Latest / CXOTalk / The Cardiovascular System, Mapped in Code as a Digital Twin | CXOTalk #901
Transcript
- 0:00Today we go inside a research lab building cardiac digital
- 0:05twins, a computational replica of your heart that could
- 0:09transform personalized medicine and how doctors treat heart
- 0:14failure. Our guest is Doctor Joe
- 0:17Alexander, director of the Medical and Health Informatics
- 0:21lab from NTT Research. I'm Michael Krigsman, welcome to
- 0:25a fascinating episode #902 of CXO Talk.
- 0:31The medical and health informatics lab belongs to NTT
- 0:35Japan and that company put a Bell Labs type innovation
- 0:40company in Silicon Valley called NTT Research.
- 0:44The original 3 labs for entity research are physics lab, a
- 0:49cryptography group and then my group which is medical and
- 0:52health informatics. We work on kind of moon shot
- 0:56ideas. The physics group is working on
- 0:58quantum computing, cryptography is working on even post quantum
- 1:03cryptography and my group is working on bio digital twins.
- 1:08How did you end up where you are now?
- 1:11Despite growing up poor in Alabama during the civil rights
- 1:14movement, I found myself in the chemical engineering program at
- 1:19Aubrey University. One once at a summer job, I was
- 1:24on a hard, hard oil explosion, and as a burn patient, that's
- 1:28what introduced me to medicine. The physicians who took care of
- 1:31me were from Johns Hopkins, as it turned out.
- 1:34So I went back to Auburn, said I want to go to medical school, I
- 1:37want to go to Johns Hopkins. And somehow that worked out.
- 1:41And Johns Hopkins is where I learned about biomedical
- 1:44engineering and cardiovascular dynamics.
- 1:47And so I did a combined MDPHD program at Johns Hopkins.
- 1:52You just came out with, hey, I want to go to Johns Hopkins.
- 1:56And you know, what was their reaction?
- 1:59It's like not a common request. So I had been purely in the
- 2:03chemical engineering program. So I went to the pre
- 2:05professional advisory committee and gave that very specific
- 2:10request and they laughed at me and said we'd never send anyone
- 2:13to Johns Hopkins. But as it turned out, things
- 2:18worked out for me. Tell us about the work that
- 2:24you're doing. What is a biological digital
- 2:29twin? We take our examples mostly from
- 2:31the aviation, commercial aviation industry where, you
- 2:35know, they build a mathematical representation of every jet
- 2:39engine. When that engine is flying in
- 2:42the air, there's a replica on the ground that's taking note of
- 2:46all kinds of sensors that are recording information in flight
- 2:50and mathematically incorporating those into the model.
- 2:53And what that does for commercial aviation is that it
- 2:56makes it possible to do predictive maintenance of the
- 2:59engines. You don't have to have a
- 3:00scheduled maintenance. The computer simulation tells
- 3:04you when it's time to maintain what.
- 3:06And so analogously, you know, we thought that if patients are
- 3:10appropriately instrumented, we had enough data from wearable
- 3:14technologies or elsewhere, we could predict when a patient
- 3:17needs some kind of intervention. And that kind of set us on the
- 3:22road to predictive health maintenance and the building of
- 3:25a biological bio digital twin. So the inspiration then came
- 3:31from aerospace. That's correct.
- 3:34Can you talk about the clinical problems that you're actually
- 3:39trying to solve? And then we'll get into the
- 3:41discussion, the complexities around building a biological
- 3:45digital twin. The core problem is that we want
- 3:49to improve cardiovascular outcomes because cardiovascular
- 3:53disease is #1 killer globally, accounting for about 20 million
- 3:58lives in the year 2022. And the way we want to go about
- 4:06it is to make use of mathematical modelling and
- 4:09simulation to tailor a model that's specific to an individual
- 4:15so that that patient can have the very best outcomes.
- 4:18Currently we rely on clinical trials that are population level
- 4:22data and usually there are many populations that don't get
- 4:28included in the clinical trials and we're not totally sure that
- 4:32the results apply. When you talk about tailoring
- 4:36specific results to a specific patient, generally as you
- 4:42indicated, you do large population studies and then take
- 4:47a solution like a like a medication and apply it to a
- 4:50large group, but you're drilling down.
- 4:52Can you talk about that? We use models that are causal,
- 4:58that we understand mechanisms for, that are physiological, so
- 5:02that we can have insights into how those models behave.
- 5:06We found that for us, rather than go to sophisticated models
- 5:11that have to do with computational fluid dynamics and
- 5:14finite element type analysis, which are computationally
- 5:18expensive, we develop simple models that allow insights.
- 5:24And our particular structure of model is called an electrical
- 5:27analogue model, meaning that the flow of blood is analogous to
- 5:31the flow of ions. Pressures are analogous to
- 5:36voltages, valves are analogous to diode, and we put together a
- 5:43structure representing the entire circulatory system and
- 5:46try to match the parameters in that model to an individual
- 5:50patient. So it's not just a model of the
- 5:55heart, but it's an entire cardiovascular model.
- 5:59Is that a correct way to put it? That's correct.
- 6:01It's an entire cardiovascular model where we have more
- 6:05specific models for the heart and the coronary circulation.
- 6:09And that cardiovascular model also has the necessary
- 6:13components to do what we want to do at the moment, which is to
- 6:17manage acute myocardial infarction and acute heart
- 6:20failure. And what that means is that that
- 6:23same simple cardiovascular system has to have lungs, it has
- 6:28to have kidneys, it has to have neural regulation.
- 6:31So it's a very broad based model that you're creating so that so
- 6:36it's far more complex than simply creating a simulation of
- 6:40the heart, for example. That's correct.
- 6:42Can you drill down into your vision of the digital twin and
- 6:50of the the way that you're designing the solution?
- 6:53Give us some insight into what you're doing, how you're doing
- 6:57it, that the challenges and so forth.
- 6:59One of the easiest ways to consider the kind of advantages
- 7:04of this sort of approach is a use case, which is autonomous
- 7:09closed loop intervention system, which you mentioned in the
- 7:12introduction. This is a situation where we're
- 7:15trying to introduce a system that manages acute myocardial
- 7:21infarction and acute heart failure.
- 7:23After the patient comes to the emergency room, goes to the
- 7:26cardiac catheterization lab, and the culprit lesion is relieved,
- 7:30then they're brought to the CCU or ICU and they're treated with
- 7:34certain drugs to stabilize the patient and have the heart rest
- 7:40and recover from having had that, let's say, heart attack.
- 7:44Those drugs consist usually of something we call catecholamines
- 7:49to help the heart contract, nitrates to release or to reduce
- 7:53blood pressure, diuretics to remove blood volume.
- 7:57These things are given serially by a physician who monitors
- 8:01patients response. What we're doing with the
- 8:05autonomous close loop intervention system of what we
- 8:07aim to do is to have a system where you just type into the
- 8:12keyboard the desired left atrial pressure, cardiac output,
- 8:15systemic arterial pressure, and then a system of syringes filled
- 8:20with each of the necessary agents automatically delivers
- 8:23the amount of drug that's necessary to achieve those
- 8:27desired endpoints. Then we monitor the patient's
- 8:30response, feed that back to our system, and depending on the
- 8:34error between what we told the machine, the syringes to give,
- 8:39and the patient's response, then we make an adjustment.
- 8:42This is a feedback control system for delivering care
- 8:47that's specified at the keyboard in terms of desired left atrial
- 8:52pressure, cardiac output and arterial pressure.
- 8:54And how is this different from existing models, right?
- 9:01I mean, there are machines that deliver drugs, so how?
- 9:05What makes this unique? Anesthesiology is a field where
- 9:10there's some level of automation in the delivery of drugs to
- 9:15maintain blood pressure in patients who are undergoing
- 9:18operations. The thing that we're doing that
- 9:23no one else has done and it's very challenging thing, is to
- 9:30control not just blood pressure, but cardiac output and left
- 9:35atrial pressure and to do these in such a way as to minimize
- 9:40myocardial oxygen consumption. So the whole point is that the
- 9:44heart needs to relax and recover, yet the periphery needs
- 9:47to be perfused. And so if we can let the heart
- 9:52work the least possible while achieving all these things, then
- 9:58that's the best situation for myocardial recovery.
- 10:03So minimization of myocardial oxygen consumption is the
- 10:07special thing that no one has been able to achieve.
- 10:11And what are you doing that you hope will allow you to achieve
- 10:16this? We have the benefit of some
- 10:20consultants and some work that gives us a way to monitor and
- 10:26quickly measure myocardial oxygen consumption live.
- 10:32And we also have some physiological construct based on
- 10:37cardiac output venous return surface that lets us do a good
- 10:40job of maintaining left atrial pressure at a specific range.
- 10:46At same time, we regulate right atrial pressure.
- 10:49So the combination of the cardiac venous return surface
- 10:54and technologies to instantaneously track or
- 10:58periodically track monocardial auto consumption are some of the
- 11:02benefits. But there's some other
- 11:04techniques that we have as well. Folks, you can ask questions.
- 11:09If you're watching on LinkedIn, pop your questions into the
- 11:13LinkedIn chat. If you're watching on Twitter X,
- 11:17use the hashtag CXO Talk and ask Doctor Alexander pretty much
- 11:23whatever you want. Now is a special opportunity to
- 11:26to do that. Joe, how does what you were just
- 11:30describing then relate back to the digital twin?
- 11:37My lab is interested in the human digital twin.
- 11:41We started with cardiovascular system because we have expertise
- 11:45and it's #1 killer global and we thought it's a good place to
- 11:49start. As well as the fact that the
- 11:50cardiovascular system even attacking something like keep
- 11:55myocardial infarction, it makes it necessary to consider other
- 11:59organ systems like the lungs and the kidneys and and
- 12:02neuroregulation, this autonomous closed loop intervention system.
- 12:07The first guess is given by our electrical analogue model of the
- 12:13human cardiovascular system that best fits a particular patient
- 12:18who walks in. Now how do we determine the best
- 12:21electrical model for the patient that we see lying in front of
- 12:25us? That relies on huge amounts of
- 12:30population level data to start with, where we build a virtual
- 12:33population of patients that are similar to the patient that
- 12:39walks in, in terms of you know, age and conditions,
- 12:43comorbidities, etcetera. We did we build a population of
- 12:48virtual population similar to that patient.
- 12:50Then based on a specific characteristic we narrow in on
- 12:53the best guess and that might be several 100 patients that are
- 12:58close to this patient. And we use that to get the
- 13:02initial first guess on this closed loop, autonomous closed
- 13:05loop intervention system. Then because it's a feedback
- 13:08system, it will automatically adjust those parameters to give
- 13:13the specified low Tatro pressure, cardiac output,
- 13:15arterial pressure. Now mind you, that information
- 13:19in the error signal, the correction signal that's being
- 13:22fed back, that teaches the model of that patient what those
- 13:29parameters should be. So the model becomes more and
- 13:32more specific for that patient as therapy is administered
- 13:38through this autonomous close dependent pension system.
- 13:40So it both zeroes in on the best parameter choices and at the
- 13:46same time teaches us what the best parameter choice choices
- 13:49are for that patient. So we will create a registry
- 13:52including that patient and those choices.
- 13:54So you are gathering broad population data and then
- 14:01isolating out of that broad group for each individual
- 14:06patient the right subset of data that most closely corresponds to
- 14:14that particular patient's situation.
- 14:18And then from there you're building the digital twin of
- 14:23that specific patient. Is is that a correct way to say
- 14:26it? That's very correct.
- 14:28We think of the digital twin as a system of characteristics that
- 14:34represent a patient, characteristics like resistors,
- 14:37capacitors, dialogues and and so forth.
- 14:40So we want the those parameters, the best choices that most
- 14:45represent that patient and that patient's characteristics.
- 14:48And so we begin with population level data that's in the
- 14:51literature. And from there, we build a
- 14:54virtual population that is relevant for this particular
- 14:57patient. And from within that virtual
- 15:00population, we find the closest patients for that patient.
- 15:05Then we begin this process of driving the autonomous closest
- 15:08intervention system. In the course of that feedback
- 15:11regulation and adjustment, then we learn to titrate or tune
- 15:17those parameters to match that particular patient.
- 15:21What kinds of data are you using as inputs into all of this?
- 15:29In the world of cardiovascular about digital twins, there are
- 15:32many places we could have started.
- 15:34We started with acute myocardial in parks and acute heart
- 15:38failure, mostly because it's acute and mostly because the
- 15:42patients will be in the ICU or CCU and heavily monitored and
- 15:48heavily instrumented where we have readings of dynamic signals
- 15:53like dynamic pressure pressures and volumes and we have
- 15:59echocardiography. So we have a lot of time
- 16:03dependent signals that we can use to train our models and
- 16:09patient and set those parameters and and gather data.
- 16:13So the inputs in the ICUCCU are, you know, measurements of left
- 16:20atrial pressure, right atrial pressure, cardiac output.
- 16:23We can do SWAN catheters, we can monitor pressure over periods of
- 16:29time. And usually the admission for
- 16:31that patient is about four or five days.
- 16:35And so we can have four or five days of acute information to
- 16:40tune to that particular patient. So we chose the easiest option
- 16:45to be honest as a place to begin.
- 16:48When you go to chronic heart failure, you have to rely on
- 16:51data that you know from patients that are not in the hospital,
- 16:57and they're static measurements and they're based on wearable
- 17:01technologies, not nearly invasive enough to parameterize
- 17:05such a system. But we'll go to chronic heart
- 17:08failure next. Where are you in the process of
- 17:13this set of developments? We have a group of
- 17:16collaborators. The ones most relevant to what
- 17:21we're talking about today are a group at National Cardiovascular
- 17:24Center in Japan, where they have cardiologist, anesthesiologist,
- 17:31experimentalist and are able to do animal experiments with
- 17:36animals ranging in size from from mice all the way up to
- 17:41cows. And so we're doing animal
- 17:44experimentation to validate these approaches and test out
- 17:47these technologies. And last year, for the first
- 17:51time ever, we demonstrated the proper functioning of an
- 17:56autonomous closed loop intervention system to treat
- 17:59heart failure in an experimental animal, a dog that that had
- 18:05experimentally induced heart failure and the system
- 18:08compensated and automatically treated that dog while
- 18:12minimizing myocardial oxygen consumption.
- 18:16We're at a stage now where we're collecting a number of
- 18:19experiments from animals where we'll prove, we intend to prove
- 18:24that the autonomous close up intervention system outperforms
- 18:29the standard of care. By showing that these
- 18:33cardiologists and their treatment compared to the
- 18:36autonomous system cannot yield the such a small myocardial
- 18:42infarct size as what we can get from the autonomous close up
- 18:45intervention system. So reduction in infarct size
- 18:49would be physical evidence of how this system could outperform
- 18:55clinicians and that's what we intend to do.
- 18:58The autonomous system then is relying on this general
- 19:04population data plus very detailed data coming from the
- 19:09particular patient who is in the ICU or the CCU, because there's
- 19:18lots of data and you're finding so far are a greater degree of
- 19:24accuracy and accuracy in what specifically that is producing
- 19:29the result. We are still in the very early
- 19:32stages and developing you know what we can do in the real life
- 19:37clinical situation. So we're all of what I'm saying
- 19:41is based on hopes that depend on what we've been able to validate
- 19:45then verify in animal experiment.
- 19:48So we're still at the animal experimentation phase and so we
- 19:53are not looking at population level data in animals.
- 19:57We're even beginning without that kind of referenced model.
- 20:07Of parameters for animals, except what we can get from the
- 20:10literature very simply and still this autonomous system is able
- 20:14to perform. We demonstrated last year.
- 20:18So, but your description of the process is correct.
- 20:23That's what we hope to achieve and there are many challenges as
- 20:26you can imagine. Again, we're at the early stages
- 20:30achieving this in animals and before we get anywhere near
- 20:35regulatory approval, we have to go through and I'm sure you're
- 20:39aware of these types of hurdles. We have to use this kind of
- 20:43system to do clinical decision support before we can go to
- 20:47complete autonomy. And so that means the feedback
- 20:51would be going through a physician, the physician would
- 20:54be in this closed loop and we advised the physician and the
- 20:57physician would make a decision whether to do what the software
- 21:01says or not. So we have many things to
- 21:04achieve before we before I can really answer your question
- 21:09well. I guess similar to autonomous
- 21:13vehicles, before you let a fleet of autonomous taxis on the road,
- 21:18you have a driver in the car because you really don't know
- 21:22what's going to happen until you die.
- 21:24It try it because it's new. So it's basically the same
- 21:27concept. That's exactly right.
- 21:29So some people have suggested that we not call it autonomous
- 21:32intervention, but like a copilot system.
- 21:35But we're ambitious in thinking that this kind of a system can
- 21:40replace the need for a specialist to be at the
- 21:44patient's bedside. It could be a system that's
- 21:46monitored by, you know, another clinical professional, but may
- 21:52not need, probably will not need a specialist, which makes it
- 21:59accessible to kind of remote hospitals with fewer resources.
- 22:04And so that's part of our vision for this, this HealthEquity.
- 22:11We have an important question from Twitter, from Arsalan Khan,
- 22:16who's a regular listener. He always asks excellent
- 22:19questions and he says this. It's this is a very good way
- 22:24into getting to to dive into the ethical and equity aspects of
- 22:30this. He says this How do you make
- 22:33sure data from these medical instruments is correct and the
- 22:38doctor makes the right decision based on the data versus his or
- 22:44her personal experience? My bias to be honest, and one
- 22:49clinician made this comment. He said that he'd rather have an
- 22:53autonomous system that focusing beat to beat on adjusting meds
- 23:00for a patient than to have an exhausted resident standing in
- 23:05the hallway checking his phone and doesn't have enough time to
- 23:08look in on the patient. We had the bias that if the
- 23:12system is built correctly, then it will correct even for errors
- 23:17that the position might make. But it's a, it's necessary like
- 23:23you described for levels of autonomy in vehicles to have
- 23:30that supervision. FDA is not going to approve an
- 23:33autonomous device no matter, no matter how well it's tested
- 23:36without that step of clinical decision support.
- 23:40And so we we have to do that. We can't take the position out
- 23:43of the loop. And then we have another related
- 23:47question from Elizabeth Shaw, who says how do you know you can
- 23:51trust the model and then the decisions in healthcare, the
- 23:56standard is let AI guide you, but the doctor should make the
- 23:59decision. Yes, it's related.
- 24:01Our first pass is to include the doctor in loop so the doctor can
- 24:06judge, you know, what the the system is telling him, him or
- 24:14her. And let me just say that the
- 24:16thing about our system that's different from AI is that it's
- 24:20causally based. Even though cardiologists don't
- 24:23do as much Physiology as they once did, the decisions that
- 24:28would be coming out of our system will be based on clinical
- 24:33practice, cardiovascular Physiology, cardiovascular
- 24:37dynamics, and it won't be a black box deep learning type
- 24:45output that cannot be reviewed and and criticized.
- 24:50I'm a technologist and I absolutely support AI and
- 24:55autonomous systems when they're used properly with the right
- 24:58type of governance. But if we look for example, at
- 25:02the Boeing 737 Max crashes that were caused by faulty sensor
- 25:10algorithms and then incorrectly documented or incompletely
- 25:15documented recovery procedures, it's an example of an autonomous
- 25:22life and death system. And, and so the idea of of
- 25:28having this kind of autonomous system relating to the to
- 25:31medical medicine is kind of scary.
- 25:34There's certainly lessons to be be learned, and some of the some
- 25:37of the lessons are avoiding shortcuts and some of the
- 25:41lessons have to do with hubris. Certainly fail safe mechanisms
- 25:44have to be built in and certainly you don't want to put
- 25:48critical components in the hands, at least I don't of AI.
- 25:53The short answer is the work we do is a beautiful dance between
- 25:56creativity and skepticism and we're skeptical always.
- 26:00And we really don't trust and certainly don't fully rely on
- 26:05systems like AI, or I would say we depend on AI for particular
- 26:12and certain algorithms for particular applications or
- 26:15particular aspects. That's why our approach is
- 26:19crucially causal and mechanistic and based on Physiology so that
- 26:24we can see better than what is being called, I guess,
- 26:31explainable AI. Explainable AI doesn't give the
- 26:34kinds of explanations that cardiologists understand in
- 26:37terms of Physiology and mechanisms and we do provide
- 26:41those. There will be rationales for
- 26:44what our systems would advise the physician when the
- 26:48physician's in the loop. Now when the physician done in
- 26:53the in the loop, then we can't say that the system will be
- 26:57perfect all at once, but we'll suddenly be designing with with
- 27:06safety in mind and risk. I should also just just mention
- 27:11that some of the, the, the deeper questions that have to do
- 27:15with testing implementation, they will involve large scale
- 27:22clinical trials just like drugs. So the same questions that are
- 27:28being raised about algorithms and how do you know if they're
- 27:32safe or not, you know, can be raised about drug development.
- 27:36And we accept the results of clinical trials even when we see
- 27:42these commercials saying that there's a risk of death with,
- 27:47you know, some of the with some of the the medications under
- 27:51certain conditions and that were proven in the trial.
- 27:54And so there is no perfect solution without some level of
- 27:59risk. And the goal would be to
- 28:00minimize those risks. Now as entity research is Bell
- 28:04Labs type moonshine ideas place, we're looking to develop proof
- 28:09of concept to develop a regulatory strategy and to hand
- 28:14off to those those companies to industry that knows how to go
- 28:20into that risk safety environment like pharmaceutical
- 28:23companies, like medical device companies.
- 28:26And we haven't touched on medical devices, but everything
- 28:28I'm describing about drug treatment also applies to the
- 28:33management of devices controlling patients who are
- 28:36more severely I'll. So we will be looking to partner
- 28:40with pharmaceutical companies, medical device companies that
- 28:44are able to afford that have expertise and ethical aspects,
- 28:48safety aspects and conducting those, affording those large
- 28:52scale clinical trials. Then we'll move on to the next
- 28:55place so we can paint beautiful pictures and brilliant
- 28:59landscapes on our small prison walls.
- 29:03You mentioned several times that your work is causal and
- 29:07mechanistic. Can you dive into that and
- 29:11explain what you mean with in comparison to things that that
- 29:16are not causal or mechanistic? If we think about a patient who
- 29:20has heart failure, heart failure usually in patients end up with
- 29:26some degree of renal failure. The renal failure causes an
- 29:31increase in volume retention. The volume retention means that
- 29:37you get fluid in the circulatory system that flows back to the
- 29:40lungs and you get fluid in the lungs, fluid in the lungs.
- 29:44Elevation of left atrial pressure means that you have
- 29:48poor oxygen diffusion capacity, so you don't get as much oxygen
- 29:52across the lungs into the circulatory system because you
- 29:56don't have as much oxygen. The heart has to pump harder and
- 29:59faster because in order to deliver enough oxygen to confuse
- 30:04the tissues, in order to do that, the heart has to work
- 30:07harder. This patient with heart failure,
- 30:09that patient's heart has to work harder and consume more
- 30:13myocardial oxygen. It needs more oxygen, but it
- 30:16can't get more oxygen. So the heart is feeding itself,
- 30:20but it can't get enough oxygen to feed itself, and you quickly
- 30:24deteriorate. That is a sequence of events set
- 30:28in place by various causes that originate with heart failure.
- 30:34That's different from, for example, having some speckled
- 30:40tracking images that are fed into AI and making some
- 30:44hypotheses from that alone about why a patient is not ventilating
- 30:52well. That for me is pattern
- 30:54recognition with an attempt to extrapolate to causality.
- 31:01But this sequence of events I just described, which also by
- 31:04the way includes anemia and which exacerbates the situation
- 31:11more, these are all things known physiologically and
- 31:16mechanistically to play a significant roles in heart
- 31:21failure. And all of the drugs that have
- 31:23been developed are mostly addressing these various
- 31:27aspects. Diuretics to relieve, remove
- 31:29volume, contract catecholamines to improve contractile function.
- 31:36But at some point, the cost in myocardial oxygen consumption is
- 31:42just really the last draw for a patient with chronic heart
- 31:46failure. It's very, very serious.
- 31:49And the intervention that your system is doing, correct me if
- 31:54I'm wrong, is injecting certain types of medications and
- 32:00precisely the right dosages and the right times.
- 32:03Is that correct or is there more to it?
- 32:07That's correct. For our particular use case of
- 32:09acute myocardial infarctum, acute heart failure where we
- 32:14have control over everything and we're measuring everything.
- 32:17It does those injections to manage the acute case.
- 32:20But the best we can expect for that autonomous close up
- 32:25intervention system is the acute case.
- 32:27And so four or five day hospital stay it doesn't.
- 32:32It gives us some knowledge that's helpful in the chronic
- 32:36case when the patient's not in the hospital.
- 32:39We have some starting places, but we need much more data, we
- 32:42need much more laboratory chemistry.
- 32:45We need to be monitoring patients with wearable
- 32:49technologies and so forth over longer periods of time.
- 32:53However, we expect that the excellent care in the acute
- 32:58situation will have benefits for the patient when they're
- 33:02released from the hospital. They won't be readmitted.
- 33:07For example, the readmission rate we expect would be less or
- 33:11would be over a longer period of time compared to the standard of
- 33:16care. You described earlier the
- 33:21analogy of an electrical system with diodes and resistors, and
- 33:30as I was reading through some of your papers and presentations, I
- 33:36was struck by these electrical diagrams.
- 33:40I mean literally like schematic diagrams that you'd see for
- 33:43electrical devices. Can you tell us more about that
- 33:47that aspect? When we naturally, when we look
- 33:51at that diagram, we see what looks like the electrical
- 33:53circuit and the circulatory system is a circuit.
- 33:58And so that part, it kind of works for us naturally.
- 34:01And actually, elements in that electrical analogue diagram
- 34:04represent left ventricular contraction, right atrial, right
- 34:08ventricular contraction, atrial contractions.
- 34:13The person who is familiar with those symbols can see the entire
- 34:17circulatory system represented there, but they're represented
- 34:21very simply. For example, just to give you a
- 34:24notion, the left ventricle ejects into the systemic
- 34:29arterial system across the aortic valve.
- 34:33So for us the left ventricle is a capacitor where for a certain
- 34:37volume, you know it generates a certain force and we call it a
- 34:42time varying capacitor because we let the capacitor level
- 34:45change or electric ventricle become stiffer to generate
- 34:49pressure. That pressure in the electrical
- 34:52allogue sense is a voltage, and that voltage in the heart in
- 34:56that left ventricle gets higher than what's downstream across
- 35:00the valve. Then that valve will open or
- 35:03that diode will open and flow will go out of the diode or
- 35:07across the aortic valve into the another set of resistors and
- 35:12capacitors which we say represent the systemic arterial
- 35:16system. So resistance in the electrical
- 35:20analogue network is the relationship between flow and
- 35:26pressure in the real system capacitor.
- 35:31A capacitor in that electrical analogue network is
- 35:34representation between volume that goes into an elastic
- 35:39artery, it stretches the wall of the artery and that generates a
- 35:43certain pressure. So by a series or a system of
- 35:48resistors and capacitors and diodes representing different
- 35:52compartments around the arterial systemic arterial system, the
- 35:58venous systemic venous system, the pulmonary arterial system,
- 36:02the pulmonary venous system, we can represent the entire, simply
- 36:07represent and I know simple, simple is a relative term, but
- 36:11simply represent the entire circulatory loop in the
- 36:14essential measurements or parameters or variables that are
- 36:18important for our application to acute myocardial infarction and
- 36:23acute heart failure. And how does this type of
- 36:27representation interplay with the creation of the digital
- 36:33twin? This representation is our
- 36:37digital twin. It's a mathematical
- 36:39representation of the patient in these terms.
- 36:44Now what I described to you was the simple circulatory loop.
- 36:48We have other subsystems that are more specialized that are
- 36:52not represented in probably the diagrams you looked at that
- 36:56represent the the coronary circulation.
- 37:00Now, if you think about the heart, the heart generates a
- 37:04pressure and it generates a flow.
- 37:06That flow goes into the systemic arteries across the aorta.
- 37:10But that same aorta is has a coronary circulation that
- 37:15supplies blood to the heart itself.
- 37:18That's why I say the heart feeds itself in supplying its own
- 37:21blood supply. That coronary, that coronary
- 37:24circulation is has a little more sophistication that we need to
- 37:29account for. If there's auto regulation of
- 37:32coronary flow, we need to account for that.
- 37:35When the heart beats, it squeezes the vessel surrounding
- 37:38it, so it squeezes its own coronary vessels.
- 37:41And so we need to account for that in describing what flow
- 37:46gets to an ischemic heart. So we have to do a more
- 37:50sophisticated model drilling in on the heart and this
- 37:54metabolism. Now it turns out that myocardial
- 37:57oxygen consumption, cardiac metabolism has been thoroughly
- 38:02described in simple terms that relate to pressure and volume
- 38:07mechanics. Again, those signals that we're
- 38:09measuring all the time, we're interested in pressures and
- 38:12volumes you can measure. You can use a pressure volume
- 38:16loop to represent the mechanics of a Chamber of the heart, where
- 38:20the slope of the pressure volume relationship is an index of the
- 38:24ability of the heart to contract.
- 38:26So a simple index of the ability of the heart to contract can be
- 38:30derived by pressure volume measurements as well.
- 38:33Those same simple signals and the areas in various aspects for
- 38:39the pressure volume can show potential energy of the heart,
- 38:42the work that the heart does, and they correlate with
- 38:45myocardial oxygen consumption. So there's a tremendous amount
- 38:48of data about the heart that's implicit in the way that we
- 38:52study it using pressures, flows and volumes.
- 38:57And Arsalan Khan on Twitter comes back again and he asks,
- 39:02with your research, what did you find or did you find that the
- 39:06systems and processes, traditional ways of describing
- 39:13the body were either not useful or overly complicated?
- 39:19How did you how did you end up with this electrical analog?
- 39:25In the world of cardiology, although the electrical analog
- 39:28model is not so thoroughly represented as what we're using,
- 39:34it has become a traditional approach to characterizing
- 39:39hemodynamics. And so pressure volume area in
- 39:46systolic, pressure volume relationship, the representation
- 39:51of the dynamics of the circulatory system and neural
- 39:57control of the circulation using baroreceptors and so forth.
- 40:01This way electrical analogue way is traditional, It's just that
- 40:07many practitioners don't understand it on this level.
- 40:13They're using volume conductance catheters in the Cath lab to
- 40:16measure volume and pressure and generate pressure volume loops
- 40:20and generating in indices of or indexes of contractility.
- 40:24So it's not non traditional, it's just that we're taking it
- 40:29on a whole nother level and extending it and using some non
- 40:33traditional or, or some elements that we discover and that have
- 40:37been published but have not been traditionally used because of,
- 40:45I'll just say, complexity of biomedical engineering compared
- 40:49to clinical medicine. So a lot of clinical
- 40:54cardiologists are using the principles but don't understand
- 40:59it so thoroughly. Now the I mentioned, I may have
- 41:02mentioned a consultant that we have on the project, his name is
- 41:07Kenji Sonogawa. He discovered and prove many of
- 41:11the principles that are being used in clinical cardiology
- 41:15today that are based on pressure volume characteristics and
- 41:19electrical analogue models. There's an electrical analogue
- 41:22model of the systemic arterial system called impedance that not
- 41:26only represent is not only represented representing the
- 41:29resistances, but the frequency dependent nature of pressure and
- 41:33flow in the dynamics of the arterial system.
- 41:36People are using results from impedance characteristics
- 41:40without being able to measure impedance, which is another
- 41:43electrical term because it's just a little bit more
- 41:46complicated than what people are are ready for.
- 41:50And so all of these insights, for example, another example is
- 41:54ventricular vascular coupling. Cardiologists talk about
- 41:58ventricular vascular coupling and one concept is to think of
- 42:03the heart that's a as a, if you will, a battery with a charge
- 42:08and the arterial system represented by pressure
- 42:14installed. Pressure and stroke volume
- 42:17relationship where you have two capacitors and opening the
- 42:20aortic valve is the closing of a switch between the capacitors
- 42:23and volume will move from one capacitor to the other.
- 42:27That notion of ventricular vascular coupling is something
- 42:29also derived by Doctor Kenji Senegal, who's a consultant.
- 42:35And I'll just add that a lot of the work I did together with
- 42:40Kenji Sunugawa when I was a graduate student at Johns
- 42:44Hopkins. And so we have a long history of
- 42:48work using electrical analog models dating back to the 70s.
- 42:55How much of the math did you need to develop versus relying
- 43:02on traditional or existing understanding of these various
- 43:07relationships? As a biomedical engineer, we've
- 43:09used, you know, whatever popular mathematics was available at the
- 43:13time that might render some new piece of information about
- 43:18biology to be understandable or to make sense.
- 43:23So I, I started way back in the 70s doing neural networks, you
- 43:28know, when it was a fashionable thing.
- 43:31We did wavelet analysis of waveforms and reflection
- 43:37coefficients reflected waves. So we as biomedical engineers,
- 43:41we love complexity. And you know, we'll use
- 43:45homeomorphic deconvolution mathematics of, you know,
- 43:50measuring earthquakes, anything that's available, we'll learn
- 43:53and use how much, as much as necessary and practical.
- 43:59So we built back in the day we built, we had RadioShack
- 44:02computers controlling analog computers that we patched in in
- 44:08order to have real time control of an isolated perfused dog
- 44:13heart preparation. That was just the way we do
- 44:18things. So I don't know how much
- 44:20mathematics, but as much as necessary to be practical.
- 44:24There's a image near Ian Hunter at MIT who once said to me, he
- 44:31said scientists like taking things apart and you engineers
- 44:35like putting things together. So I in retrospect, everything
- 44:41I've done, I'm more an engineer than a scientist, has been
- 44:45integrative. How to explain the system when
- 44:48it's already put together. And sometimes that takes a lot
- 44:54of mathematics, and we'll do what's necessary about them.
- 44:58Sounds like you're building the bionic system.
- 45:01Remember there used to be ATV show called the $6 million man
- 45:05that was sounds like you're you're building the bionic
- 45:11approach. What about the the team required
- 45:16to build this? There are so many different
- 45:20skills that are involved. One comment to your comment
- 45:24about the bionic man by like woman, woman is that Kenji and I
- 45:30once wrote a paper around the bionic baroreceptor reflex with
- 45:35exactly that notion in mind, where you have a device that
- 45:38automatically can regulate arterial pressure reproducing a
- 45:45failed autonomic system. And there's a syndrome called
- 45:49Scheidrager syndrome where the receptor doesn't doesn't work
- 45:53and that kind of system has been proven to work.
- 45:56So we back in the 70s or 80s, we were developing A bionic
- 46:00baroreceptor reflex. Yes, on our team in the May lab,
- 46:06immediately we have a number of PhDs in biomedical engineers, in
- 46:11biomedical engineering, we have MD's, we have masters people and
- 46:19everybody can do programming. Everybody knows Physiology
- 46:24either from the beginning or they've learned from our
- 46:28consultants, and we have routine conferences with cardiologists
- 46:33and anesthesiologists doing experiments together with them.
- 46:37So we have that level of expertise.
- 46:40And at the National Cardiovascular Center, I
- 46:42mentioned earlier, those guys are seeing patients on the one
- 46:47hand, then doing experiments in animals, then writing programs
- 46:52and then building electrical analog circuits.
- 46:54The same person is doing all of those things.
- 46:57And so we have access to all of their training and their
- 47:00expertise. And this is an occasion where I
- 47:03should also mention that I've talked about our main
- 47:07collaborator and our main projects on the larger scale.
- 47:12We also have collaborations looking at other aspects of
- 47:17digital twins with Harvard Disease Biophysics Group, where
- 47:22we're working on building developing a bio hybrid digital
- 47:28twin strategy using stem cells to manufacture tissue engineer
- 47:34hearts from patients own stem cells and then study those
- 47:38chambers beating and isolation. And we have a goal or they have
- 47:42a goal and we're trying to facilitate them to build a heart
- 47:48based on on patients own stem cells and tissues generated in a
- 47:53bioreactor from those stem cells.
- 47:55So there were tapping into tissue engineering and lots of
- 48:01machine learning. We published a paper with them
- 48:05in Science Robotics on machine learning inspired design of a
- 48:11ray, where they posited that it may be possible with AI and
- 48:17machine learning to have biological design evolve at a
- 48:21more rapid rate than evolution on the basis of what would be
- 48:25predicted from some first principles that we know about
- 48:30how tissues function. Not necessarily a trial and
- 48:33error approach to revolution, but a specific design inspired
- 48:37by machine learning. Then we have a Group, A
- 48:40collaboration with Technical University of Munich, where
- 48:44they're developing different types of biosensors and
- 48:50different types of technologies for in vivo measurements of
- 48:55electrical signals. And so we have a lot of
- 48:59expertise available in our collaboration done, a lot of
- 49:04expertise local to our small team, which is only about 9
- 49:09people at the moment. So you really are fusing the
- 49:15Physiology with medical practice, research with robotics
- 49:24and the math abstracted to create your models?
- 49:29I mean, it's so interdisciplinary.
- 49:31It is. And while we have a strategy,
- 49:34we're also again we're that Bell Labs type group where you get a
- 49:38bunch of creative people who have lots of new ideas.
- 49:41So some of this, the work that we're doing is born out of many
- 49:47great ideas that need to be pursued at the same time that we
- 49:51try to keep a larger strategy in mind so that everyone is working
- 49:57around a kind of organizing principle for what we want to
- 50:02achieve. And one of the main goals is
- 50:05improved patient outcomes. Some of those goals we expect to
- 50:10achieve in the longer term, but some we want to demonstrate in
- 50:15the shorter term. And this autonomous closed loop
- 50:18intervention system is turning out to be at least in terms of
- 50:25first in human studies. We expect to see some first in
- 50:29human studies within five years. You brought up earlier the
- 50:35HealthEquity vision that you have and how does this research
- 50:39relate to that and very quickly, because again, we're going to
- 50:41run out of time. If you think of the quintuple
- 50:44aims of healthcare and that's improved experience of the
- 50:48patient, improved experience for the clinician, improved
- 50:53population health, decreased cost, improved HealthEquity,
- 50:59then I think for example, the autonomous flows of an invention
- 51:02system does all of that. So improves HealthEquity by not
- 51:07requiring so many specialists in order to treat a very deadly
- 51:14cardiovascular disease. The cost, we hope, should be
- 51:19less. Although the history of medicine
- 51:22would suggest that no matter how dramatic an improvement we make
- 51:26in scientific advancement, the entrepreneurs find a way to make
- 51:30it more costly. But we think that this
- 51:34technology will be less, less costly and should improve the
- 51:41physician experience. I won't be so exhausted having
- 51:45to give so much care to managing particular patients and the
- 51:50benefits just strike directly, I think with the quintuple aims of
- 51:57improved healthcare. Where is all this going?
- 51:59Where is your research headed right now?
- 52:01In the immediate term, we're headed toward those first in
- 52:04human studies that have to do with clinical decision support
- 52:09and we're looking to develop regulatory strategy with a
- 52:13partner and hand off the expertise, hand off to them and
- 52:22their expertise and investments in large scale clinical trials
- 52:27and developing technology devices or pharmaceuticals.
- 52:31And then we would move to focusing on chronic heart
- 52:37failure and that's where we would expect to make use more of
- 52:44the collaborations we have with Harvard Disease Biophysics Group
- 52:48and with our folks at TUM wearable devices and such.
- 52:54And again, this is all starting from the notion of a human bio
- 53:00digital twin or human bio digital twins, not just
- 53:03cardiovascular. It was a starting place and I'm
- 53:06personally interested in neurodegenerative diseases and
- 53:09so if I'm around long enough, I hope we can get to to that
- 53:15branch of work as well. All right.
- 53:17And with that, I'm afraid we're out of time.
- 53:22A huge thank you to Doctor Joe Alexander.
- 53:26He is director of the medical and health Informatics lab at
- 53:31NTT Research. Joe, thank you so much for
- 53:35taking time to be with us. It's been a very fascinating
- 53:38glimpse into your work. Thank you so much Michael for
- 53:41having me. Everybody, thank you for
- 53:44watching. Before you go, subscribe to the
- 53:48CXO Talk newsletter so we can notify you about upcoming shows.
- 53:53We have great shows that are coming up.
- 53:55Thanks so much everybody for watching and we'll see you again
- 53:58next time. Have a great day.