 innovative technologies and techniques. The topic of today's presentation is artificial intelligence in cardiovascular medicine, current status and what does the future hold. I'm Zvonmy Kraser. I'm an international cardiologist at Texas Heart Institute and Baylor St. Luke's Medical Center in Houston, Texas. Joining me today is Dr. William Kahn or Billy Kahn. He's a cardiovascular surgeon and vice president of Johnson & Johnson Medical Device Companies and executive director at the Center for Device Innovation at Texas Medical Center. He's also a professor of surgery at Baylor College of Medicine as well as an adjunct professor of bioengineering at Rice University and also at the University of Houston in Houston, Texas. Another expert that's joining me today in this program is Mehdi Rizavi. He's a director of electrophysiology clinical research and innovations at Texas Heart Institute. He's also an associate professor at Baylor College of Medicine and also adjunct professor of bioengineering at Rice University. He's a cardiologist and electrophysiologist at Texas Heart Institute and CHI Health Baylor St. Luke's Medical Center in Houston, Texas. Welcome gentlemen to this program. Thank you. I do not have any disclosures related to this presentation and also Dr. Kahn doesn't have any conflict of interest and Dr. Rizavi doesn't report any conflict of interest related to this presentation. Now to understand a little bit better about artificial intelligence, artificial intelligence is an attempt to explore the architecture of human brain to perform tasks that conventionally were not able to be resolved with standard algorithms in this field. It's an attempt to mimic cognitive function of human brain in the process of learning and problem solving, understanding speech and strategy and many other aspects. The important thing about artificial intelligence and efforts and goals is to be able to better collect the data with an effort to improve accuracy of diagnosis in cardiovascular medicine. Also it offers the opportunity for early detection of disease and also the prediction of outcomes. Also the benefits could be increasing access of quality of care and better disease surveillance and timing of intervention as well as to uncover novel associations between data and disease and to reduce human errors as well as to decrease cost of medical care and also improvement in imaging such as equipment performance, adding new algorithms as far as imaging is concerned and improvement in techniques. We also hope that this could add to improvement in professional data sharing either in publications or presentations or in statistical analysis. There are many, many companies that are exploring this option and one is this particular one, IDXDR. It's an AI diagnostic system that can autonomously analyze images of retina for signs of diabetic retinopathy and this is very important in early diagnosis. There are a huge number of patients with diabetes that are at risk of developing diabetic retinopathy and progression of it to blind this and we can identify diabetic retinopathy in very early stages outside of the spectrum of specialists. I think it would be a great benefit to the society. So this can be used in a primary care setting without any expert interference or observation or performing the procedure and this particular technique can very accurately establish the presence of diabetic retinopathy as we can see with very high sensitivity, specificity and imageability and then if the diagnosis is established, the primary care physician can refer the patient to diabetic retinopathy specialties that can address this and treat it in early stages of this disease. There are many other examples that we can share with you and one of them is a very common one that occurs relatively frequently in an international field and this is a progression of renal functional impairment. Let's take a typical scenario of performing a diagnostic angiogram in a patient that has a moderately impaired renal function and that contrast load obviously can lead to a worsening of the renal function either on a temporary or permanent basis and it will be extremely helpful to know and predict whether any particular patient in this situation would end up with ATN and might require additional hospitalization. So this particular algorithm can predict worsening of the renal failure on the basis of obtaining serum creatinine on day one and day two after the procedure and predicting how high is creatinine going to rise and whether this is going to lead to ATN and this has been reported in the literature as well. Here is another very useful example that I was directly involved with and that is related to the use of AI in clinical research and publications. Billy, you mentioned IBM Watson computer as one of the mega computers that can help us in many different areas in variation and treatment of patients with cardiovascular diseases. We wanted to determine in this particular scenario what happens as far as complications are concerned with the use of large border arterial sheets during various kind of interventions such as EVAR, T-VAR and TAVR and we wanted to compare it with surgical access and repair of femoral arteries in this subset of patients. Now as I mentioned our particular interest was in this population of patients but to analyze real life experience in this type of scenario surgery versus endovascular repair of the access site is an important thing and we for this particular reason implored IBM Watson explorers data system with IBM Watson health and the reason for it is we wanted to get objective analysis in a huge number of patients without introducing bias and finding comparative group of individuals that either had a surgical or endovascular repair of the access site so we were able with Watson computer analysis to actually search longitudinal data among 55 million US patients that underwent those type of procedures between 2012 and 2017. Now this would have been impossible to do without computer access and analysis of our if I could just add some Zvonko you know now that's being done prospectively there are companies startup companies that are focusing on using AI to identify patients to enroll prospectively in trials and one such company deep six that uses AI machine learning and was brought in there was a drug study the Texas Medical Center wanted to be involved they went combed through medical records and only came up with two patients that met the criteria deep six when hooked to the online medical record identified 46 patients in two hours. What we did is we use Watson IBM sister which was in a lot of instances actually for educational purposes could be free of charge which a lot of scientists are not aware of it but this computer can analyze language pattern so it can do language processing image recognition including tone analysis and deep machine learning which is very important it also has a Watson avatar that can guide you through the steps how to design the data analysis and it can also perform cognitive computer computing technology as far as text mining is concerned. What is also very important and that was useful to us to look at EMRs of those patients Watson has access to 90 different servers will combine data storage of over 200 million pages of information so again what we wanted to do is comparative analysis of percutaneous access and closure with one of the devices so called for a closed device versus cut down we wanted to make sure that we match those groups as far as all the important parameters of concern such as age gender index type of procedure the way the procedure when the procedure was done baseline blood work to make sure how much of blood loss occurred when a patient needed blood transfusion post procedure and presence or absence of peripheral lasko disease and then we also wanted to perform multivariate regression controlled analysis for baseline such as use of anticoagulants and presence of atherosclerosis malignancy COPD history of MI stroke and blood transfusion and many other variables that could influence the outcomes and here is what we found using a logistic regression model performed for mortality and morbidity and we can see actually that there is a quite significant difference between cut down and use of perclose at 30 days where perclose patients at 70% less likelihood to die at 30 days which was absolutely amazing and remarkable and also the hospitalization was for 43% shorter for perclose patient and patients that underwent cut down and this was published in one of the journals but you know maybe one of the major difficulties that we had in publishing this manuscript would you guess what was the reason for it people thought you had a bias or that the patients weren't completely weren't randomized well let me tell you interestingly enough when we submitted this for the publication some of the reviewers didn't even know that Watson IBM Watson exists number two is they didn't realize that IBM Watson doesn't have a bias because it chooses the algorithms that are totally outside of scope of bias and and can match the data more accurately than anybody else could do it in that way that was so lack of knowledge and lack of application on regular basis in research and education and publication was a major deterrent actually to have this published and then of course there was a skepticism why would surgery be so inferior to percutaneous access and repair you know one point about bias though that's sort of true but AI can have a bias because AI generates rules based on the the information you give it and that's a big point in AI is to avoid AI bias because the computer is rules based and as it develops its algorithms you can unintentionally cause the computer to develop its own biases they're different than the biases that we as practitioners might have but that's an active area of study and a look out for the bias will always exist because it basically depends on the input of data correct if somebody labels a certain complication inaccurately that data will be recorded as reliable and if the information is not there again it cannot be used so yes there is always a possibility of bias. Meta analysis could be full of bias as well particularly when you mix studies that have less than optimal information and database and small small set of data and you mix it with a large data that has a totally different information as far as reliability of the data is concerned but anyhow this just tells you that there are many aspects as far as AI is concerned but in this particular scenario I think it was very meaningful information now another aspect that is very important where AI plays a major role and that's in the cardiovascular imaging and one of the example is the Stanford University mega computer and database that has been used for those purposes. Stanford was one of the earliest to initiate the project of AI in imaging and they used a cloud-based, beta-based scale searchable repository of diagnostic imaging studies for developing AI image analysis systems that would guide physicians in interpretation of the imaging and also as far as improvement in outcomes are concerned as well. Now what's really important is that that data is available to anybody for instance like related to chest x-ray analysis they gathered over well close to a quarter million chest x-ray to be able to determine and interpret various kind of conditions that might be present on regular chest x-ray and then to determine whether AI can accurately predict the diagnosis. Another one is from RSNA as far as chest x-ray collected from NIH and also RSNA as far as CT brain over 25,000 exams for diagnosis and evaluation of CT of the brain analysis in echocardiography over 10,000 echos have been collected for that particular purpose as well as low extremity radiographic imaging and MRNet all of those have been collected and are available for clinicians analysis and use and as we can see for instance like in a chest x-ray interpretation of different pathologies the correlation or sensitivity and specificity between expert interpretation and a computer or AI interpretation is very very close it's very reliable so we have achieved that goal in the fields of chest x-ray and many other things that I have listed here. Here are those two typical examples of what can happen when the patient with coronary disease as far as indication for intervention is concerned. Here we have a patient A and we have a patient B we have a CTA of the heart looking at a coronary artery anatomy and we can see patient A on the left hand side panel upper panel we can see there is an error angle of the left anterior descending there is a calcification and we would estimate this that stenosis is 70% or so. Now what is important and what we have available now we know that from the international cardiology field that fractional flow reserve is more reliable way of assessing the severity of the disease and the outcome as far as this severity is concerned then just interpretation of a visual perception of severity of the disease and we can see here in the middle upper panel that actually FFR is 0.86 which is basically relatively normal flow and also as far as distal flow is concerned on the right hand side again we can see that it's relatively normal flow however when we look in a pattern of a patient B in the lower panel again we can see that there is calcium and we can see that there is a narrowing in the LAD that we would estimate again to be 70% stenosis but when we look at the FFR then we can see that actually that there is a drop in FFR in this particular scenario so we can therefore more accurately predict the severity stenosis and the decrease in flow and the problems related to the outcomes as far as the risk of myocardial infarction and the need for intervention so this is a very useful way of using imaging to establish a accurate diagnosis and to guide you in what is appropriate time to perform intervention and in this particular scenario using FFR with heart flow we can see that the correlation was very good as far as diagnostic accuracy in concern in this particular clinical trial so a lot of patients will avoid unnecessary procedures operations and even diagnostic cats and so that's a great example of what AI and image processing can bring to the population and healthcare now let's go to one of the examples where an ultrasound it can be greatly used in diagnosis using artificial intelligence one of them is this GE venue with auto IVC analysis and another one is that you already have mentioned DIA or VIVO EF that can actually assess the ejection fraction from a handheld device and also from Philips so-called EPIC study using AI where we can assess peak systolic longitudinal strain just from the echocardiography and that will certainly have importance in guiding the physician as far as any particular issues or concern might be related to any particular patient now one of the issues with ultrasound imaging particularly in echocardiography is an unnatural hand-eye coordination that is required to perform imaging so it's truly unintuitive in a certain way because whoever is performing the imaging has to look intermittently back and forth between the screen and the position of the probe and that also is complicated by variety of issues that you might deal with with different patients whether it's a different patient's body constitution or due to let's say imaging of the heart in a patient that has a COPD where the imaging could be cumbersome so how to come over those obstacles and obviously AI can assist and help in this particular scenario and one of them is this captioned AI that actually emulates the guidance of an expert sonographer by providing more than 90 types of real-time feedback instructions during the procedure how to position appropriately the probe how to obtain the optimal image and when to obtain the optimal image actually this particular software has so-called auto capture that automatically captures the image when it's optimally acquired and that can be achieved in multiple view and then the measurements can be obtained and this was validated in two centers and as far as sensitivity was concerned between the expert and the non-expert obtaining the images using artificial intelligence was very good as we can see here for LV size for LV function RV size pericardial infusion and therefore this technique is very very useful because it can be used in the triage for instance an emergency room in the field in the rural areas where expert in acro cardiography or ultrasonography is not present I wanted to add there's this other company that's been out for a couple years called butterfly network and they're leveraging your iPhone so the ultrasound beans the image to your phone but it also leverages the camera so it's showing you're showing your hand and how you're holding it and like you just showed it's telling you how to change and move to capture the best images and all these things are trying to democratize good ultrasonography because if we're going to be able to provide quality care all over the world and all over our country to the rural areas it's going to be replacing technique with technology and that's isn't that one of the huge things that AI brings into focus is the ability to leverage leverage artificial intelligence to do that another way to use AI is in a trans esophageal acro cardiography which is typically performed by an expert in the field where you can do a volume analysis real-time 3d Doppler 3d valve analysis which is very important during international procedures such as valve procedures, tower, mitral valve repair and so on that we can see here using AI systems we can demonstrate clearly what the problem is and how to address any particular issues that we might encounter during complex international procedures one interesting thing as far as radiology is concerned all the technological advances that have occurred within the last decade or so related to the use of AI in reduction in radiation dose in integration of imaging modalities and improvement in image processing all of those led to better faster safer and more effective care and better outcomes for our patients and here are a few examples this is one of the MR applications where we have a MR scanner that then reconstructs images in real time and you can use AI algorithm that analyzes the findings and establish the proper findings to guide you in the diagnosis of any particular condition now the great advantage of it is that not only it can guide you as far as good acquisition during MR but it also reduces the scan time from typical 90 minutes in complex scenarios to less than 15 minutes which is beneficial to the patient and beneficial to the individual that's performing the procedure and also individual that is interpreting the procedure another important feature is using AI technology in image acquisition using so-called avatar system that can identify patient's position on the x-ray table for landmark detection for the range detection for the range adaptation over time and isocenter positioning so you can obtain images in the most optimal way and patient direction analysis it can identify head and toes position of the arms and so on and all of this can be achieved with advances in artificial intelligence and computer analysis and you know that's going to be very important as you have seen more than most the number of things we can do percutaneously is expanding geometrically with the introduction of percutaneous valves percutaneous repair of aortic aneurysms percutaneous interventions in the skull on aneurysms and clot extraction or many the stuff you guys are doing in the heart for electrophysiology and so when you are navigating not blindly you've got ultrasound you've got fluoroscopy but bringing AI in to help you make better decisions to do procedures in a shorter period of time and get better results that's going to be huge and there's a proliferation of new technology in that area as well if you have Zvonko I think you use it extensively in aortic work don't you right right well let me share with you some of the benefits of AI guidance for international procedures and actually uh many if you would be so kind to talk a little bit how we interventionists can be helped using large bore sheets avoiding serious life-threatening complications such as retinoneal bleed and complications related to it and you are an inventor you're a scientist and you have this amazing idea how to apply existing technology to this particular problem so maybe you can comment a little bit about it sure thank you so you know in the image that you're showing what we are doing is one of the major problems everyone here knows is internal bleeding during the course of these invasive procedures especially when you're getting vascular access to the large bore arteries and so the this the technology that you are showing here basically it creates it introduces another variable if you will another dimension this specific technology is right now not being used with AI but the fact that there's another variable that's introduced in the decision-making progress process actually lends itself to to more precise potential even predictions of bleeding but what this system does is it is measuring the impedance of the tissue around the sheet in real time and is blood accumulates blood has a very very low impedance part of the lowest of all biological tissues and so his blood accumulates in the perivascular space the impedance drops and that's correlated pretty nicely with the amount of blood that's extravasated and you can kind of imagine without really trying that hard that you know as the number of patients who are who have this who imaging and these sheets utilized increases over time then you could get to a stage where not only are you saying level one level two level three based on volume but when level one happens perhaps you could actually predict is this patient now going to go to level two level three and so it's not only a warning that an event has occurred very early in the process but actually the next step with AI would be that I'm going to predict that an event will occur or there's a very high likelihood of it and therefore can I actually do something to prevent the problem before it even happens and that would be that would obviously be very significant in the management of patients so what is the sensitivity and specificity of early bird in detecting it's sure it's it's it's both the sensitivity specificity are upwards of 92 percent and this is both in series of preclinical and clinical studies so it's it's quite high and you know if anything the impedance variables can change you know very very quickly it's the first thing that we have noticed that changes is the impedance before anything else of of the variables itself and again that nice correlation that relatively robust sensitivity and specificity we think potentially down the road again could be a very important additional dimension towards predicting the boots thank you very much for this very meaningful information another very important advances as far as imaging is concerned that is very useful in my almost everyday practice is so-called fusion imaging in case planning during EVAR, T-VAR and TAVR by obtaining preoperative CTA which then can be corrected for parallax correction, automated CRM positioning and also to assess vessel origin obviously this can then eliminate errors during the procedure and also in pre-procedural situation offer virtual procedural planning here are some of the examples we can see of the use of CT images prior to the procedure and then fusing it with the images on the right hand side during the procedure where we can identify clearly the position of the renal osteo or origin of internal iliac arteries or any other vessels of importance during EVAR or T-VAR which is extremely helpful because this way we can save time save amount of contrast that we use and perform the procedure in safer and more reliable way one of very exciting fields is so-called force imaging or fiber optic real shape imaging during EVAR that has been introduced by Philips and it uses a fiber optic technique in a catheter where actually this is achieved with the fusion of the CT images which is obtained prior to the procedure and then imaging just with the fluoroscopy during the procedure where actually by looking again optical impedance this fiber that is in the catheter can guide the catheter without any difficulties to the desired location as we can see here marking with the ring the origin of the renal arteries and then also easily cannulating with the force technique the contralateral gait of the sten graft and deploying the sten graft in desired location and we can see at the completion angiogram excellent results in this complex anatomy perfect placement of the sten graft and significantly shorter procedure than it would otherwise be the case if we would not have this technology available another exciting field is again with Philips and also Microsoft partnering is holographic imaging during the interventional procedures such as EVAR or T-VAR procedures are concerned where we can see that the individual that's performing the procedure has in front of him all the tools that are necessary in manipulating the image and manipulating the devices during the procedure now there is a very broad potential not only in what I do EVAR, T-VAR and TAVRs but in many other areas where holographic imaging can be used and just recently I have seen this being used in the field of electrophysiology so maybe you can comment a little bit about it. Sure so what has been done is using the you know the holographic imaging what can actually be done is you can actually just by the manipulation of the images itself you can actually significantly you know probably like by a factor of you know five or so decrease the number of ablations that are delivered and you know unfortunately in the field of VP there's kind of a saying that you know with a little bit of an edge to it it says learning as you're burning and and the fact is that you know a lot of times you deliver lesions in a blade in different areas looking for a specific outcome of termination of an arrhythmia or whatever other variable business of interest and so as but of course delivering those lesions does have effect on the underlying substrate you don't want to burn tissue if you if you don't have to but the holographic imaging you actually are much more precise at least in this this was a limited study but in the number of lesions that are required to be delivered and so that's that's a that's a you know that's a very significant improvement in what our patients undergo in terms of the extent of ablation exposure you know this is an interesting question you raise is there a risk for cardiologists radiologists and echospatialists to replace by AI in the near future and there there is no risk we will always need domain experts to take care of these critically ill patients AI is a tool like any other tool and if you don't learn to use it you you do that at your peril very good what about cardiovascular surgery Billy what is your opinion on the use of AI in this field yeah I think so uh decision support I mean AI is critical and we're already sort of doing that we have scoring systems to decide who to operate on and who not to and who's likely to require different things but once once we start uh acquiring big data and doing big analytics we'll be able to do that with much greater accuracy here's some great examples of robotics in surgery and really that by the true definition of robot none of these well some of these have aspects of robotics but the DaVinci surgical uh system the one that most people identify as a surgical robot is not a robot there is no autonomy it is a tele-surgery meaning the surgeon makes all the decisions does all the moves if the surgeon wants to do something very ill advise the robot will let them that said we are right on the cusp of that changing and you may know that johnson and johnson get a large uh interaction with uh google to create verb robotics and now verb robotics is sort of uh being folded into aris robotics it's a complex a lot of big pieces moving on the johnson johnson board but other big companies as well are saying how can we give some element of intelligence uh uh to surgical robotic systems again historically it's just been a tool it does make smaller incisions and you can be incredibly precise in the visualization the 3d visualization through the robotic uh uh console is amazing but what if and this is this was the hypothesis if we actually looked at and deeply analyze the thousand or two thousand robotic procedures done every day in the united states and in europe and gleaned key insights and then use those while we're watching a live surgery go prospectively could we give insights to that surgeon that can make them better faster safer that's an unknown question right now well surgeons will their first uh actionable items say turn that off or will they listen who knows it's certainly going to have an impact on training surgeons though to do robotics um and we're seeing that already why what's good about robotic surgery well the obvious advance of not opening you up not spreading it's much less morbid post-op pains better you get up and around quicker but it takes a lot of training for the surgeon and a very dedicated team that said their teams around the country that are doing a brilliant job with lobectomy with mitral surgery very few people are doing coronary surgery with it but there are some advances advantages to robotics other than the small incision first of all your view you have this beautiful 3d camera that you can put very very close to what you're looking at and so in the console you have this beautiful three-dimensional view that is much better than what you see during open surgery and if you've got a bit of a tremor you can take that tremor out and if you stop and leave your instrument your instruments will stay exactly where you put them so these surgical systems are are brilliantly designed and for those teams that have learned to use them they can get great results that said only three to five percent of surgeries are done robotically now so it's been a wonderful evolution of technology but it hasn't had epidemiological impact yet they don't have them in the third world they don't have them in rural centers and so the next big step in robotics is coming up with ways that it can be globalized and hopefully that's what's going on now with the big robotic organizations now there's a lot of technology that allow you to use robots and intervascular interventions these again are not robotic you are controlling them yourselves one of the big ones is um Magellan uh Zvonka you're going to talk about that I think we're also going to talk about Corindus but uh auras is a robotic bronchoscope and you want to tell us about Magellan that's a fascinating platform right uh I think this was uh very informational uh we're just at the very beginning uh using robotics in well cardiovascular surgery and endovascular interventions and here are a few of the examples uh like Magellan robotics can be used to uh enter into very complex anatomy such as complex renal artery origin occlusion of the internal iliac arteries for a variety of reasons uh this can certainly uh simplify the procedure make the procedure shorter and more reliable one uh particular technology that's very exciting and has been existing for a while it was relatively recently actually acquired by Siemens is Corpath by Corindus and uh this has been used uh extensively uh abroad and less here in the United States in coroner interventions but also in peripheral vascular interventions and here we have an example of a cardiologist in India actually performing the procedure remotely from uh one town to another town where he has a less experienced cardiologist just placing caters and wires and once they are placed there then this particular cardiologist international cardiologist remotely uh is advancing the wires balloons and stents and actually he performed uh in less than uh five hours five different international procedures with excellent outcomes now for us here in United States uh this is uh still an issue and we will discuss this a little bit more in detail later but one of the major issues is uh uh credentialing for that and there is also uh liability issues that we have to take into consideration and uh what if complications occur who's going to uh resolve those complex complex scenarios if the internationalist is not present uh Michael let me ask you a question because you know we keep saying robotics and and that's a conventional term for these type of systems but a robot is supposed to cognate it's supposed to be thinking it's supposed to self-correct it's supposed to learn these systems do none of those it's a person driving and if that person wants to drive something to the wrong position the robot the robot says yeah do it there is no AI there's no learning there's no insight but those things we've seen in other venues self-driving cars the car is looking at everything doing threat analysis looking at pedestrians it's amazing that is true AI how long before we have a robot do you think that you can actually do access and say go out get in the left corner and you would cross your arms and it would navigate it would watch where the wire was going it would wait until it was down the left main and then you could say now go selective in the LAD and get that first diagonal is that an unimaginable future state or is that coming and if it came how how would you accept it as a clinician well uh we know that AI is trying to mimic the cognitive function of humans basically using uh the same process of thinking using logic using common sense and figuring out and solving problems in the most effective way and as you mentioned this already exists this technology exists in driverless cars and so on are we there yet in medicine in cardiovascular medicine we are not but what we are showing here this is early stages of using computers to help and facilitate the process of whatever we do whether it's for diagnostic purposes or interventional purposes like in this particular scenario there is a software that's dedicated to that particular purpose and there is a certain reliability and sensitivity of that software that is very important in being able to achieve those procedures it's not just a movement of devices caters or balloons or surgical instruments but it will also give you a control to prevent certain untoward effects if that is implemented in that particular software and this is where we have to work on and this is where actually simulation has to play a role and we will talk about it shortly but a simulation is certainly important and we know in aviation industry and how simulation we would not be able to have reliability and safety that we have in aviation industry because the pilot fully depends on AI or the computer on the plane to be able to safely guide the plane to a certain destination and do it in the most effective way and threat avoidance you know Maddie I want to ask you the same question if you had something that you said okay go get transeptile access and you stood back and it did it and then you said go ahead and map the whole left atrium and it did it and then you said do a pulmonary vein isolation and it would do it with that would you embrace that or do you think there's just no way it could do it as good as you because I think that it will be able to do it as good as me and someday perhaps better even right now we have some technologies actually you know your to your own company the biosense webster where we are trying for example something called pace mapping that we do all the time in our procedures where you pace from a specific region and you look at the morphology of the paste impulse you know on the 12 lead ECG and you compare it to the clinical btr you know pvc morphology that you have that you're trying to chase and what we do is we try to give it a score we say okay this this paste morphology when I paste from this spot it looks 80 similar 90 percent similar 100 percent similar to what I know is my target clinical based on an EKG that I have you know collected and the the the system in your case the cardio biosense system biosense webster system tells me 97 versus 96 versus 95 percent and this morning just before we came here I had a case that I used that specific precision that that that the technology offered me to actually successfully ablate in a case that otherwise probably would have been extremely challenging and likely it wouldn't have worked I don't feel threatened by it in this case it enabled it and I think that yeah in other situations if it increases the you know the safety for the patient again another example is contact force measurement again from the tip of the blation catheters it tells you you're pressing too hard you're not pressing hard enough you know it if it makes the patient's outcomes better I don't think that the physicians should ever be threatened by that as it's basically saying you know some of these eventually AI to me will be as if it's saying I have the experience of two million cases something I as a physician will never have and so I embrace it I'm going to work with it and you know I'm going to find more innovative things for myself in terms of you know trying to push forward the science so great so uh you know I think when you look at the role that AI is going to play in cardiovascular medicine in the future it's just going to be greater and greater the slope of discovery is exponential so which means in the last five years we progressed like we did in the previous 10 years and that 10 years was very similar to the previous 20 years things are changing so quickly and pieces of the puzzle like seranus like force contact like you just said mapping for ectopic foci all that all that sensor computer aided sensor stuff is now input into this big giant AI data resource that's going to allow us to really understand what the future looks like for any given patient and to be able to do that anywhere in the world so you know it's I think it's it's coming and it's coming fast I think uh what's his name uh Ray Ray Kurzweil said it's coming in 2027 which is only seven years from now but exponential who knows well I certainly hope it will happen that fast but here is a relatively simple presentation of schematic scenario what could potentially happen as far as AI is concerned in cardiovascular medicine of course one of the most important big data that we have is EMR and the issue that we have with EMR at the present time is that they don't necessarily talk to each other so we we definitely need standardization as far as EMR is concerned because for data mining we need EMR to be able to get all the information that is pertinent to certain disease certain patients and then use that information in the best way to be able to add to other components such as imaging that we talked about and then implementing artificial intelligence algorithms using cognitive computing using deep learning and machine learning to advance whatever we do in cardiovascular medicine whether it's in physician in training education whether it's physicians that are in remote locations to guide them and help them in complex procedures where they don't have access to all the technology that we normally have in very sophisticated medical centers and also what's very important for endovascular specialists that would like to gain additional expertise or recredentialing for their procedures now as far as physicians in training are concerned it's important that they undergo supervised learning but before they actually get involved in any interventional complex procedures they should be able to pass the test of simulation and then lead to a hands-on and simulation even though it exists for several decades it's still in very primitive and rudimentally way existing at the present time and the reason for it is that it's basically driven by industry needs to promote their product but it's not designed in a true sense for extensive educational purposes and what I mean by that is that simulation in cardiovascular field does not have all the components that exist for instance in aviation industry where the pilots in training going through simulation have a variety of options how to deal with certain complexities whether it's a loss of engine control loss of hydraulics whether it's fire whether it's a sudden drop in the plane altitude for whatever reason or a hydraulic failure of any kind and so on we do not have that in our education and that is something that needs to be developed and used more extensively for better patient care and then as I mentioned remote procedural guidance it exists to a certain degree but in a very primitive way and particularly nowadays with we live in a COVID era god knows for how long and it's almost impossible to organize courses with hands-on experience from physicians coming to institutions with symbols of excellence to train them in the procedure so remote guidance will be extremely extremely important and we have to explore into that field using artificial intelligence and again credentialing and recredentialing is another aspect particularly now where we have to do it in our own institution with some kind of remote guidance and all of it I have no doubt with the use of artificial intelligence will lead to improved patient care and better outcomes you know that's one of the things I've heard that's going to be a big indication for things like Watson IBM or anything that can parse student speech is taking over the whole oral examination field because now you can sit down with an examiner and if they've had a bad day or they have some bias they don't like the way you look that can change the way the whole thing goes or they can just fortunately take you just into superficial stuff you know or take you very deeply into stuff you don't know it's very capricious the outcome of an oral exam not with leveraging AI and natural language parsing and so they think they'll be able to objectify that so you've said challenges in AI medicine there are a lot of them and let's go through those because I think that's a hot hot we already touched on some of them and not only in the cardiology not only in the interventional field but also in the cardiovascular and vascular surgery but one of the greatest challenge in the present time is truly data integration and I mentioned as far as EMR data is concerned also imaging data is concerned and how to integrate in a critical flow is it's still a challenge for most of you know when we're talking AI there's weak AI and they're strong AI and those are terms using AI weak AI is very rules-based simple task if you wanted to integrate data you could use weak AI strong AI uses the ability to present to face problems that it hasn't been prepped for so strong AI you don't need to integrate the data it'll sit down it'll comb through it it'll make sense out of it and put it in the format that it likes and that's coming very very soon so trying to get EMRs to talk to each other may be something that we don't need to worry about because very soon the AI will be able to navigate a totally unfamiliar data set with totally unfamiliar data fields parsing natural human speech will be able to make sense out of it and put it no you know the idea of somebody having to input data just right will be a problem of the past absolutely correct and that leads to this point of interpretation understanding machine learning models actually machine learning is becoming so sophisticated that the computers and the software and algorithms can already predict what is the best and what is the most optimal way to analyze the data but obviously we have to make sure that we're using the most appropriate algorithms and we are inputting the data that is essential and we have to be able to create machine learning models they are only as good as the training data and you mentioned Billy that already we have to make sure that the appropriate and most important data is included and that data is used in the best and most optimal way now what's also very important is of course no computers are perfect no machine learning or AI algorithms are perfect and then my misdiagnosis can occur and again that is obviously related to not understanding machine learning models and not being able to put a good quality training data in those machine learning models so we always have to carefully analyze the data and to see if there was any bias either by the operator that included that data in analysis or by that particular AI algorithm that led to misdiagnosis and you know one of the challenges if I may with that is that there's two different kinds of AI there's supervised learning and unsupervised learning and unsupervised learning which is strong AI it comes up it digs deep into the data and makes its own associations you don't you don't control that so it may come up with correlations and insights that we've always missed the problem with that is how it came about its decisions is opaque to you they call that black box AI and in the banking industry for example when someone applies for credit or a loan if you turn them down you have to be able to say why it can't be just I just don't like you when you when they brought AI into the banking business sometimes there'll be something that everybody assumed they would get and the and the AI says no this is not a good loan we're not going to give them credit and you kind of try to query the AI and say why and you can't find it so you may get some bad diagnoses and study knows may teach us a lot about AI that aren't such bad diagnoses we think they're bad diagnoses but the computer had insights that we couldn't follow because it was you know black box AI mining insights that we don't have and finally we have regulatory challenges which are always challenges in regardless of what we do in cardiovascular medicine and we have to be able to convince regulatory agencies to work with us on the new AI models and how to implement them in a most appropriate way there are also social and legal implications and we talked about malpractice in remote proctoring or remote performance of the procedures that will happen in the future and that has to be that has to be addressed which has not been resolved the foremost part particularly not in the United States so those are the issue of major concern that will slow down the whole process of implementation of AI in cardiovascular medicine because I wanted to end this program with a very optimistic note and I am quoting Sunder Pichai CEO of Google who stated that AI is one of the most important things humanity is working on it is more profound he says than electricity or fire so that's true optimism and I hope it materializes that's a great great quote to end on electricity yeah fire no very good so gentlemen thank you very much for the opportunity to join me for another educational program at Texas Heart Institute on innovative technologies and techniques it was a pleasure discussing this subject with you thank you Zvanko see you thank you thank you