EURETINA Symposium 2022 - Heidelberg AppWay

Well, good afternoon, ladies and gentlemen, thank you for joining us for the Heidelberg Engineering Evening Symposium this afternoon. The focus of today's discussion is Heidelberg's engineering application and this is part of the program to discover more. Today we are joined by two experts. speakers Professor Urger Smith Air Force, who needs no introduction, who will discuss the fluid monitoring application developed by Retin Site and its integration into precision medicine and workflow optimization, and Sandra Desennett of Retin-ai , who is part of the product development team. at retin-ai, who will discuss with us advances in the evaluation and examination of features related to geographic atrophy, the first speaker this afternoon is my colleague jonathan shangle, who is part of our US team, jonathan will talk about the

heidelberg

engineering application feature that allows secure transfer of uh data from uh clinical data for uh evaluation, so some quick housekeeping rules.

Can I ask you to put your phones on silent mode? Please, for the benefit of our speakers. If you have any questions, could you use the microphones in the center of the room and please state your name and any conflicts of interest you may have. Okay, thank you very much, turn to Jonathan, thanks for the introduction, Chris, I appreciate it. Good afternoon everyone. Hello from South Carolina. I appreciate that you invited me. here to germany, so today we are going to talk a little bit very, very quickly because the real meat of this conversation is with our two colleagues who will talk later, but basically we are going to talk about

appway

what it is, how it works and what we can use it for potentially within the Hayek or Heidelberg environment.

So what is Appway Heidelberg Engineering recently developed a secure transfer method specifically for raw data to be deleted and moved from Heidelberg Engineering? ecosystem and securely transferred to a third-party application in this particular scenario, we are collaborating with two individual groups using artificial intelligence and big data to determine and quantify fluid outcomes and ultimately predict fluid outcomes through artificial intelligence, The idea, of course, is that the app Providers would work hand in hand with the image quality that Heidelberg is known for to improve diagnosis and ultimately investigate whether it is a standalone solution in the sense that it can technically used outside of High Environment x2, although best used within High Environment x2 specifically with Hikes 2.6 releasing very soon.

It's standalone in the sense that it will run at high x1 and it's an external app so it's much smoother. The bottom line is that it is quite safe and simple. method of transporting information from one application to another, so who is it for professionals who need significant additional information about their patients' imaging data? So ultimately, I don't think anyone in the room needs to be told, well, potentially we'll see where the liquid is and what type of liquid it is, having said that it's always beneficial to keep an eye on things for doctors. who want data-driven support, have this enhanced diagnostic guide available to researchers, we have a way to advance their research but more importantly communicate very easily potentially internally.

No wonder we had a lot of interest in the United States when we started talking about the app. A way for researchers at academic institutions to be very interested in the capacity and likelihood of internal transfer. of information, perhaps between the clinical side and the research side and of course the professionals who may not really know what they are looking at, is a good way for data-driven solutions to determine the results of their flow of clinical work, so what are its characteristics? a very secure and easy way to access in this particular scenario AI-powered data analysis in the shortest possible time and we mean the shortest possible time.

We'll show you what the response time basically looks like depending on your bandwidth because this is all done. over the Internet and our cloud solution is a matter of minutes, so from the moment you send the information to the application and the server algorithm, you get a result quite quickly. It seems like there are more and more ophthalmic application solutions every day. available and this is true with the app waste scenario, so the idea for Heidelberg is that this app functionality basically becomes an app store of available apps that you can potentially use within your clinical environment.

People are already using the hi-x2 app and so it's familiar, it's integrated and it's really a matter of math, it's just a matter of clicking and dragging the raw potential data, the volume scans that we're sending so that all these scans b are entered directly into these default applications and we will show it to you. What we mean here in just a second. So when can we get it? It's a great question, so we'll be showing it on the calendar for fall

2022

at this show, so please come by the booth and I'll be happy to give you a little demo, officially. launch with hayek's 2.6, which as I mentioned will be a fall experience, we will show it here, we will show it in Milan, we will show it in Chicago and of course also in the dog, very happy to see you all at the booth and to give it a big tour, so how is it done?

How do we do it? Within the Hikes 2 application we can configure the particular application. It is simply a matter of contacting the appropriate parties and obtaining permissions and logging in. the information from there, it's basically a matter of configuring it like you would with anything else when it comes to an application and then you're done, drag and drop the information during the entire response cycle, it sends it from the application that encrypts . it comes back through the cloud and you have an external report that you can actually and these people will show you in a moment.

We'll go over this pretty quickly halfway through. Of course, it's important that the mmv is what it's built for. and this is the application where we can show other imaging modalities within the Heidelberg environment, of course, not just the Spectralis family or the products, it has something with the diacom interface, we can see it, so you can potentially see these reports one by one side of the other. on the side of your other imaging modalities, it doesn't matter what you have, it doesn't matter where you are and of course the encryption that comes with cloud sharing is also pretty crucial, so how does it work?

See these volume scans here on the left side. you see these default apps here on the right side, you see the

heidelberg

appway

app as highlighted there, very, very simply, you're just going to drag and drop it onto the appropriate app and at that point it's sent to the app wherever you are. it may be where the server is hosted through the cloud through the secure cloud and at that point the report comes to you and of course they'll go over these types of things in great detail and that's what they're really here for. Viewing it is pretty intuitive and it's with your other reports, uh, immediately integrated into the patient encounter, so a little outline here we start with the hyx2 side of things, we choose the data we want to see, we can see it, we can look at it. the structural oct we can see it immediately inside the clinic it is sent over the Internet it is encrypted it is pseudonymised the exchange takes place in the cloud that is where we host the images while they are transferred it is received on the algorithm side through the link from appway that the solution provider takes a look, out comes their magic diacom, so anything that goes in or out of the x2 high app is a true.com object.dcm module in this particular scenario is an epdf, which means it's actually available to extract all the discrete data points, the reverse process comes back, the cloud exchange handles the transfer back to the walks to the app and there we are, you can see the original data, you can see the analysis and at that point it's up to you what to do with them correctly. you can send it to external emr, external packages, you can save it inside hi-x2 app which is a true full package solution of course encryption, security gestures, very crucial data projection, it is always very important for anyone and we believe We are taking this very seriously here in Heidelberg, so who are we going to work with?

Basically, on a daily basis, we talk to more and more people that we've been lucky enough to partner with, some people who are already on the business side of things. It is always changing every day, we have been focusing on AI technology but we are not limited to that, as I mentioned earlier it is very exciting and interesting for research and research tools hosting applications controlled by independent vendors, so these are the people who are taking the information, they use their own proprietary algorithm and their own segmentation when determining these results, so really all Heidelberg is doing is providing a secure transmission method.

You will have to consult with the supplier. They may require an individual, one-time scan. types and of course it's up to them how much it costs and we don't have much to say about it but that's something they can discuss as well and it's just the beginning, like I said we're talking to people more and more every day. We will have new applications available. It will be like the app store on your iPhone in the Hikes 2 app of particular interests. We have new results and pharmaceutical tests that are doing potentially very important things, even with geographic atrophy. and of course we are very interested in this kind of thing and for the new devices that come out with Heidelberg in the Heidelberg family, of course Hikes 2 will be the backbone of everything in the future, so everything will go hand in hand and, yes, we are happy to accept any talking points or we can move on to the next speaker.

Do you have any questions for Jonathan? Jonathan, a question for me in terms of implementing Appway in a clinical setting. What are the types of information governance questions that hospitals and clinics ask? departments about sharing data outside of the clinic with third-party providers, you know, I think the biggest question that we see and of course it's very early, but the biggest question that we see is around encryption, so we have documents related to each one. provider we work with that handles cloud sharing and can describe it accurately okay, thanks, any other questions okay, thank you very much, okay, thank you, okay, our next Peter is Professor Ursula Smith, who I said which really requires no introduction to this audience and As we will focus on the obstacle accuracy and workload of the fluid monitor, thank you.

I thank the Heidelberg team for inviting me and talking to you about something that kept us very pleasantly busy for many years and which has now materialized as the fluid monitor, which is the first automated fluid measurement tool approved by mdr which is available to the community, which means to all of you, everyone outside of everyone in the world, so why is it about optimal accuracy and workflow, what are actually the real world limitations in terms of quality and workflow, the role of oct in patient management is enormous and this sequence of events as shown here is done a dozen times, a hundred times, a billion times in the world with patients who have dme or occlusion venous, so, however, we have I learned that even though we have beautiful diagnostic tools and therapeutic substances in the real world, the benefit is limited and that is quite surprising because oct is doing an excellent job and in the meantime it has been become the tool used exclusively by half of the eye.

Doctors, particularly in the US, monitor a patient and no longer perform an actual eye exam, which means that clinical evaluation is much more important as it is the only basis for treatment decisions. What we did so far was observe where the liquid is and what it is. very clear that with the introduction of anti-vagina therapy oct has been revolutionized and that was the first revolution we went through, it started with oct in the time domain, but meanwhile we are in a 3d universe and a 3d universe comes with fluid volumes and volumes Of course, you need other tools, you need a second AI-based revolution that can take care of three-dimensional features, so the solution is AI-based clinical decision support systems to improve flow.of work in the clinical world and as shown here, classical machine learning still involves a human, but usually when I look at this graph, I always identify with this and how often, how many hours, months and years of I spend my life watching flights, looking at octaves and retrieving them, etc., so deep learning methods will help.

Let's do this in an automated way and improve quality and not just workflow. This is the graph from the Google group where they showed it. Look at the blue bars. The failure rate. The error rate. It means misdiagnosis and misdiagnosis in human experts, even if being more skilled in the field is much higher than the misdiagnosis made by an AI system, so this is the kind of quality we talk about, so what Are they real world needs? The real world need is that we need some parameter, the parameter that is most established and the only essential semi-quantitative parameter, retinal thickness, central retinal disease, is used in all clinical studies.

Whoever participates in your clinical study knows that CRT is always the characteristic that is used to decide on retreatment and to define what treatment intervals it means. treatment burden we have shown, by observing more than 40 000 oct volumes, that there is no correlation between the central thickness of the retina and the true amount of fluid, which means volumes particularly in neovascular, and the correlation between the crt and the true amount of intraretinal fluid is less than 0.5, so the CRT does not really reflect the activity of the fluids, which is very worrying because we have to take into account that the new blockbuster treatments that come and promote longer intervals have tried to introduce some personalized measure of disease activity, but it is still tied to the central retinal disease and it is also very complicated because for pharisma, for example, if you look at the main publications, you will get the recommendation that an increase of 50 microns in the thickness of the central subfield measured by sdoct compared to the average value of cst in the two previous scheduled visits or an increase of 70 microns compared to the lowest value recorded before this is not something that is a routine testing, so we need to be able to measure accurately and reliably and do it quickly.

Why do we need volumes? It's something that's a fancy new idea just because we can measure it. No, it's because it has relevance to function, what we want to address is the patient's function and not just the oct and when we look at the role of fluid volumes in the Hawk assay, which is the top line and in the Harrier assay, the bottom line, remember this is papa sisumab. a very efficient substance, then you can see an increase in volume and these are the darker lines. This is the quartile of patients where the volume in each intraretinal, subretinal, and pediatric compartment is highest.

They all come with vision loss, so yes, the change in volume is very important. To a large extent, it is decisive for the visual result, so we need something, but we need something that is robust and that we can trust, so if you have an artificial intelligence algorithm, you must not only design it, but also validate it, validate it in hundreds of thousands of In many volumes we did this in the field of clinical trials, since the Vienna reading center is the main reading center for every phase three study of substances that you can think of, so We have access to many octs that show any type of disease, particularly neovascular. disease, we developed an algorithm that is capable of automatically identifying, localizing and quantifying fluid, particularly intraretinal fluid and subretinal fluid, and we have validated it in this enormous volume of data sets, then another decisive step must be taken that Goes to a real world scenario, the real world data of course is more pitiful since no one gets paid for it, there are no study nurses or doctors taking all the time in the world to make a perfect picture , these are just images that are made while we are there. seeing patient after patient, this is taken from the Wipes of the Vienna Imaging Biobank data, these are basically all the patients that we have treated over the last 13 years in our own outpatient world and look, we then extracted the 25,000 and the 14 000.

The spectral volumes of 1127 patients with neovascular disease that we treated for five years and these five years are shown here and you can clearly say that the tool is very capable of showing us where the deficiencies are over the last four years, is there recurrence of liquid and can be measured very accurately in terms of volumes at the nanoliter level, so this should not happen. It means that the real world results from simply using qualitative tools are nowhere near good, so this is in a way the path of developing an expert tool for experts like you or for any ophthalmologist out there. training and on the way to becoming an expert, so a proof of concept was provided in 2017 and then validated in a controlled clinical trial setting in 2020, a real-world retrospective analysis was published in 22 prospective.

The clinical study started a year ago and the fluid monitor validation publication was also submitted, so we have done it in a completely reliable way. This is the roadmap to success from code to clinic. World-class clinical research was there at the beginning, but no. More than 10 years ago, robust AI development was developed over years with extensive testing and validation of biomarker algorithms to ensure the highest quality, then regulatory authorization is needed, so this is not just a tool of AI that is inserted into any instrument like a coffee machine and then we say: use it, it worked in our hands, it needs regulatory authorization and this is very strict and the fluid monitor is the only automated tool that has approval mdr mdr is the approval, the new approval required by the European authorities, who are also ensuring that, in particular, data protection as the data is sent to the cloud works at a thousand percent, so the result is the rate within the monitor of fluids, a reliable artificial intelligence software for clinical use in clinical routine.

What do we need in clinical routine? We need the big picture at a glance. This is what we seem to already know, but this is just a purely clinical qualitative interpretation with all the subjective variability that entails. It's just qualitative. Everyone would have another approach, even when you have readers in the reading center, different readers would make a different type. of annotation, so there is a complete lack of consistent decision making, the evolutionary step is now an accurate fluid activity meter that converts pixels into nanoliters of volumes and provides reliable objective and quantitative results, as seen here, and this It is very plausible since the doctor can not only follow the flight as if it were walking across the retina, but at the same time also have the location on an office image, so this is all at a glance now that we have a blood pressure meter. accurate fluid activity and we know exactly why the result is what it is and It's not just about understanding that this is an individual disease in an individual patient at an individual time, but it gives you everything you need to know in one report of fluid measurement so you can see here that in this cloud based clinical decision system which is currently marked with CE mark in mdr conditions for Heidelberg Spectralis system which provides a report while the patient is still with you in a few minutes .

It depends on the speed of your own Internet. You see intraretinal fluid in an office user. Sub ipe of retinal fluid. you see the scan, the b scan and this can now be used for the clinical path, what does the clinical path look like? An oct image is made, then the data from your image management is integrated and then comes the access to the AI ​​application, there is an upload of a specific image that the doctor selects in a cloud located in aeu that will then work with the data from the image and all the changes and will provide a fluid monitoring report, so now you are looking over the shoulder of one of the doctors who is using that is selecting a specific image and sending it to the cloud where the fluid monitor of the Marketplace app is working on it and produces the report that I just shared with you, so the report is based on real-time image analysis, as you can see here. in a fast flight mode but also in an office scan b and gives you exactly the numbers you are interested in for the three different liquid locations you see.

Red is intraretinal fluid. Yellow is subretinal fluid. Blue is detachment of the pigment epithelium. There is the indication. of the volume in nanoliters here in this case there is no subretinal fluid however there is a large amount of intraretinal fluid in the central millimeter the central millimeter in neovascular and it is more important as it determines the visual outcome we also add a six millimeter scan because this is important for diabetic muscle edema and for venous occlusion and anything that relates to off-center disease activity, of course this needs to be used in clinical trials, even more so because clinical trials are laying the foundation for the new future therapies and this is where we need to know as much as possible about the effectiveness of new substances, so a group of red inside is not only a fluid monitor for clinical practice, but for an excellence in reading driven by Yo.

Everyone can know about the Vienna reading center, which is a large, fairly established reading center. reading center that has been operating for 17 years already started manually over time a lot of automation has been introduced as a basis for human annotation, but it is very clear that humans cannot annotate fluids, particularly not in clinical studies where a rapid response is needed . This AI-powered reading excellence provides AI-based image analysis services to improve performance and quality in clinical trials from start to finish using only validated biomarkers and at the moment there are 18 different biomarkers but for Of course, with each type of treatment, each therapeutic objective.

The new biomarkers will be relevant, it greatly accelerates the analysis of image data and optimizes clinical trial workflows because one of the other characteristics is not only accuracy, but it is real time in the real world. So what does real time mean? Everyone involved in clinical trials knows how to do it. It is tedious even for the researcher to evaluate patients and wait for a reading center evaluation that arrives a few days later, which will then confirm that the patient is in or tell him that the patient is out, and data collection again becomes very tedious Many images must be taken, they must be uploaded.

Manual reading requires a lot of time for fluent annotation. It takes at least five hours of an expert reader, but here we can have the liquid volume with a click of the mouse and that during a busy study sequence. Of course, it has a very high value. The analysis of the study in the end is already done, since the study is analyzed with each patient, so there is total transparency. There is complete communication between the researchers, patients and study sponsors, which basically reduces time. To finish a study up to half of what was normally needed, we have done it in one of the studies that is stunning and has been presented at the top.

It is very easy to acquire in October, it is uploaded and then the image is processed in the cloud for central. disease registration in the subfield Four-wheel registration was important and retinal thickness segmentation, the image returns immediately while the patient is still there and there is an interactive validation by the ophthalmologist if necessary and then a report and this study visit. It is completely made with the highest precision. This is such a simple report for a central subfield thickness and you will see that the report can be adjusted to whatever the needs of this specific protocol and this specific FDA setup are. um, yes, I'm doing well, so a multicenter prospective clinical trial is currently being conducted in the Vienna department and you can see how it works, not in a single visit, but on an ongoing basis.

This patient has been treated for nine months and you can see how well fluid control guided the treatment. a perfect visual result of 20 out of 20 and these are the liquid volume peaks that have been measured this is the benefit on an individual scale on a larger scale it is very clear that anti-vegf therapy has been referred to as the biggest budget drainIn medicine there is a lot of budget burnt and the data from the United States, as well as the German data, the European data and the British data show us that this is very inefficient so far and that we have to do something about it.

Would you like to take this next step with us, you are cordially invited. You can learn more at the Heidelberg Engineering booth or the Retina Side booth so you feel like you can buy a piano and play. Thank you so much. Do we have any questions? for professor smith's effort, a very quick question from me, for doctors to take advantage of the added precision of fluid segmentation, are they going to need to change their approach to the treatment regimen in the US? Do you believe that the vast majority of doctors are using treatments and extend a small prn ratio and fixed doses.

Is there a preferred treatment regimen to take advantage of fluid segmentation? This is a measurement tool that will improve and bring accuracy and speed up your workflow. Whatever type of regimen you are using is independent, it is your own choice. but whenever you see the patient you would like to know what exactly the disease activity is and what exactly the perfect therapeutic response is. Thank you. Okay, one more question. Sorry, sorry, Fabio from Portugal, but it depends on the segmentation lines. because the segmentation takes into account the oct, sometimes when you want to see the volume, the segmentation is not correct, that is what makes the segment, the commercial segmentations are not correct, that is why even for the core disease of the retina you have to correct this, but this algorithm um The fluid algorithm, as well as the ga algorithms that we offer, are all validated and the accuracy and sensitivity, everything has been annotated on a very large data set.

There is a post pending to prove that yes, no errors, okay, okay, so our final speaker. in the afternoon is uh sandra desanat from uh wrestling ai and um sandra is going to discuss the use of ai to examine new advances in ga treatment. Thank you very much for the introduction and thank you for the opportunity to speak here and also for the nice introductions. Before me, yeah, today I'm going to talk about the different apps that we have at Appway and focus on the next ga, but first, maybe, what company are we? Maybe you don't know us.

Retina, simply put, is software. company we make software is a medical device to improve the efficiency and quality of diagnosis and treatment in the ophthalmology space and we help in clinical trials and also for the clinic to improve workflows and we are a global company that works in different countries and our main focus And you saw it very well before in the presentation. That's what we're already doing. On the one hand, we have retrospective analyzes of studies. Large clinical studies and we also do clinical trials using our tool and we do it with CE mark. and platform approved by the FDI and also see the fluid and the layer marked uhsegmentation that you will be able to see in our booth or also with the appway application and um yes, thanks to the hard work that we are launching today or we did in this presentation, we are launching the heidelberg way apps uh and them You can see in the booth and also in our booth how they perform and our approach to the different applications that we offer are based on diseases, so we want to use the different models that we have to really focus on what which is important for different diseases, so we have applications uh for amd drdme cme and soon also for ga um and we believe that it is important to focus on the different areas, especially in the clinic, and this will allow monitoring the progression of the disease to over time having very precise measurements, reproducible measurements that can be compared. also between clinics in different countries and we hope that this will also improve patient engagement because simply showing raw images to a patient may not be convincing, but if you have the analysis and you see a change in these measurements, it could give you more confidence. the patient, for example, continued treatment, especially for the next medications in the ga space, and we think that beyond the fluid alone, we think that also having macular biomarkers of layer segmentation or the presence of absence of different types of biomarkers is It is important to combine them all depending on the disease in the different applications that we have and here, for example, for dryness and liquid, maybe they are not so important, but you want to know in a quantifiable way in the different sectors how thick the subfield is. central of the retina, but also in the periphery, is there hrf or not?, and how much is there. and then for neovascular diseases, of course, we also want to have fluid, we want to know the space sub rpe uh drusen pds resonate pds we want to know how much of that there is and how it changes over time and yes, coming soon and that's what I'm going to focus this talk on geographic atrophy and should come at the end of the third quarter, as soon as an investigation uses only the first application that you can also try in Heidelberg against the wind, so what would that report look like?

It looks like, as you know, there is a PDF report coming from this software and as you've seen before, we have different types of models that give you information about the disease, from different types of modalities, so on the left side is the that is. Seeing is more of a static view of the disease, so what is the current state? How much atrophy lesion does he have according to the oct? How much do we have in faf? and what that looks like in the oct. We usually select an interesting b. Scan from the entire volume so you can't quickly see the worst scan, for example the burst area in the scan or even the center if that's important and take advantage of what we already have in terms of layer segmentation and also look at hyper streaming in the choroid and we also look at the reflectance in the photoreceptor layers and the rpe on the right side is another report that is also in the gs base and that helps you see what will happen in the future, so we have developed a model which takes a current oct, a volumetric oct and predicts what the lesion growth will be like in the future and this is based on our argument that we have already done this but I will go into more detail but basically as a report so you will know . this patient is a rapid progressive will there be forward progression or not or when will this happen um or even for the collateral eye will there be ga and when will it happen so let's start with the model that already addresses what is being used in studies currently in clinical studies as a criterion main valuation and that is our faf model.

We do geographic atrophy segmentation in faf and this allows you to quickly and during the study quantifiably measure the ga area and get the reports continuously during your study, um then based on our models for segmentation and we have fluids, we have the space sub-rpa srf irf but then we also have different layer segmentations rnfl and so on up to the choroid and we use this as input and also the oct information so we can do the ga segmentation in ocd and this is what it looks like so here we have the aurora segmentation on the left in the b-scan, we do that for each b-scan and we add this information and project it onto the localizer so you can see the extent of the oval affectation and of course you can do it over time and then you'll actually see the trend, the actual trend of the patient, um, and you'll see if what's happening to this patient and there's always a chance that all of these applications will disappear as well.

Go back to our platform and have it have interactive feedback on those patients, if so, if you want to share that data, and then as the last model, there's progression prediction, as you've seen before, where we really leverage all the information that we have. . of the oct and the certified algorithms that have also been validated and are quite robust and we take them, make an artificial AI model and predict the growth and spread of ga at different times in the future and for this I only need the oct of course , we could add more information, but since Oct we can do it and I just wanted to highlight this patient who seems interesting to me, that is what I was mentioning before, this is an eye, the collateral eye of a patient with atrophy already progressed in the another eye and as you can see in the baseline, there is no ga, there are particular rules recited, but there is no ga and then what our model does predicts that there will be this. ring of injuries that will happen in the future um and if we look at the actual data that we have from that future um it's actually between those two points that I show, I'm showing both of them um that the injury actually starts at that same Of course, we can never be pixel precise.

There's also tracking over time that might move it around a little bit, but this really gives you an idea of ​​how big the size of this injury is going to be, when it's going to happen and this is a tool that we think is going to be important. with the new medications that are coming in to really inform the patient that the treatment will actually prevent this type of lesion growth or at least parlay its occurrence and here it's the same as an animation so maybe you visualize it better I don't think so . This will be something that will be easy to do in a pdf, but it is something that is also for the doctor to see how it evolves and then it can be compared with different treatments, so this would allow a good natural progression to be seen.

It says like this, but if you treat it, you will have much less growth, it could prolong your vision for x many years and I think this is a very powerful tool to convince patients to continue taking the medication or even start taking it. although they have perfect vision at the point of presentation and this is more of a static view, it's the same information, but it could give you a snapshot view of the growth patterns and when this atrophy is going to occur in this patient, so what you're seeing. Basically every caller tells you when this region is atrophic and you see that there is a bout of asparini from the front region and this typical growth around the fog, yeah, in terms of posts and what we've been doing, all these things are published in In terms of models, our fluid, layer and biomarker segmentations have been evaluated.

There we have been marked since 2021 and we have also been publishing a lot within the company and with our many collaborators, and something that might also be of interest is the Razer Bill study that we are doing with Novartis in the public space looks at how this works type of automated analysis analysis, how does that change the way doctors treat patients or at least diagnose patients. be a better treatment better diagnosis is there a significant difference and is this an ongoing study? You can read the article and I hope we get some preliminary results soon and then for the other algorithms, we also have publications that you can read online, they are free. available on our website but also in the editor, yes, so in conclusion, what we have is a validated algorithm and for research we use algorithms both on our platform and now also on upway so that the group can directly access them from the Heidelberg machine and are already being used in many different studies, allowing for comparable and reproducible quantifiable data and, in my opinion, this will help to really improve the treatment because now it can be compared from one clinic in one country to another. what's happening and when to switch, for example, from one medication to another and those are my final points. um yeah, these quantifiable methods will help you decide better because now you can trust what's been found in the clinical study because it's exactly the same analysis if you don't do it.

To get the right result of the treatment, now maybe you can know why you can get those measurements, compare them with the study of other doctors because they become comparable, which actually hasn't been comparable and finally, I think it's a very point , very important for the near future. It's really letting the patient know what's happening, where we are with the disease, what might happen in the future and what the likely outcome of the treatment is, so I would appreciate you being here and visiting us and in our 40s, stand or write us an email email if you're interested, and of course try the apps at the Heidelberg booth, which is quite interesting to see in action.

Thank you so much. OK. Do you have a question for Senator Desantis? Maybe I can add Oops. Add. for ga for the audience so ga is going to become one of the biggest challenges for all of us retina experts or ophthalmologists just keep in mind that we are talking about five times more cases of ga um than neovascular cases and we are already inundated and we also continue to note that the human expert can see the bubbles in terms of fluid, but the changes in geographic atrophy are subclinical and we need to be able to identify patients who have disease activity, so we had the privilege of analyzing the dataof the appeal. trials or the first complement inhibitor that showed proof of principle and yes the therapy works but it doesn't work in all sectors, it works in those patients who have disease activity and disease activity can be defined the old fashioned way used in rpe, but what?

What we suggest is to measure photoreceptor loss because it is a disease that starts at the level of the photoreceptors and the rpe just follows and you can see very well in the models that the AI ​​tools provide where the photoreceptors are falling, this is where treatment will have an opportunity to work and predict disease activity based on the level of photoreceptors, which is key, even for the conversion of patients who only have drusen and nothing is seen in October and nothing is seen clinically or see nothing on faf up to 18 months before. the algorithm can recognize where the photoreceptors will fall 18 months 12 months later, so this will be a completely different treatment approach and not be surprised by the liquid.

Thank you, thank you for the good presentation. I wanted to ask what the difference is between validated and certified models. The models that you have shown on a slide in terms of validating AI performance are if, when you do the validation studies, you have a validated algorithm, but you still need to follow the steps of the regulations, so your company needs to be certified ISO. Have you gone through the process of the regulator that is in Europe? You have different bodies in the FDA in the United States and you are certified, at least how I used it was after you went through all the steps of the regulation.

Validation is just a study. evaluation okay, so in terms of AI performance, it would be the same steps that you would apply in both cases in performance, yes, then the validation study will give you the performance and then you need to have all the processes, qms, all that and the ga tool you need. presented, it will be a ce certified tool, yes, including prediction function, the plan for ga is first segmentation and then progression prediction, and it is for 2023, end of 2023. Thanks, okay, a question from jonathan, yes, this is actually it. To both of you, thank you very much for the introductions.

Regarding the analysis of the b-scans with the volume scan, is there any kind of restriction or requirement for the size or number of scans to perform your analysis? So for this segmentation we don't do it. you have a restriction, you can use circular scans, marked patterns or cross patterns, full volumes, it doesn't really matter, yes of course the lesion must be within view, that's the only restriction and we think depending on the disease, sometimes a centimeter is a millimeter. The most important thing is that sometimes you need to go to six millimeters and you will be surprised how much fluid you will find, and particularly for ga, what we do, of course, is the six millimeters, since the ga injury begins outside the central millimeter and that would be in different layers than those of the photoreceptors, even the thinning of the photoreceptors, thank you, okay, is there any final question in that case?

Could we show our thanks to our speakers one last time? Thank you for joining us this afternoon, thank you for your time, may I encourage you to join? If you would like to learn more about appway or the apps themselves, visit the Heidelberg engineering booth or visit the retin-air retina site booths for a more detailed discussion. Thanks again. Good day.