:00 – 10:00: Google LLC, Andy Kim, Aravind Eye Care System, diabetic retinopathy
3. 10:00 – 15:00: diabetic retinopathy
4. 15:00 – 20:00: glaucoma
5. 20:00 – 21:38: Andy Kim

Notable quotes from "AI for Eye Care: A fireside chat on AI in clinical workflows"

• Dr. Kim is the Chief Medical Officer and Director of IT and Systems at Aravind Eye Care System.

  about Andy Kim · at 0:23

• For having me here today and where we are in this whole AI process, I really thank the entire team at Google for helping us to help people at Madurai and many other places where we are deploying AI.

  about Madurai · at 1:43

• Last week, we reached an important milestone of screening half a million patients with ARDA in India, which is quite a milestone.

  about ARDA · at 3:28

• I could give you an example on how before we deployed AI, the number of patients that we screened for diabetic retinopathy almost went up eight times when I looked at the data for one year versus the 2018 data when we deployed ARDA in 2019,.

  about diabetic retinopathy · at 4:17

• Now, Google and Aravind have had a longstanding partnership, and we are honored to have done this work with you, and obviously to have positive impact on patient care over the last decade.

  about Google LLC · at 5:50

• Is what we deploy at Aravind, and one of them is AI, and then thankfully, we started with ARDA, which has been a very successful process, and we are looking at so many other aspects of patient care.

  about Aravind Eye Care System · at 9:17

Who and what is mentioned in this episode?

People

• Andy Kim — Person · Main guest / guest of honor · 4 mentions · first at 0:23
  Central speaker describing Aravind’s AI and teleophthalmology work, including ARDA deployment and broader clinical AI strategy.

Companies & orgs

• Aravind Eye Care System — Company · Business focus · 2 mentions · first at 0:27
  Central organization discussed for its innovative use of AI in eye care, providing affordable and high-quality services.
• Google LLC — Company · Partner · 4 mentions · first at 1:39
  Key collaborator with Aravind Eye Care System in developing and deploying AI solutions for eye care.

Products & tech

• ARDA — Product · Case study · 1 mention · first at 3:31
  AI-based product developed for diabetic retinopathy screening, highlighted for its impact and deployment scale.

Concepts & topics

• Artificial Intelligence — Concept · Core topic · 2 mentions · first at 0:40
  Primary technology discussed for its application in healthcare, specifically in eye disease screening and diagnosis.
  Anything that helps us to accelerate that, anything that works for that, is what we deploy at Aravind, and one of them is AI, and then thankfully, we started with ARDA,.
• diabetic retinopathy — Concept · Key argument · 5 mentions · first at 0:40
  Primary disease targeted by the AI screening system ARDA, demonstrating significant improvements in detection rates.
• glaucoma — Concept · Related topic · 2 mentions · first at 9:47
  One of the eye diseases mentioned as a potential target for future AI-based screening tools.
  In diabetic retinopathy, there are so many conditions like glaucoma, age-related macular degeneration, and quite a few stuff.
• age-related macular degeneration — Concept · Related topic · 1 mention · first at 9:47
  Another eye disease identified for potential AI-based screening, expanding the scope of AI applications in eye care.
  In diabetic retinopathy, there are so many conditions like glaucoma, age-related macular degeneration, and quite a few stuff.
• corneal ulcers — Concept · Related topic · 1 mention · first at 9:53
  Highlighted as a major problem in India, with potential for AI-based solutions to improve diagnosis and treatment.
  In India, the corneal ulcers is a major problem.

Places & events

• Madurai — Place · Headquarters · 1 mention · first at 1:50
  Location where Aravind Eye Care System is deploying AI solutions, mentioned as a key site for their operations.

Full transcript with timestamps

0:22 - So, a few words of introduction. Dr. Kim is the Chief Medical Officer and Director of IT and Systems at Aravind Eye Care System. He pioneer teleophthalmology in India and developed an EMR system used in many eye hospitals. He has been one of our very close collaborators for the past decade for the ARDA project and had envisioned the AI-based automated diabetic retinopathy screening system. He was recognized by WIRED25 list of people who will change the future for the groundbreaking work in artificial intelligence and diabetic retinopathy, with nomination from Sundar.

0:58 He continues to focus on AI for diabetic retinopathy, screening and deployed ARDA at over a hundred sites in the Aravind ecosystem, so let's give a huge round of applause to our guest of honor, Dr. Kim. (audience applauds) - Thank you. - You know, I heard that staggering number of 3,500+ of patients come through the doors of Aravind Eye Hospital every day and get world-class care, and a fraction of the price that otherwise they would pay. So, how do you do that, and perhaps you can say a few things about Aravind Eye Care Center.

1:37 - Thank you, Yossi, and at the outset, I just want to thank everyone at Google for having me here today and where we are in this whole AI process, I really thank the entire team at Google for helping us to help people at Madurai and many other places where we are deploying AI. Well, to your question about so many patients coming in, it's generally, you know, any hospital tends to treat people who walk into their doors. Basically, if you look at the barriers to healthcare, there are three barriers. One is the awareness barrier, accessibility barrier, and the affordability barrier.

2:26 Now, this is something that I guess we recognized over a period of time, and then we started addressing each one of them. As you know, that we do the lot of awareness, accessibility by going into the community, what we call as outreach into the community, and making it affordable for people. Like, people who can afford to pay, we let them pay, and then those who cannot pay, we don't charge them anything, and then there's, we also want to maintain the dignity of the person, so it's they can walk into a free hospital or the paying hospital where the care is the same.

3:04 So, that's something I think over a period of time has helped and people recognized that. Since we are addressing all these barriers, that's one of these reason why we see more patients at Aravind. - The picture we see here is from the 2018 when the first clinical patient was screened with ARDA, so it's been quite a journey. Last week, we reached an important milestone of screening half a million patients with ARDA in India, which is quite a milestone. And by the way, we have here among the audience, some of the members of the ARDA team.

3:41 This was only possible because of the support from partners, you know, like yourself, and we really learned so much from the experience. Now, when you reflect back on the journey, and you were there not only from the beginning. In fact posing the questions that actually helped kind of start this effort, what surprised you the most? What are the learnings that you observed from this journey? - It's just the sheer magnitude of how much you can do with AI. That's the learning, I think. I could give you an example on how before we deployed AI, the number of patients that we screened for diabetic retinopathy almost went up eight times when I looked at the data for one year versus the 2018 data when we deployed ARDA in 2019, and in 2022, I took the data because we didn't want to use the COVID period.

4:43 When we were in full scale, the number of patients that we had identified having diabetic retinopathy went up by eight times that number. We were doing the screening earlier on too, but just the sheer number, how we can scale up with just use of technology amazes me, and so I think that that's a big potential, that you can deploy in places that really need these kind of- - Yeah, one thing that I observed, by the way, is that as we learn how to deploy technology, the pace is increasing. So, we have a lot of learnings.

5:17 In fact, we published a paper about some of the learnings of the ARDA deployment, and now when I think about deployment of AI systems into AI results, research results into systems, it's actually much faster than it used to be, so again, this was, a lot of those learnings actually are from these efforts, which was so instrumental to understand about how to take scientific work that makes a difference into the clinical world with all the learnings of how to deal with the data and the quality, and all the other things that we need.

5:50 Now, Google and Aravind have had a longstanding partnership, and we are honored to have done this work with you, and obviously to have positive impact on patient care over the last decade. What is it, what would you wish to have done differently in the journey, 'cause obviously in such a long journey, there are some learnings and perhaps wrong turns. (Dr. Kim laughs) What would we have done differently that would make it better? - You know, thinking back, it's different. When the technology started at that time, we were very focused on, 100% on the technology, getting that there, getting the technology, but what we needed to also look at is the environment that we need to build the technology to uptake this technology into the system.

6:38 This is on the hindsight, we can say this, but at that time, the technology was really amazing, because I personally was working, trying to get this AI since I heard this about, I'm a clinician, I'm a retina specialist, but technology always amazes me. 2006, I was working with many of the engineering institutions when people talked about AI imaging, image recognition and stuff. So, everybody would come build an AI, so-called AI tool, which will recognize part of it, like a hemorrhage, a red spot. It'll recognize an yellow spot.

7:16 They would write a paper on it, finish their thesis. They move on, and so that's when, when Google came in and we were introduced to Google and we both had the same purpose of bringing out a tool that can really help people, I think that is really, really was something really amazing when Google came, and it really was a change maker for us so we were able to reach more people. And this is just a beginning, and there's so many things that are going to come. So, I think it's been really a wonderful, wonderful journey for us and learning experience also.

8:00 - One of the reasons that make me so excited about this effort is this notion that taking this deep technology and deep science and then looking what needs to be done in order to actually get it to right impact by actually making it work end to end, taking into consideration everything around it, and then the potential of doing that more, of course, is quite great. Well, talking of each, well, Aravind Eye Care System is known for its patient-centric vision, mission, providing high quality, affordable care for all.

8:33 Can you share your thoughts on how you see AI playing a role in this and helping accelerate the Aravind mission, not only with this problem, which we are already on good track, but actually beyond that, and also, what do you see as the potential of AI more broadly, where you are in, being a pioneer on this notion of using technology to have impact at scale? - Yeah, good that you asked this question, because one of our, we are a purpose-driven organization. So is Google. You know, our purpose is to eliminate needless blindness.

9:11 Anything that helps us to accelerate that, anything that works for that, is what we deploy at Aravind, and one of them is AI, and then thankfully, we started with ARDA, which has been a very successful process, and we are looking at so many other aspects of patient care. One is accessibility. The other one is the quality of care that can improve with the using AI. And the number of diseases that you can identify for screening. Like, in diabetic retinopathy, there are so many conditions like glaucoma, age-related macular degeneration, and quite a few stuff.

9:53 In India, the corneal ulcers is a major problem. So, these are some of the things we are currently trying to work with to develop AI-based tools to identify, to screen these diseases. Beyond that, improving the quality of care, and make sure that there is a lot of safety inbuilt in how you use AI in patient care, meaning when a doctor is examining the patient, is AI able to help him to not make a wrong entry into the EMR? So, his observation is getting into the EMR. If it's a wrong entry, the AI should be able to alert the doctor that, you know, what you're doing is inconsistent to the patients there, and that's how I see AI can go forward, and more importantly, it can also personalize treatment.

10:48 Today, I know if I see a patient with diabetic retinopathy, I have to either give them an injection in the eye or treat them with lasers. Is that something that is acceptable? Is all the patients same? So, all these will come out I'm sure with using AI, understanding the patient. The patient. Not any patient; the patient. And trying to treat them, personalize the treatment. I think AI will go a long way in that, and I feel those are some things that AI will definitely play a major role. - So, what I'm hearing is obviously empowering the healthcare workers, helping for example, reduce some of the errors that are inevitable, obviously, in the healthcare system, which obviously translates into a lot of what's happening, and obviously with now with generative AI, so for example on the, you know, work such as MedLM, Med-PaLM Med-Gemini MedLM obviously the opportunity to identify that, but you also touched on this notion

11:52 of personalized medicine, which, of course, AI can be very helpful. Talking about personal experience, you know, one that I cherish is a year and a half ago, actually visiting in clinics in Bangkok actually. Can see the picture here. My biggest surprise was actually there was no surprise, in the sense that it just worked as we were hoping to see it work. Seeing a person coming, sitting in front of the camera, in two minutes actually getting the diagnostics that I was told is 20% better than what they used to get because there weren't enough eye experts to actually examine that, so it was actually typically examined by generalists who would not have the same proficiency as eye experts.

12:36 And then the fact, one thing that also I didn't realize at the time was that the fact that people were getting the results right on the spot made them actually more likely to actually follow through. Where the statistics are heard is that if somebody's getting the result five years, five weeks after the exam, then only one in four or five would actually follow through, so these are all kind of, were very encouraging things to see, so I find it very inspiring to see how quickly AI is helpful to so many people, so thanks for initiating this kind of research.

13:11 When you think about the future, how do you envision future plans involving technology, AI models? You touched on that a little bit. What would be other goals and aspirations? Where would you envision, say in three to five years, you know, Aravind and ecosystem? - Yeah, as I mentioned, we do have a larger picture in front of us to use AI in the betterment of patient care. One of the things that currently we are trying to see is we have a vision center model where we have primary eye care clinics which are run by ophthalmic technicians, and that they see the patient, do a complete eye examination, just like a ophthalmologist would do.

13:57 They will do a slit lamp examination, do a retinal examination, take pictures, and enter the entire data on an electronic medical record, and they are connected to a doctor sitting in the base hospital, one of the ophthalmologists. They are managing, that ophthalmologist will be managing four or five clinics like this. So, the whole day, what they're doing is they're simultaneously as the patient is there, they are able to see the patient data, look at the images, and make a diagnosis or whatever that they need to do, and then they talk to the patient over video.

14:35 That was very important, because the patient knows that they are talking to the doctor, and the doctor has seen all this data and that's how we have been treating these patients in the vision center. This has been happening since 2004. We have deployed and we have almost 113 vision centers just in the state of Tamil Nadu where there are no ophthalmologists. Like, for every 50,000 population, we have put one vision center. Now, these vision centers we feel will go a long way in using AI, because the doctor's time is very limited, so if they can ensure that these AI-based devices can triage the patients, identify those who are having more problems versus, who really need a doctor's evaluation, they are the ones who are kept for the doctors, and the rest, the system will take care.

15:31 This is something we want to do it at the primary care level. And of course, I mentioned about screening for other conditions like glaucoma, many other diseases. In the rural areas or primary care settings, it's something that we want. But at the tertiary level, we feel it's to go a long way in improving the quality of care in personalizing treatment for the patients, summarizing patient visits, analyzing the images that we have, hundreds of images taken for the patient over a period of time. Would that give new information for us?

16:08 These are some things that we think bigger picture that we have for using AI in Aravind. - One thing that I'm really intrigued about is the potential connectivity. Once you connect people with telehealth or the opportunity for experts to review material from far away, and not only for conversation, but also to look at the content, if you bring that together with AI, then it opens up all sorts of other opportunities, because then, you can use AI for prioritization, for triaging, for helping to make sure that the right information gets to the right experts, and do that in a global scale, and then actually provide the kind of opportunity of access regardless of where people are, so I think this actually is very intriguing.

16:57 And another aspect, you mentioned this notion of looking into all these material. Images, I guess videos and others. With the new technology of generative AI that can also be multimodal, obviously this opens up some pretty exciting opportunities for the future to see how can we use actually AI for analysis of data and as minimum to help out, identify, and help and be a greater help to the healthcare workers. What would be your advice to researchers we have here, some of which have worked on ARDA in the past, others are working on other areas of health or otherwise.

17:34 Researchers are looking to have the same level of patient impact as ARDA in general. Any advice you would have based on your interaction? - Technology is, like everybody, every day, we are seeing newer technologies coming in, and then many a times, we try to find use cases for technology, which is when many of the technologies kind of fail in that process. What is important is to understand, look for the use cases that people have. What are the challenges that the doctors have or whatever processes, and then try to fit in the technology or the use of the technology becomes much better, and then you see the translation from research to uptake is really high.

18:19 - It's always a good advice to ask what would be the impact of what I'd like to do if it would be successful, which in this case was, you know, solving the problem of screening. On the flip side, of course the other side of the equation, is the healthcare providers, and you were a pioneer in identifying technology, and fortunately, we do see that we're beyond an inflection point where healthcare providers are more eager to use technology for that, but still, many are just learning of how to do that. What would be your advice to healthcare providers who want to incorporate AI into their clinical settings?

19:00 - Yeah, so there is still a big divide in this between those who are the believers of technology, and there was some people who say, "No, that's not the way to go." We are seeing the positive impact of it. There are people who are seeing the negative impact of this. So, that barrier will continue to be there for some time until the larger groups take over and start using AI. People will be very much hesitant to use technology for the simple reason, one, anytime I present something about the AI, the immediate question is, "Am I at the risk of losing my job?"

19:45 They don't understand the impact, that it can bring in more patients to them and more meaningful patients to them, rather than screening the patients and identifying those who have serious problem and treating them. It is going to take some time for people to believe in technology. It's not like replacing doctors. They have to understand, it's only aiding doctors in making correct decisions and supporting them to do what they can do the best. - Certainly something that we strongly believe in that AI has a role to play in helping out the healthcare systems to the extent possible, and one of the important questions is to identify where it can be helpful at best.

20:28 And any closing thoughts on your part? - I just want to say thank you to this entire wonderful team for how you have supported the journey at Aravind for developing this AI, and you continue to be, we still continue to do some work with each other. I think, I just want to say thank you. - Well, I'd like to thank you for your partnership, and actually, it's not often that we have a partner that has been not only inspiring a project and a partner to the research, and then partner to deployment, and then having impact all the way on the clinical stage, so I think it's a very special partnership that I can only thank you, and also take the opportunity to thank the team that has been working on that over the years on everything, from the original research to deployment, and up to today.

21:22 So, thank you all, and thanks again for everybody who joining. Thank you, Dr. Kim. - Thank you. Thank you, sir. Thank you. (audience applauds)

Questions about this episode

Who speaks in "AI for Eye Care: A fireside chat on AI in clinical workflows"?

Dr. Kim (Guest), Yossi (Host).

What is "AI for Eye Care: A fireside chat on AI in clinical workflows" about?

Dr. Kim, Chief Medical Officer at Aravind Eye Care System, discusses the integration of AI in clinical workflows, particularly for diabetic retinopathy screening, highlighting the system's ability to provide world-class eye care to over 3,500 patients daily at a fraction of traditional costs. He emphasizes overcoming barriers of awareness, accessibility, and affordability through community…

What is mentioned in this Google Research episode?

10 people, works and ideas are mentioned, including Andy Kim, Aravind Eye Care System, Artificial Intelligence, Google LLC, ARDA, diabetic retinopathy. Each one on this page links to its own page with every episode that mentions it.

How long is "AI for Eye Care: A fireside chat on AI in clinical workflows"?

The episode runs 22 min; the transcript is about 3,295 words in 30 timestamped sections.

More from Google Research

• All analyzed episodes
• People mentioned on Google Research
• Topics mentioned on Google Research