 Good morning. Today I wanted to talk to you about how we can use augmented intelligence as a co-pilot in the operating room. My name is Dr. Zane Kelpie and I'm in Northwest Medical Center in Tucson and Arizona and I have no disclosures. The purpose of changing the way we do things in the operating room and bringing it to the 21st century has to be with absorbing the technology that we're using but also use the technology to our advantage as a form of a co-pilot to make better decisions and so I've always felt that the gestalt of our education and the experience, the 10,000 hours, that really wasn't going to be good enough. So I thought we could harness the power of the technology such as EEGs, waveforms, nearest monitoring, our ventilators and our eco-machines in cut-ex surgery to try to make better decisions and what I'm going to describe is how we've harnessed all of this and put it into our AI lab and essentially drilled down on key decisions to make it real for us to make better decisions in our operating room and really the purpose is to improve our decision-making and most of the trauma that we inflict upon patients happens in the operating room and when we try to do no harm I think we should try to predict to do no harm as a prelude for getting better outcomes and I think this will lead us to better personalised or social care and I think that's the key message that I want to talk to you about today and what I'm going to talk about really is how we built this AI powered ecosystem to make decisions very close to real time so we can influence what's going on with each patient in the operating room and that personalises our surgical care. So the things that we thought about was well what's the flow like of a patient seeing coronary artery bypass surgery and can we minimise some of the complications so for example if we use EEGs can we predict or influence or diagnose neuro-inflammation or to reduce things like post-pump center or could we use the dopplers to identify tissue resistance to look at graft patency could we use Nero's monitoring and put that all together to influence and see what happens whether you're pulsatile or non-pulsatile and does that actually influence the factors for renal failure. Of course there's labs that are available and we send that off every time in the operating room along with putting our patients on ventilators so could we use the ventilator settings in a more novel way to reduce post-operative information which the bypass circuit actually inflicts upon our patients. You know one of the other biggest tools that we have in the operating room for cardiac surgery is our echo well could we use the echo in a better way to predict heart recovery or revascularization or regional wall abnormalities and you know the other thing is that we send tissue off to our buyback which we then do metabolomics and genomics on so how about if we take these ecosystem this ecosystem and we stack all these features up and I'll show you how we do that to make a better decision for the patients and going into each one more deeply I think EEGs can actually if we look at the sequence of alpha beta theta and even gamma ray gamma waves of patients who are on bypass we can actually predict strokes not just looking at their cerebral sat because by that time it's a bit too late can we actually predict that something is happening and the waveforms are a powerful tool to help us decide that by looking at the preamble of ischemia with the waveforms but also neuro inflammation and you know could we do something in the operating room to help prevent that so that's what one of these this tool is for. The other thing that we're using is we're looking at pulsatility indices when we're looking at graphs during bypass surgery and we look at the flow we look at the the definition really of pulsatile indices can be derived to give you an idea of what what's going on with tissue resistance and tissue compliance when we you know when we you know plug in a graft into that into the irrigation system of the coronary arteries it's very heterogeneous because you can't you know you hear from the cardiologists that you've got 80 per cent stenosis well yes pi r4 but to the power of four gives you your flow and your resistance and your diameter across the the valve across the the coronary artery but truly what is the longevity going to depend on of that graft and if we're able to change that at the time of when we put the graft in or put it at a place where we feel we could have better downstream effects you know there's many times that we sometimes put our grafts on to coronary arteries that are actually clogged up and the graft fails and so that is highly irritating as a as a cardiac surgeon so we're working on using pulsatile indices to do that and how can we change the tissue resistance when we've been using tmr transmicardial laser revascularization which increases veg f which allows us to increase the um microvascular status and most of the problems that happen with people who are in heart failure or have scar or are diabetic the tissue resistance and so can we change that so we're evaluating that in the operating room and we're trying to predict the errors in which we can do this much better um with echo we're using myocardial work and this algorithm is a relationship that we have with um ge and we basically look at our echoes before and after say revascularization and we follow the concept of myocardial work using tissue Doppler and we do this in real time in the operating room so we actually know when we've been successful at revascularizing the patient because we look at the regions to look at myocardial work and we also uh calculate the regional global strain which is um really powerful because we're also looking at building heart recovery and global strain essentially determines um you know uh strain which is basically squeeze in vertical horizontal and twist um of a muscle that's squeezing and we can look at how much load there is on that muscle and that changes whether you're doing a valve uh in mitral valve surgery to change the pressures such as the left ventricular endostatic pressure or you're looking at um a revascularization for a muscle that you want to see improve and we've got uh papers submitted already which actually shows our heart recovery program on patients that are getting offloaded with uh the impeller and we use a surgical impeller the impeller 55 in the aortic position uh with open heart surgery in sick patients who have ejection fractions less than 35 for 25 percent and there are the undergoing bypassal valve surgery um and we evaluate how we wean our patients to recovery uh using this process with the impeller every single day without weaning protocol and uh the initial tool in the baseline for this is in the operating one of the interesting projects that we've also got is using near monitoring to measure cerebral oxygenation which of course everybody everybody in the OR has but what about if we slap that on to the kidneys and if we do that we're able to look at um kind of oxygen delivery and uh that oxygen delivery uh is what we calculate from the perfusion machine uh before during and after bypass but also if we change the modes we are looking at this pulsatile flow have a difference on oxygen delivery this reduce your renal failure and so what we're also doing is we're looking at urinary um i g f b binding protein 7 and tissue factor inhibitor which is an m m p 2 which essentially predicts using elysis at 0 12 24 and 7 48 and 72 hours is this patient going to get aki or cute kidney injury before they actually do and can we intervene much earlier from three hours before we even know and so we're doing this with near monitoring in the operating room uh and looking at oxygen delivery with our patients and that was a really cool project we embarked on uh to see whether manateau would actually have an effect and uh our first student from texas heart first perfusion student victor mendosa uh this is his project and what we're doing with our ai lab is we're looking at answering the question of whether manateau um actually he has an effect and uh on our kidneys or was it just a placebo or is it something that we just do it to make ourselves feels better well we actually developed a protocol what he did um uh to design that to answer that question does manateau have an effect and is it hurting us causing a cute kidney injury or helping us uh with the free radical scavenging properties of manateau as we all seem to think and this is a heat map which looks at all the different parameters that we've got uh evaluating a cute kidney injury and um we're looking at uh intra pre and postoperative um stories along with the morbidity sorry the demographics of the patient and we we look at positive and negative coronary factors which bear relevance significant and so um um we i'm not going to take his thunder away but the key message that um we found is that what we give in terms of net volume um and crystalloids as well as ejection fraction all have significance uh in whether or not we give manateau uh in his study of uh before and after bypass as we come off the cross clamp or just as a cardioplegia in the cardioplegia solution and are we making an effect so we looked at two different doses and we found that there were mitigating factors which were positively and negatively correlated and the nearest monitoring actually did help us um decide and uncovered a lot of what we found to be true and I know that the purpose of me discussing this is not to take his thunder away but is to show you how complex um our relationships can be and what augmented intelligence or artificial intelligence unravels uh is the complexity of the relationship uh which is hidden and can have significance for example knowing what the pre-hydration statuses of a patient um has a positive relevance we all know that cross clamp does as well which is there but the ultra-filtrate volume as well as sodium levels are also important along with bypass times and that coupled with other things that we do uh you know when you're on a bypass circuit uh we often have uh uh inflicted as I said earlier a lot of information that information is um can be quite damaging same patients who are already pro-inflammatory such as heart failure patients so we devised a way to induce a form of um uh vagal stimulation which would allow while we're on bypass to keep the ventilator going uh but at a much reduced rate and a ratio that is akin to meditation while you're sleeping while you're on the bypass machine and um there's many studies that show that heart rate variability actually increases and there are many studies that actually show that there's a link between heart rate variability and inflammation and so what we decided was well can we quantify this and we're looking at IL-6 and inflammatory markers pre and post surgery uh but we uh and pre and post bypass specifically and people have been ventilated not ventilated and what we have found is also interesting in that our HRV does increase uh which is heart rate variability and we calculate that with various different waveforms not just the EKG um but looking at different ratios with high frequency and low frequency ratios but we also found that the burden of inflammation is much reduced in these patients and that's unpublished right now um you know as we also look at uh our relationship with people like at the Broad Institute um very good guy called Penit who we are looking at uh giving tissue uh from our Biobank from each patient we are trying to look at the genomics but genomics for me is also a little bit too slow so uh you know why would I in it's great that I can influence something that'll change the left ventricular mass in a year's time after I've done a valve and we all know this lb mass regression when we do a ordered valve replacements or mitral valve repairs but um how about if I try to use the tissue to try to get uh data that's much more important such as proteomics and genomics and I'm sorry metabolomics so we're looking at specific pathways particularly in mitochondrial protection um which when we give cardioplegia would allow us to change the real-time status of ATP in a heart that stopped or is beating um and we're using that data to actually make better decisions well it's great we've gone through all of these things but how can we actually tie this together well we're actually um streaming this data from the OR picture that I showed you at the beginning of the case uh beginning of the presentation to uh show that we've got these waveforms but also these static demographics which go through the electronic health record through hl7 feeds into our ai lab and where we've got a data warehouse and in that data warehouse we're using a lot of analytical tools we've got relationships with data robot and with google and it's on that platform uh and amazon web surface so we've it's all under hipper firewalls and um we are trying to develop predictive risk scores predictive scores for stroke predictive scores for um morbidity and mortality because frankly speaking there's sts risk score which you see in every h&p before surgery of a patient is frankly crap it's predictive at all it doesn't tell you anything um and it vaguely simulates a story of risk which is not truly accurate and I feel uncomfortable telling my patients that they've got a 1.4 percent chance of um mortality and which is frankly not true and so it um kind of emboldened us to try to do something a little different which is what led to this whole idea and this project um with trying to then make real-time decisions in the operating room and this just summarizes kind of the story of what we've got from our operating room and by the way we've got similar formulae in the ICU and in our step-down unit um with all these monitors that we're working with neon coden and it streams to the lab so we've got petabytes of data from our electronic health records with medications our op reports our lab results the nephro checks I talked about earlier um device data from our monitors our perfusion system um is feeding into our G9 monitors in the operating room where we can see the positive flow and so now we're doing time series analysis and we're using our waveforms from our EKGs, EEGs and our ventilators and we're looking at data from our echoes and also pre-operatively from our CT scans and our ultrasounds and our angiograms and so this gets fed into our data lake and where there's a lot of unstructured data and as we do that in our cloud in the lab we drill down on projects say for example looking at graph papency we're looking at creating a data mark where we develop our algorithms for positivity index and we then say okay well how can we stack all of this information um that we've got of this one patient in from multiple modes into a neural network and we're using various algorithms such as uh picarat various uh neural networks which uses static and dynamic data and we're then able to personalize a predictive score and this video which is four minutes long summarizes the future of what we're doing so i thought i'd show it to you to end the presentation before i ask for any questions as we stand here at the dawn of the fourth industrial revolution the potential of ai is evident but the value is yet to be unlocked for so many people like you seekers solvers with unrivaled expertise in your field you're driven by curiosity but where do you start the race is on to bring the power of machine intelligence to as many innovators as possible our unique machine brain was purpose built for continuous learning by 600 of the best and brightest data scientists and engineers all united behind the idea that ai won't reach its true potential until the power of ai is available to all so how do we do it begins with pattern recognition in a learning session rapidly accelerating your model and blueprint development through our chaplain challenger approach that ranks your models to help you find the best insights for your desired outcomes but models are fueled by data and to put it simply data's experience your organization's experience this needs to be transformed rapidly into actionable insights while remaining secure and within your control of all times regardless of where your data is data robot is able to help you harness that knowledge on the prem or in the cloud structured or unstructured text numeric geospatial visual and normal but unlocking transformational growth isn't possible till you can scale models into production with confidence enabling continuous learning and real-time improvement as a result your learning session gets smarter over time but as the data robot learning core evolves those learnings are shared with all of the sessions increasing the collective knowledge of the entire machine and what does it look like when all of these sessions are running in real time let's take a look in an effort to inspire our r&d product customers and partners we created a real time 3d visualization of our data robot machine break visualizing data helps humans process and make decisions more confidently visualizing in three dimensions over time unlocks even greater ways to explain insights and drive action and seeing the entire dynamic system in concert shines a light on the real-time potential of the human and machine intelligence revolution it can be a bit overwhelming at the scale but data robot handles all this complexity so you don't have to freeing up human resources to focus on the most challenging and interesting problems while enabling better trusted critical decisions empowering citizen data scientists to be the human heroes of the machine intelligence revolution being the impact of ai out of the box and into our lives helping regular people through revolutionary advances in health finance climate government and so much more they are the human heart beating within our machine there is nothing artificial about the problems of the real world so should we settle for just artificial intelligence would you trust your doctor if you practice artificial medicine would you trust me to perform surgery on your heart if the experience ahead is artificial well i'm not a voiceover artist i'm a heart surgeon today when i make decisions on my patients i use their lab data their images and their waveform data but this is too slow tomorrow we as physicians need to make better faster human decisions decisions we can trust but we can do this by leveraging the power of augmented intelligence and if we do so then we would be able to predict medical interventions and deploy preventative strategies in healthcare you know the impact of this on human lives if we get this right we'll be immeasurable i'm excited to work with data robot because together we're able to make augmented medical decision making this is a new formula which will be the future of healthcare driven by ai i'm dr zane calpy and i'm a citizen data scientist this relationship is important because um i think industry drives the future of healthcare and we've got to embrace it because i'm happy to announce that we we are also um of the first partner in a health incubator with data robot where we're able to push forward the new way of thinking and i hope this presentation opens your mind to new ideas that we do and for things that we do every day so with that i'm very happy to take questions thank you very much