← Power Hour Optometry
Power Hour Optometry
Artificial Intelligence in Eye Care: What's Real, What's Overhyped & How to Evaluate It | Eugene Shatsman & Abed Sarhan
01:01:48
Loading summary
Transcript82 lines
- [00:00]
A
Foreign. Welcome to the Power Hour, Optometry's biggest and longest running show. I'm your host, Eugene Shotsman, and today's episode is an interesting take on a topic. You know, I love AI in eye care. My guest today is Dr. Abed Saron. And while he has a deep tactical background in AI and deep learning, by the way, from deep over a decade before AI became something that we all talk about all the time, what I found most interesting about this conversation is that it's a practical conversation about what AI is and is not and is ready to do in clinics today and what's being oversold and really how doctors should be evaluating these tools for themselves. And Abed gives some really interesting criteria there. And I was taking notes as we were, as we were talking, because his perspective as a PhD is really something I haven't heard a lot of. And we get into things like where can AI genuinely improve workflow, workflows, patient experience? We talk about things like voice agents, automation, data integration, parts of the pre testing and education process. But we also spend a lot of time on where people should be a little bit more skeptical when it comes to black box diagnostics, for example, or bold claims that salespeople use to sound impressive that may not be necessarily realistic under some scrutiny. We also talk kind of philosophically about what's naturally human and what's going to be kind of the clinic of the future and what AI is and is not going to do for, for patients and for practices in the future. In terms of really just thoughtful discussion about where technology can help, where human relationships still matter most, and the kinds of questions that practice owners should be asking before they invest in anything AI related. So before we get into the episode, quick reminder, make sure that you're subscribed on your favorite platform, YouTube, Spotify, Apple Podcasts, or wherever you get your shows. Just so you know, every time a new episode drops, as always, if you've got feedback, questions, ideas for future episodes, or just need help with anything, reach out to me directly@eugene shotsman.com I love hearing from our audiences. You can reach me there or at the Power Hour website. Really do enjoy hearing from you. And now let's get into today's Power Hour. Abed, welcome to the Power Hour. Excited to have you on the show.
- [02:30]
B
Yeah, thank you for that invitation, Eugene.
- [02:33]
A
So, you know, I think it might make sense to. The question I've been wondering is you have a PhD in the field of deep learning, and I'm curious, what does that even mean? How does that Relate to what's going on today?
- [02:49]
B
Yeah, that's a great question. You know, in the field of AI, AI becomes like very more like a black book these days where there are so many things happening, but when you dive deep into the AI field, there are like multiple sectors in it, multiple fields in it. My specialization is when, when I mentioned deep learning is basically when we're talking about hundreds of thousands of like data inputs when it comes to analyzing flows, when it comes to like working with like deep learning models that goes over the images, integrate or analyze each pixel with their four dimensions and basically use that toward like segmentation, classification, all those different aspects of it. So the word deep learning is basically when you're going beyond traditional approaches in AI and going to the more sophisticated data modules that analyze big amounts of data.
- [03:48]
A
Yeah, so it's really how machines learn. But also is it, is there a difference between deep learning and machine learning? Is what's the nuance?
- [04:00]
B
Yeah, so machine learning is also like an umbrella. So machine learning can learn things, but there are things where. Let's take an example. If you're trying to build a model and you feed it with 10 variables, based on those 10 different variables a model will output if this person have a condition or not. Like for example, support vector machines, which is like a type of machine learning model, deep learning goes way ahead. Deep learning, we don't usually talk about 10, 15, 20, we're talking about 10,000 of variables that are actually fed into a series of networks that get filtered from one stage to another based on something called lost functions, or how the basically the activation methods done for those layers and then give you an output. So it's a more complex problem we are trying to solve using large amount of variables.
- [04:56]
A
So, so when you get a PhD in that, like what is your research? Like, are you basically building these networks? Are you? Yeah.
- [05:06]
B
So basically what I've done in my PhD is that I didn't actually just take off the shelf model. I actually went into the, off into basically the deep layers of building those layers of network on the C level, on the low programmed level to do that. And I also have to manipulate something called the activation method and loss functions to build like things on my own and validate those. When you're talking about general machine like svm, you're not actually doing that. You're just taking like a library and putting it, feeding the data and then basically driving. My specialization is basically on the engineering level of building those networks.
- [05:50]
A
Okay. So I think it's fair to say that you were in AI before AI became the popular thing that it is today, right?
- [06:01]
B
Yeah, before it became like so much like famous. I would say AI has been there for like decades, but before getting into the waves of it now. Yeah, we used to like, we used to work data mining almost different terminologies for like different aspects of building modules.
- [06:19]
A
Yeah. So the, I guess the question I have for you is that, you know, what point did you become convinced that AI, and specifically in eye care, because I know you have a glaucoma specialization was going to matter in a practical way and not just academically. At what point did you realize oh man, we're actually going to build something that people are going to use?
- [06:44]
B
I came more convinced about that with the advancements on the TPU level. So the thing when, when you're talking about deep learning, you're talking about big amount of data that needs to be analyzed. The way that our things structured in the vision space or like in, in the medical field in general, you have so much data but not deeply analyzed yet. The traditional approach for machine learning algorithms, they, they weren't ready to do that. They are like very basic. They don't, they can't really handle this big amount of variables. With, with the advancement on the TPU level, it actually opened a high, a big door to be able to be innovative on the modules and also to build your own models and training and build upon previous data sets to basically build a high precision module based on segmentation. That's basically what the tipping point on to know like oh, now we're entering an era where confidentially is feasible. Data is getting, people are getting more and more accepted, accepted for sharing data or like integrating it with their flow. So it becomes more and more apparent there.
- [08:01]
A
Yeah. And what chronologically, like what year was that that you became convinced that man, this might actually be super practical one day?
- [08:12]
B
Like in 2015?
- [08:14]
A
I would say so 11 years ago you knew that AI was going to be a big thing in healthcare.
- [08:21]
B
Yeah, like I do remember my master's was actually on LLMs. Like my master's was how I can ask a question freely and then capture the data from a relational database. It wasn't like unstructured document or structured document and capture that. And then now we have like chat GPT, we have OpenAI, we have Cloud, we have all those different platforms everyone using those days, but it's more on a general, on unstructured data sets or databases develop. Yeah.
- [08:52]
A
So do you think, I mean what, what's your opinion on where we are today 11 years later after that realization, I mean, how are we? Have we exceeded what you expected to be happening? Or have we, or have we, or have we not reached the potential that you expect us to be able to reach?
- [09:13]
B
I think 11 years ago people were promising over promising what AI and what things can be delivered. I don't think we hit that. I think there's so much hype about AI. We are still in the middle of like, what's gonna be a good evaluation, what's gonna be a good module to be used in the clinic. And if you check the research out there, there's no one structured way. Even, even those that are translated in the industry, the way things are being evaluated, the way things are being structured or trained, it's not sufficient enough in my opinion. Still, there's still big gaps and that could be a reason why they are not being still widely used or people are not very convinced yet with some of those models.
- [10:00]
A
Well, yeah, so maybe you can explain or elaborate on that just a little bit, like what's missing or maybe what structure is it a unified approach or. And maybe you can even elaborate further to just kind of describe what you think most people misunderstand about AI in healthcare today.
- [10:20]
B
I think many people miss asking the right question for the right tool. So when you're trying to evaluate a voice, let's assume you're working on a staff trying to basically have a voice agent. The way that you evaluate a voice agent is different than how you evaluate an OCT machine learning module is different than evaluating low karma classification model. And the problem that you see in the field is people use words like accuracy, mean average, error, auc, for instance, area under curve. Those are good metrics, but they are not specialized metrics. That tells you if a model works properly or not. I'll give you an example. Let's assume we're building a model for segmenting disk and cup from a fundus image. You know, for a disk is larger than a car. And the way that you segment the image, if you say I can segment a disk with 98% accuracy, well, that doesn't mean, if you mean your model is actually working well, because if you're actually taking a region of interest of that disk and that disk occupies 98% of your image, you're already like, doesn't really reflect how your model is performing. What you, what actually reflect is if you measure the distance between the center of the disc and you try for instance, showing beams and measure the difference between the edge of the segmented disk versus the ground tool. And if you check the papers out there, many of them miss that type of evaluation. And this evaluation is very important because it can tell you if that segmentation is detecting the borders properly, if it is differentiating between a disk and the ppa, for instance, around the disk and tells you, like, what's what, the precision of that, of that type of segmentation. The same with voice agents. For instance, people report mean word of error or 4%. But have you tested that voice agent in stress? Different accents, different environments, different keywords, different emotional situations. Those are missing when we're trying to read the research behind many of those tools out there.
- [12:36]
A
I see. So you're saying there's levels of permutation. So, you know, and even, even simply speaking, like if you have a voice agent.
- [12:44]
B
Yeah.
- [12:45]
A
One of the levels of, I guess, analysis that's missing is, you know, can it detect emotional.
- [12:52]
B
The.
- [12:52]
A
The emotional context that humans would naturally be able to detect. Because if you ask me, would you like to schedule an appointment? And I say yes, that's a different statement than yes, exactly. And that, you know, I don't know. Now we have a product that's a voice agent. And now I'm thinking is, did we really think about that? You know, is it. Is it possible that. And I guess, you know, what, what model needs to be built to do that? And practically speaking, how, how different is the outcome going to be?
- [13:30]
B
Absolutely. And I think there are ways to do that. I think there are ways to detect if the person is saying like, like, you know, there are different variables or different indicators that can be collected. But it feels those are usually I missed out or like not focused on when they are. I, in my opinion, are those are what makes a product very niche and make it outstand and perform well on clinics. This applies for every aspect in the vision space going on, even like in other spaces. But we're focusing on the vision space because I, I'm seeing a lot of hype, but I am, I feel it's missing the proper validation and proper training. I'm not saying all of them. No, for sure, but I'm seeing that needs to be addressed properly in studies to build the confidence with doctors and clinics.
- [14:25]
A
So when you look at eye care today, and if we kind of zoom in on that, what do you see as the biggest gap between what technology could do and what it's actually doing today? Because I think there's been so many advances that are hyped right now, and many of them are very practical and very helpful. And you know, they're early to market but you know, is still. But it is still something that's practical. But then there's also like the what can we actually see it possibly doing in the future? And I don't know sometimes what's hype when you ask that question and what's realistic? And it sounds like you've got a pretty realistic view.
- [15:09]
B
Yeah, no, absolutely. And what I like about the vision space, it feels good at the crisp at the cusp. Basically people are open for change, people want a change, it connects. But it feels like they're kind of in the middle. Okay, who we want to trust and what we want to use. I would say my rule usually is yes, clinics, some of the tools may be a little bit immature, which is fine. You deploy and you start experimenting. Although as long as you're not like overselling or like over promising with stuff. But a tool that can have the speed and the resources to keep evolving as the field is evolving is the tool that can really win in connect mistakes. Because when we're talking about for instance, fundus images, one of them missing stuff. I feel like it goes back to the topic you mentioned before. How you're dealing with high precision for segmentation, how you're providing the explanation that the doctor needs, which nowadays it feels like a black box. That's okay, this is a glaucoma, this is the likelihood of having diabetic retinopathy. But how did you come up with that decision? How a doctor will be able to see what are the variables used, which is a big gap missing now nowadays in the clinic. And that goes back to the segmentation object. Because if you want to detect glaucoma, for instance, you want to check the cup, the disc, the RNFL area, all those different type of things that should have a high segmentation precision to be able to show it to the doctor and show the measurements, how they are being done. So the gap in the field now is not. It's about how you provide explanation to the doctor about specific decision that's been made by the AI. That's if we're talking about diagnostics, if we're talking about voice and other ones like for instance, voice agents, it's about just publishing those metrics and be transparent about them and just say, okay, this tool is evolving, this application is evolving. And people, people are more open for that cause they know, okay, this is the limitation, this is what we know, this is what we don't know. And the tool is a whole loop, for instance, over time so that's, that's what I feel the big gap that's happening nowadays.
- [17:35]
A
Got it. And you know, when you think about, based off of what you know from an eye care standpoint, what, what parts of the eye care patient experience and maybe even the clinicians experience are most likely to be truly impacted or transformed by AI in the next couple years, do you think?
- [17:58]
B
I would say treatment selection and examinations. The way that exams are being done in the prescreening lane in clinics, it is very dependent on staff and it's very prone for friction sometimes because staff are under a lot of pressure. They see so many patients, patients coming stressed, they don't really know what to expect. And there's always that, that like oh, how I can explain how I can adapt to that specific exam. So examinations, it's one factor that I think can transform how, how eye care is being delivered and the treatment being done. Because in many situations doctors analyze or doctor check the data. But at the end of the day, if the patient doesn't select the treatment, their situation is going to worse and treatment selection is going to be depend how you explain to them and how, how they are being convinced about that treatment. So they don't take it to stay on the approach rather than this is scientific approach that you really need to use that treatment and if you don't use that treatment, this how your vision look like and two years, three years, four years. So those are the two areas where I feel can I really impact the patient journey and clin.
- [19:12]
A
So, so do you think that the treatment explanation will be done by AI or how will AI be used in the treatment explanation?
- [19:20]
B
I do think we're heading to that, to that direction. I do think because data is there, patients have their data and clinics. AI will come into a place where it's gonna show how a treatment can impact the patient over the long term. And that's gonna be integrated with medical. I would say could be like eye drops, could be like with surgeries, could be like with like glasses, like different aspect of the treatment that usually dry eye, all those different type of treatment that patients sometimes needs to go through. I do believe AI will be able to reach a place where it can assimilate or going to show to the patient how a treatment can impact their, their journey.
- [20:08]
A
Now on one end I'm going to, I'm going to kind of look at both sides of this thing. So on one end I imagine a world a what, what you're saying to me is what I'm. When you're talking about it. What I'm, what I'm envisioning is there's a mechanism. Let's call it a headset, let's call it a machine. Let's call it what? Let's call it a computer, let's call it an AI phone, device, whatever. There's something that's going to talk to the patient and it's going to know, based off of the diagnostic procedures that have been completed, it's going to know that patients like you are experiencing symptoms like this. And it's even going to know the treatment acceptance rate across thousands and thousands and thousands of other cases. Yes, if it says this or if it says that or if it says, and if the goal is to get the what's good for the patient, which is treatment acceptance, then it's going to know the best way to talk about a particular disease state. It's going to know how to talk about dry eye or glaucoma or cataracts in a way that gets the patient to pick the premium lens or the most, or the most sophisticated form of treatment or whatever. It's going to have all of that, we'll call it information from dozens or hundreds of practices that it's got the, it's got the level of intelligence to be able to explain this in a way that makes sense for a patient. So that the patient says, yes, that's world number one, but I see world number two as okay. But then we're natural, we're taking out the inherently human part of the doctor patient relationship. Because the doc, the reason I, and maybe I'm silly for assuming this, but the reason that I take action on a treatment is because I trust my doctor and I believe that that doctor is going to be the one who provides the solution.
- [21:59]
B
Absolutely. And when you're talking about treatment, we're not talking about the doctors that being involved in that selective treatment, we're talking about the doctor is validating the last level of a treatment. But the patient will be able to simulate it to remove any bias, will be able to see how their journey would look like if they don't take that treatment. Because in many days patients come to the clinic, the doctor prescribed an eye job for answer for that patient, patient doesn't always take it seriously or doesn't always commit to that. So how you want to convince or how you want to show the patient that this is the magnitude of how things can be worse, for instance, over time if medication is not taken properly through that. So the doctor will be able to decide what type of treatment or validate what type of treatment this patient will go through and we'll have a talk conversation on that relationship with them. But the patient will have the capabilities of seeing that in real life or virtual or be able to simulate that or like see it in a way so they could experience how serious this is.
- [23:03]
A
I would say I, I like your idea of the treatment non treatment simulator. It is also like, it's kind of, you know, I guess something like dry eye. It's, you know, the simulation is going to be like, you know, like you're really itchy. Like, I don't know, like what are you going to show somebody to be? Like your eyes are just really itchy and you're really uncomfortable. I understand, you know, simulating like, I don't know, glaucoma or AMD or something like that. You, you, you will you look at the, you, you look at someone's vision and you. Even with cataracts, it's easy to kind of simulate and say, if untreated, this is what your life is going to be or listen to what your field of vision is going to be or whatever. I, I don't know if you can really do that with, with something like dry eye, but maybe you can and
- [23:54]
B
I don't know, you could show them how you don't take or take care of the eye treatment, how their itchiness can ideally impact their daily operations over time and how those can influence how that things would look like because that itchy will lead to people getting upset, annoying, frustrated. So you could basically develop something to provide that type of education for the patient to be able to realize how important this can be for them and that education will make an influence or
- [24:32]
A
like, okay, very, very interesting. When we come back on the show, we're going to take a quick break, but when we come back on the show and I try to get practical and we're going to start talking about, like, if I'm evaluating AI solutions, what the heck should I be doing and how do I determine if I should take the risk or not take the risk. And we'll talk about what's keeping people from doing it and what's getting people excited and then what's frustrating people. So we'll be right back on the Power Hour and continue the conversation. Hey there, it's Eugene and I want to let you in on something. So you've been to conferences before. You come home fired up and then Monday morning hits and it's back to the grind. The ideas don't stick, the Plan never gets made and six months later your practice is in the same place. So I know that pain. I've been to those conferences with you and that is not happening. At this new event called icare Boss Live. You've heard the story of icare Boss and now there's an event, ICARE Boss Live. It's September 16th through 18th in Cleveland. Two and a half days. We're bringing together 200 of the best practice owners in Icare for a one of a kind event that combines speakers, peer learning, mastermind groups and industry innovation, all designed around one goal. You leave with a 90 day plan and you can actually execute it and get stuff done. And we're going to tackle some real stuff. Exam only rates, revenue per patient, people problems, leadership, AI and technology, specialty growth, the things that keep people up at night. We're going after it and we're doing it in a room full of practice owners that are just as serious about growth as you are. This is not a conference, it's not a seminar. It's something different. There are only 200 spots. So if you want to be in on this, this is not publicly announced, just on this podcast. Go to thepowerpractice.com, click events. Click apply now. This is invite only. It's not for everybody. So you have to apply. We'll ask you a few questions and if it's a fit, we'll invite you to register this event. I care Boss Live is gonna sell out. Do not sit on it. I invite you to apply. Right now, we're back in the power hour with Abed Saron and I. I guess maybe this is where I'll start our, the, the second segment of our show. Do you think there are technologies right now that every eye care practice should be using?
- [26:57]
B
Yeah, I do think so. I do think we're going into a place where there is technologies out there that I think will make significant impact on the clinical flow and will really help clinicians, staff and patients. The challenge is trying to select the right, the right one. I would say it's going to be okay.
- [27:18]
A
So in your opinion, and you have to say specific brands or anything like that, but like, what are, what are the solutions that weren't available a couple years ago that people should really be looking at right now that could potentially help them a lot, both either in practice or, or with the patient flow?
- [27:37]
B
As you were saying, I do think there are two parts of it. There is the part on the patient journey in the clinic and there's a part of like the operational Clinic I do believe voice agents can really like offload most of the pressure. And clinics that I think can really
- [27:56]
A
help
- [27:58]
B
just selecting the right one for those and trying to have one at triage it, I would say for clinic that's going to be the key. But integrating voice agents from now gonna really help you in the future because it's gonna just keep getting better for that now.
- [28:14]
A
Voice agents in what context?
- [28:15]
B
Haben voice agents. It means when patient calls in, they want update about their medication or they want update about their appointments. I like those type of things. They don't really need always to fall on the staff or the front of the desk staff because those are things that could be integrated into your EMR or clinical flow and just move things in less friction.
- [28:43]
A
Yeah, so completely agree. And I think the first, first and foremost, I always talk about how 26% of. And it got just a little bit worse in 2025. Actually 26% of calls that come into practices tend to be unanswered during business hours, which is a shocking, shocking number. And many of those are new patient opportunities and many of those are people looking for appointments and many of those are people who are having less than adequate customer service. And so, you know, I have always talked about prioritizing. If somebody needs an appointment, at least serve them now because they're going to go somewhere else if you can't get them there. So I think those voice agents now exist. I think, you know, there's a lot of the. And the barriers to those being successful before were latency. It was uncomfortable to talk to one of those and you know, wait for the weird pause that, you know, you kind of have. So latency was one of them. I think the inability to integrate with eye care ehrs, that was another one. I think the lack of training initially for the models to say, like, what should. What are the things that patients are going to ask? How much is an eye exam? Do you take my insurance? And what are the best answers? Right. Because you shouldn't just say like $150 or no, we don't take that insurance. You really have to make it so that the patient still wants to come. Depending on what you're saying. Yes, our eye exam is if you don't have insurance, our exam is $150. But that includes the following technology and it includes a really great experience, and it includes this, that and the other thing so that it feels like there's value, not just a random number. But all of those things aside, that's what I think was Necessary as the input to a good voice agent just to schedule appointments, but to make it also more special.
- [30:41]
B
Now we're getting into a stage where voice agents can clone basically the voice of one of you. So it looks more humanized because one of the challenges that were happened before, in addition to the one that he's mentioned, it felt very robotic, very monotonic, very like. Doesn't feel that human eyes like asking of it now it feels like, oh, I'm. I'm speaking with someone who sounds like the staff. It is dangerous in other aspects. But from clinical perspective make it more like patients are more accepting the fact to be able to speak with a voice agent because it feels more natural.
- [31:18]
A
Yeah, I completely agree with you. I think that somebody who, if I'm in the southern United States, someone who has a southern dialect is way. Is way more likely to get. And I have no data on this. I'm just assuming, right. And I guess I, I'm using data that I have from call centers. When you think that call center agents need to sound like the people who are who are calling. So if you have someone who is in the north and talks fast, you want to have an agent that's, you know, closer to in the north and talks fast. If you. And so I, I get that. I think we naturally get that. Again, I don't think there's enough data that I know of yet to support that. Maybe you have some data that, that kind of. That explains that. But have voice agents been studied by the way? Is that something that's. That. That there is a lot of academia
- [32:08]
B
around, there is some research on it. The problem, the thing with what I'm seeing now in the field, that majority of them are being based on like three or four providers. Like for instance, like now ChatGPT have their own people build on their LLMs and like build up that. So I think majority of the tools out there, there's no resource to be able to support what they're saying. They are just basically like some metrics for user experience. But there are underlying providers, some research out there that shows that's getting better. And this is where what I mentioned, like some of those research papers they mentioned, oh, 4% word error rate. But this is not the only metrics that you need to report on there. There is need to be a report about how this agent work with different emotions, how how it works with under stress, how it works with people, different accent, those different variables, they may not be great, but when they are being reported, people are able to track how Those variables are improving over time. So that's going to be the key for like some doctor was selecting some of those words item. And I don't blame, blame anyone in the sense that doctors will go and see the tool or stuff will be able to see the tool but there is no clear presentation about what to do matters or what was the underlying tool that just a new tool is being built on top of it. And that's, that's where the gaps can happen. That's where trust gonna be impacted because I've seen that also clinics. Someone tried that tool, oh, it's working good. Someone tried that other tool, oh, it's not working good. And some people will not, will not trust that tool anymore. And this why it's many situation. You feel doctors rely on what their peers are using. Although not all the time. Maybe like always the right choice, but they had good experience with it and that's going to be the drive for them.
- [34:14]
A
And that's where I say, you know, if you're going to consider any tool like this, you really do need to ask your team to call it. You really do need to ask your, you, you need to call it yourself. Yet, you know, it's great that you know, Dr. Abed is using it and he's having a good experience. But like let me call it myself a couple times. Let me have my office manager call it, let me have people on my team call it. And you know, my team, the people on my team are the ones who are going to be most resistant in the beginning sometimes because they don't necessarily understand that it's an opportunity to elevate. They might be thinking this thing is coming from my job. So they're the ones who are going to criticize it and I'm going to challenge them. I always tell people like, call it and try to break it. I want you to break it. Like. And we, as I mentioned, we have a product that's just coming out of beta right now and it took longer because we specifically said to people like, hey, I want you to break it. I want you to figure out a way to like get it to go off script. I want you to figure out a way to challenge it with a question where it gives you a really stupid answer that would be super embarrassing. And at the end of the day, look, 26% of people are not even having an experience on the phone. And of the ones who are having an experience on the phone, like I'm going to guess that more than 50% and not even guess it's an educated guess because we listen to a lot of these phone calls. I'm going to guess that at least 50% of those people are having a bad experience. Yeah, right. The staff is not giving them the time that they need and they, you know, they're, they're stretched super thin. There may be very well meaning staff, but they are stretched super thin. They don't, you know, they just answer questions and minimally, whatever and they. Okay, do you have any other questions? Okay, no buy, you know, and sometimes it's, I don't. So it's not a great experience anyway. But so, but if we're really trying to say, hey, this is an agent that can meet certain expectations, I agree with you. Being transparent about what expectations it can meet, what expectations it can't. You know, for example, our agent cannot do anything but schedule appointments right now you try to ask it, where are my glasses? Is going to say I can't answer that. I'm going to transfer you to the office. Right? You try to ask it, you know, when are you guys open? It knows how to do that. It says I'll do this is when we're open until. And also like would you like to schedule an appointment? Because that's its primary function. It's been trained to schedule the appointment. But that's, you know, being, I think transparent about those limitations is key.
- [36:37]
B
But here's the beauty of it as well. Once you know the limitations, you don't really need the voice agent to replace everything that's happened on your front desk. You can start integrating it in a stage ways so it actually complement what you have. Why I'm saying this because if, you know, if you don't start integrating that voice agent or something else in an incremental way, we're going to reach a place where the clinic or the flow going to look outdated and patients can start looking for other clinics who are who jumped on this and adjusted their workflow, their configuration, their settings to be, to be adaptive to such change. I would say even if it's gonna happen in the next four or five years, but at least you start adapting to it from now, even if you cannot.
- [37:29]
A
Okay, I agree with you and I think the point about voice agent is well made. You and I could not be. You and I could not be more aligned on that front because I think, and it is, and you know, we're going to get to this point, but I think we're going to get to the context of what is truly human. Where are we going to actually be Truly human. Where are humans actually necessary in the flow of eye care right now?
- [37:58]
B
Specialization. Because a human going to be necessary on like specializing on cases and analyzing the special cases of data and building relationship with patients. AI going to facilitate that process for patients or that process for flow. But AI will not replace, in my opinion the relationship building between the doctor and the patient will not be able, not at stage now where AI will be able to analyze or handle those special cases or like specialized cases for like doctors. This is why I am seeing a trend between even optometry I would say where there's like scope expansion or specialization, I would say in specific fields because I do believe with the current tool and the current technology that we have is going to enable multiple people even like to monitor sheet for many, for many conditions uses.
- [39:01]
A
Yeah. So when you say specialization, do you mean adding specialty to the practice like I don't know, myopia management or are you saying specialization for treating complex cases that are not routine?
- [39:17]
B
Both. So there are different type of doctors. Some of them they want to expand their practice to include those special cases and basically expand to those special cases. Authors want to expand their practice to basically are able like to treat more cases. Like for instance, some of them now are adding dry eye a lot dry eye treatment. Not so many clinics were having that. I'll say some of them are adding like ear tests for as additional score for instance for their practice. Some of them they are trying to add more specialization in the topography field for instance because some of them they weren't tapping that field was mostly like they are generalists. But now they are trying to dive into the details of how they can for example.
- [40:12]
A
Or scleral lenses.
- [40:13]
B
Yep, exactly. Yeah. So I like for instance where I am there's only two clinics or three clinics who work with scleral lenses. Now we're trying to see more people are interested in that field. I've had to explain expand the scope because just relying on the general aspect of like optometry or even ophthalmology, we're getting into a place where that's going to become more I wouldn't say straightforward but doesn't like it's requiring doctor to dive deeper with the patient.
- [40:43]
A
Okay, so let's, let's back up to the earlier list. And I know we went down the path of agents but. But you know, what are, what are the pieces of technology that people should be looking at closely or investing in right now? So you mentioned agents was one of them for sure.
- [41:00]
B
What else I do think there is A good trend in the field of funders and octs and octas. The geography one. I do think with an asterisk for that, that people need to know how to select those tools. Analyzing OCT is great. Knowing how a tool for analyzing OCT is being developed and being, is working, I think it's gonna help with the clinic. I would say that if we're talking about ar, if we're talking about general operations, I do believe the data, like the way data is being captured in a clinic nowadays is very fragmented and that doesn't require an AI to connect all those different type of equipments together and visualize it for doctors. So that's another aspect of like I do believe investing in that can really help down the road and streamline the patient flow a little bit.
- [42:05]
A
Well, and I think patient experience is a differentiator now for practices. Right. It is not enough to do the test, but you have to explain the test. Well, the practices and I constantly talk about this on the show, but the practices that do really well are the ones that can articulate, articulate to the patient what's going on, what's going on, what treatment they recommend, why they looked at something, what's the test. Even if there's nothing that was, you know, even if nothing happened, even if there's no disease detected, like we did this test so that we can find X on this, on this image, you would have X, Y and Z be potentially wrong if, and look, yours are not wrong, you know, and that's the, that, that whatever you know, insert, insert whatever you want into that explanation. But being able to show things to patients and being able to do so in an efficient way, I think is, is critical to a good patient experience
- [43:07]
B
and I do think there are now ways to do that and a personalized methods for the patient. So before it used to be very staff dependent and we know, we all know staff are overwhelmed. Sometimes you have staff coming stuff going. I would say like turn over part there is the sickness part. There are ways that you could start integrating in the clinic and I've seen that in a few of them where basically you could personalize that education aspect for the patient that can help them perform better on those exams and actually improve their journey when they come to the clinic. And that's going to really influence the patient's selection of which clinic they want to go to. They're going to influence do I trust that treatment or do I trust that doctor or that clinic to be able to select that treatment and make them to come more? Because now they're going to see the value of how that visit actually helped them explore more and more about their vision and current health situation.
- [44:06]
A
Yeah, I'm going to make some of your technologies that you're mentioning available to the folks who are, who are signed up for our AI Advisory Council. So we'll put a note in there for it and I'll put the note in the show notes. But you can give me the list of some of the technologies that can actually do that. But you know, the other thing I was going to ask, and this kind of makes sense when you were talking about relationship, right? Like this is all part of the experience. Slash, relationship. What is human about the experience? Well, what's human is the relationship. I like your point about specialization, although I think it's still, you know, part of specialization is relationship.
- [44:49]
B
Yeah.
- [44:51]
A
Now let's talk about like, big picture. How does eye care change, like, overall, like really, like, how does eye care change? Are you, are you in the group of people? And I've heard a lot of talk about this, that like, look, we're going to be able to deliver eye care in a very, you know, I, I'll call it like in a relatively commoditized way. Right. The sensors are going to get better, the technology is going to get better. You don't need a human to do a refraction. You can do, you know, people can do a refraction from home. The remote. People can get access to a good refraction in a remote area without ever talking to, you know, without having to see an optometrist or walk into an office. Like, how good is this technology going to be and how, and how far is AI going to take it?
- [45:42]
B
It's a good question and I've heard about that. But let's take refraction as an example, automated refraction. At this point, I don't think it's at stage where we could actually use it for remo. One of the challenges with refraction, I know people speak about liquid lenses, but even with the current liquid lenses available on the market, they are small and they are basically with any, with any person with high prescription. They don't really work for wall, for instance. So there is a lot of research needs to be done on the components that can allow you to do, I would say, automated refraction or like have the patient do refraction at home before being able to start deploying it with patients, before being able to configure the flow, how the communication can be between the patient and the doctor. But there's going to Be trend where you could control a refractor and basically automate the refractor so the doctor doesn't really need to spend so much time trying to figure out the prescription rather than trying to connect with the patient about what's going on in the healthcare where. Where I do see the field going on, going, going into a direction where a clinic going to operate and I be like very transparent about that with less resources and has a higher impact. What do I mean by that? A clinic who's very busy with seeing patients. I don't think they're going to need the same number of resources they currently need to operate that clinic. But also at the same time the amount of time a doctor needs to handle an equipment or figure out the prescription or figure out the treatment gonna reduce. So it could spend more time in building relationship with the patient. That's where the field I believe is heading. If I wanna summarize it in the word is going into more digitization. The clinic gonna be more and more digitized. Machine is gonna be more and more connected with each other. So you need less resources to control or to manage your clinic with more time to spend with patients. Now the asterisk on that is there's lots to be done still to achieve that many people claim that oh we can connect your whole clinic or we can do 1, 2, 3, 4 and make sure everything is connected with each other. We're still behind I would say for a stable reliable product. I would say that is device agnostic. Some manufacturer may have their own tools for connect their machines. But if you go to reality clinics doesn't always have same machine from the same manufacturer. And more than 90% if you talk about optometry they don't really. They're not with one manufacturer. So there's still a big gap in that. That regards what do you think that
- [48:45]
A
the people who are designing these technologies, what do you think they misunderstand oftentimes about eye care and clinics in many
- [48:55]
B
situations they don't really speak with the customer. Many situation they feel oh we know how the field operates, we know what we need to do. We're just going to be like basically like develop it that way. That's one. Optometry and ophthalmology is a very niche field that many of the big days don't really focus on building specified solutions for the. For that at the third one, the field is heavy on machines. There's not a lot on software and it's very. It's very important to differentiate that people who build Great machines doesn't necessarily mean they can build great softwares to be able to build that different mentality, different way of thinking, different way of approaching things, I would say so. Those are the three pillars that I believe that are missing between like people who are trying to. Some people do it better than others, I would say, and some people succeed in that more than others. But I see that as a gap or something that's overlooked.
- [49:55]
A
Yeah. So, you know, and I guess maybe that kind of brings us to the what should I be looking at? What the opposite of that is? What should I be ignoring? Right. So there's this concept of hype, hype, hype, hype, hype. That's all we've heard for two years of it's going to be able to do this, it's going to be able to do this. Whatever. If you had to advise the thousands of people who tune in and listen to this show and say, listen, I'm a technologist, I've been doing this for a long time. This is definitely overhyped. What's overhyped, Abed?
- [50:38]
B
People who's telling you we can detect glaucoma? We elected Dr. Diabetic Retinopathy. I think this is overhyped. I do think for clinicians trying to evaluate those different type of tools, I would say check the research. There are some good tools out there. I don't think there is great tools of being used abroad for like diagnostics. Let's take glaucoma for instance. So I would say pay a lot of attention to the research and the data selection part when it comes to diagnostics, other tools. When someone comes and tell you that I could integrate any device in your, your clinic and make it very streamlined, I would say you need to ask the right questions to make sure it actually does that. And the right questions would be how do you handle data that's not dicom. How do you handle interruption of machines? How do you, how do you make it secure inside the clinic? There are like multiples of different type of questions that needs to be asked to make sure you're making the right decision. I would say for those interesting. And that's where the overhype I believe is happening.
- [51:50]
A
And maybe let's expand on that because we really should think about if somebody is going to be investing in AI related technology.
- [52:00]
B
Yeah.
- [52:01]
A
In the next three to six years or three to six months. I'm sorry, what questions should every one of these listeners and every one of the doctors be asking before trusting an AI driven technology or AI Driven platform.
- [52:17]
B
If you're talking about diagnostics, if on a make it very simple, doctors need to know how a decision has been made. AI modules that are black boxes in my opinion needs to be avoided. So if someone is coming and tell you like, oh, I could detect glaucoma for you, and this is the model, this is the tool without providing explanation about how glaucoma is being detected, I think needs to like, it's not the right tool in my team. So black box type AI, I think it's a flag. It doesn't require explanation which is like that's why it's black box.
- [52:57]
A
Okay, that's on the diagnostic side.
- [52:58]
B
What else on the, let's say data connection inside the clinic? First of all, you need to demo it. Like can, can I, can I implement this in my clinic for a specific period of time, they will try it or speak with someone who already tried it, was not investor, who's not a consultant, who's not like all those like bias things from that one, like pick up a client and see, okay, how, how that's happening for them and ask questions about what's going to happen with the new clinics, how support going to look like, how a clinic that doesn't have a DICOM integration for a specific type of machine, how that's going to be integrated, Will that be integrated with my emr, sorry EMR or not? Those are the questions that I'm going to ask. And how stable and in many situations, to be honest, to know how robust or reliable that tool is, Check the team and you could even speak with the, with the, with the founders because many of those tools are still like startups, I would say they're not like big and there's still some opportunity to speak with the founder or the executive team or like see how the team works, check the team who's building that. And if the team is not focused engineering as not like focused on that, that's usually drives some of the questions like, okay, how much are you able to keep up with the evolution of the field and how much are you able to improve the platform as you move forward? Because the last thing you want to do is spend tens of thousands of dollars integrate specific tool and then three years from now that tool dies or the team doesn't really improve.
- [54:42]
A
Yeah, interesting. And then when we think about, you know, because you mentioned connection, whatever, what about things that purport to replace human inefficiency? So whether it's the AI agent stuff that we were talking about or whether it's automating something else inside the office, whether it's some steps or insurance verification or whatever. Right. And something that it frees up your staff for a higher value task or higher value activity. What other questions should I be asking about those types of tools? If I'm thinking about. Well, we'll call this category of efficiency and staff productivity.
- [55:28]
B
Yeah. And efficiency can be applied in different aspects or different stages. When it comes to clinic. There are efficiencies on the front desk. There are efficiencies on like capturing data in the clinic and showing it to the doctor. There is efficiency with the doctor visualizing that data and making analysis and deciding that efficiency on pre screening and efficiency on treatment selection. There's like four or five different faces that are like many departments inside the clinic that each one, if it's not being handled properly is going to add up to like basically bottlenecks, frustration and interrupted workflows and koenigs. So there are some of those many departments that can, you can start really implementing some automation into it that's going to free the staff to basically focus more on the treatment selection or on different aspects of it. What I can see that is on, as you mentioned on the admin stuff like the front desk type things, I do believe there are areas where you could basically streamline that pre screening. I do believe we're getting into, into an area where machines can become more portable and less dependent on, on, on staff to be able to run exams. And the third part is data integration. We're getting into a phase where I'm not saying analyzing the data, I'm saying just integrating the data in a very seamless way without having a very high cost burden on clinics. I think that's where also could be an area where it's going to really free up a lot of this task type. Yeah.
- [57:09]
A
And the other area I would think about would be EHR integration. Just adding to what you said because I think your answer is brilliant. But I think I'd add just a couple more things. Number one is the integration with the central, you know, for better or for worse, you know, the central intelligence system of your, of your practice is your ehr. That's where the data is. So being able to fully integrate and to pull data out of it and in some cases write back data into it, which is really important. And the other one I would say is patient experience. Because just because it makes something easier for your practice does not mean that it makes it better for the patient. So I am always like, I have seen a Number of these solutions that people have brought to me and said, you know, please include this in your AI advisory council. And I've said it does not make the patient experience better. I can see how it makes the staff better. I can see how it allows you to possibly eliminate half a person. But if it's not inherently designed with the patient in mind first, then in my opinion, it's a failure of design. We're not really thinking about what ultimately drives the practice, which is a happy, successful and excited patient.
- [58:32]
B
Absolutely. And we're getting into a phase where people selecting clinics based on experience. I've exactly. I've seen that a lot. Patients say for their colleagues or go to this clinic because it's using this technology, for instance. Or for instance, a patient goes to an ophthalmologist. Oh, that other ophthalmologist is using that technology. Why this is not here. That's just frustration.
- [58:58]
A
Yeah, exactly. Abed, we probably could geek out over AI for hours. And I think our time for today is coming to an end. So, final thoughts. Anything that you'd want to leave the audience with, especially when it comes to thinking about what the next 12 months looks like in their clinic and knowing that most of them are going to have to spend money, time, three resources and effort in identifying and applying and implementing solutions with more robust technologies in their practice.
- [59:32]
B
Absolutely. And I, I do think there is this image between many, many specialists that there are some concerns and some worries about, okay, what's AI and what's. What's going to happen, what's going to go on? I would say automation, digitization. That's how I like phrase it. I would say machine learning is going to be there. I do believe the sooner we start integrating it with the workflow, the sooner clinics can advance and basically work to their advantage. Machine learning when it comes, or AI when it comes, it's not going to be for replacing doctors. It's going to be for assisting them in making better decisions and improving their decisions, I would say over time and improving their workflows. So instead of trying to push it back, let's try to use it or try to be selective of what really works for us and embrace it in our workforce. I think that's going to be the key. And I would say don't go blind about AI tools. Make sure you always ask questions and do the due diligence companies out there, it's their responsibility to provide you the answers that makes you feel comfortable and makes you feel trust the technology. I'll say if anyone are short on that. It's basically on the people who develop the tools that they're not doing a great job in providing the explanation.
- [61:06]
A
We it fantastic. Well, thank you so much, Abed. Really great insights and always such a pleasure to speak with you. And just having the context of the fact that you had spent time in AI before AI was what it is today. Right. Is always so interesting to understand how you see the world and where you see the world going. So appreciate your time today and thank you for being on the Power Hour.
- [61:35]
B
Well, thank you for the invitation. Always fun and great speaking to you.