Chapters Transcript Video The Anatomy of the Human Cognitive and Emotional System: IA vs AI Dr. Michael Sughrue discusses how cognition and emotions are regulated by the brain and how to modify treatment plans based on Medical AI practices. Presenter: Michael Sughrue, M.D.Co-Founder & Chief Medical Officer at Omniscient Neurotechnology Well, good morning, everyone like to have Edward show up July 1st that I was wondering how many people are at judgment. But um well, thank you for uh to the coming over to the grand round today. And uh we have dem speaker Dr Sh um Dr Sh did his neurosurgery residency in U CS F. And following that, he did a fellowship in a minimally invasive neurosurgery awaited Charlie Tho over Sydney Australia. Um And then he came back to the US and became faculty at Oklahoma University of Oklahoma. And I ran into him at one of the I think the NS conference and I was immediately struck by his rep of uh knowledge and intellect and, and, and the amount of literature you can cite. And he and I got into discussion of functional neuro anatomy. And since then I learned a lot from him. I visited him at Oklahoma and I was impressed with it or now he then took a detour to work at Sydney Australia where he launched his mission neural network. And uh here here he is, well, he is the founder of that company and then he has become world renowned in this field. And then he has traveled around the world to speak and in various countries. So it is a great honor to be able to have him over and discuss this tech topic. Uh So here Dr Sugar, he thinks everyone and uh just as a back note, I did, I do have a disclosure which is obvious it's actually, I'm talking mostly about anatomy here. So it's not relevant specifically to this, But we're gonna talk about is what do we know about how cognition and emotions are regulated in the brain? What can we utilize technologically to ultimately uh do better, a better job for our patients? And how can we think through this and modify what we've been taught to adapt to what's really been discovered over the last decade or so? And we'll talk a little bit about uh what I think how medical A I should work or what, what, what isn't being done, what should be done and, and where I think things are going, uh it'll be a brief digression on that. So I told neurosurgeons if you could, basically, if I could have only had one slide to do, if I, I could basically give the talk like this, should we cut large scale brain networks during surgery or try not to? The answer is yes. Now, what we're gonna talk about is what that actually means. OK, because ultimately, where it comes from is, is recognition that even when I was a resident. Many, you know, now 20 years ago, it became obvious to me that what we were saying as neurosurgeons was a great outcome was something that looks like this that you go around the world, seeing people talk about surgery. This is usually the gold standard for I did a good job. But in reality, when you meet patients and you get to know them after a brain operation, some of them go home and do things that, that are, you know, irrational, dangerous. Um And uh and oftentimes ending in things like divorce because people's personalities change. Now, we've been talking about that for a long time. There's all the people who've come out and written papers in the literature saying everything is eloquent and, and that's all great and well, but we also know that often people need brain operations and not everyone has these issues. What we also know is that um we know that the more you remove of a Glioma, for example, the longer people live and so almost linear at so at the upper ends of resection. In fact, if, if you remove a margin, like every cancer surgery, it seems to be helpful. And you can see that I've been actually talking about that for quite a long time. But I think one of the things that we have to recognize is that we're not doing a great job in some of our domains and this is not just true for brain surgery. It's true for most parts of neuroscience, but brain surgery brings it most starkly. So if you look at it, this is a paper that's written at a Penn State, they included me on the paper. And essentially what they found in this paper was that after treating people for gliomas, the rate of a having a new mental health problem, mainly anxiety and depression is about 70% which is about 3.5 times the general population rate and higher than people with other terminal illnesses. So, pancreatic cancer is bad. It's bad to have a glioblastoma. But people with pancreatic cancer do not have a 70% rate of depression. We're doing something wrong. But what? Right. So I can sit there and say this, but the answer isn't to back off of the tumor. It's to try to balance these two competing risks and to do so with facts and one of the problems is we know it's somewhere in one of these question marks. But where because drawing that line distinction is critical for doing brain surgery and this is true for a lot of diseases. So if we look at things like Alzheimer's, if we look at stroke, these are also very bad problems. Now again, people have talked about this in neurosurgery for a while. Um This is actually for those who aren't neurosurgeons, this is a fairly well known person in the field. And ultimately, the discussion was how do we adapt things that we already do? Like inter operative brain mapping to address cognitive issues. What I'm gonna point out is we've been doing research on this now for over a decade and I'll pre prevent some evidence to say that that's probably not going to be the answer to this. In fact, I mean, it probably is impossible to map some of the functions that we're talking about that are relevant to our patients. And what I present instead is a first dive into minimizing this problem, not that we can eliminate all of it, but angling around the concept of large scale brain networks and particularly these five, which we have the most evidence for. You don't have to memorize this. Now, I'll come back to it and we have studied this extensively and the amount of evidence that these particular networks are bad, they're bad when they get damaged is actually pretty significant. So there's over 550 papers when we wrote this, which is about 2.5 years ago, since 2010, there have been 550 peer review papers across multiple diseases that say that damaging these networks is bad. It causes cognitive emotional problems. Ultimately, risk is unavoidable, right? So the patient comes in with something bad and this is going to cause a lot more problems than cognitive issues. And so we have to take some risk and so we can't be afraid of things or under operate because we don't know what's, what, what's, where and, and that everything is scary to us. Um, again, I put this up in a talk when I went to Korea because it always shocks the South Koreans that you've been to the North. But the point of it is is that when we talk about the benefits of doing slightly more surgery, they're real and they've been reproducible, but they're not enormous, right? So taking this to its logical extreme as Schopenhauer pointed out, anything taken to its logical extreme leads to something illogical. And so we have to have is a better model than what we've had for a long time. And again, we've been talking about where things located in the brain functionally for, for over 100 and 100 and something years. Now, the problem with it is, and this is something that if you, if you don't dive into the discussion of co how cognition was perceived of in the early neuroscience era, you might be aware that there was a decision made that affects how we think about the brain. Now. And so in the early days, there was a debate between localization is which is that certain parts of the brain were specialized for certain things and distribute distribution or non localization is or globalism is another way it's been put. Now it's safe to say that you probably are more familiar with the thing on the left because the localization is largely one, this was due to neuropsychology, bro, uh healing Jackson and many other people who showed that there's certain uh certain problems that happen in certain areas of the brain with certain injuries. And that's a pretty good thing for certain functions that are highly localizable and for reasons that will make sense later, um really kind of can happen from a single lesion. But what we found is we've um learned more about the brain is that bringing distributive or global ideas back into the model actually helps us understand higher functions. And there's a very good reason for it. So it isn't that this is wrong, it's that it needs modification and we need to rethink some of the assumptions or simplifications that people made a very long time ago when we didn't have function in their imaging, I keep going back to this cause I cause I, you know, you're, when they talk about teaching, you should, they say that you should give a topic and go over this over and over again. Ultimately, what these networks come from is the idea that there are areas of the brain that are not next to each other that usually have physical white matter connections that are mixed in with other white matter connections. But that these areas have highly synchronized activity, they talk to each other very often. And part of the reason they're doing that is the same reason why the motor system talks to other parts of the motor system or the visual system. On the right side, talks the visual system on the left side are broke on war, you talk to each other because they're doing functions in common. And the brain is telling you by the fact that these areas are co fring together that they're probably useful states that the brain uses very often. So we we this first bit kind of came around around 2008 in the scientific literature and around uh a few years after that, I became familiar with it because it made it over to my world. And this paper was published in 2016 in nature and it was basically determining how many areas of the cerebral cortex are there. Now, Robin said 47 or 52 depending on how you describe it. But ultimately, that's obviously not ex he didn't nail that in 1908. It's amazing that he got anything uh that's vaguely reproducible. But ultimately, what we know is it's closer to about 100 and 80 to 200. And this is the the human connector projects version of it. Um a couple of things I think are interesting. So I don't know if my cursor is showing up here very well. One is that this is the insular here, the under surface of Broca's area does something different than the outer surface. And there's also also areas that are sitting just inside Sulci, there's areas that cross s boundaries. So the brain doesn't care what a gyrus and Sulcus are. There are actually no functional organization related to that folding pattern. The the brain folds around white matter connections and that it happens to put the function where it thinks the function needs to be. And so we looked at this and said this is a really useful alphabet for doing reproducible science on brain anatomy. So if I said eight C, you could go and determine whether that was true or not in a way that was reproducible. And we could share information more importantly, we could start to describe the brain in an appropriate and learnable level of detail. So we did this a lot. Um So this is probably about 75 papers. Now, at this point, I'm not sure I haven't kept count. But what we've really did is what's called that analysis where we go out into the world's literature, we do an assessment of where the F MRI activation points are. We locate what area lives in those coordinates and then we can get a map. OK? And this is language if you're paying attention, but we can get maps of lots of things. And, and again, like I said, we've done this a lot, um actually had, this is an entire issue of neurosurgery that I published a couple of years ago. This is about now six years ago. So I'm getting old, but ultimately, what we did is we, again, we did lots of detailed analysis of who these areas are, who do they talk to. And ultimately, what we wanted to do after that when we first published it is is that everyone said that this was really difficult um to learn. And partly because just a lot of information that's 500 pages. The book is this thick and uh I probably can't remember all of the things we published in there. And what the ultimate issue came down to is how do you find one of those areas on the cortex? It's all great for us to have literature on it. But if you can't do it in real people that it's not useful. And the, the real challenge gets down to when we have people who have pathology in the brain tumor, strokes, brain surgeries, et cetera, the brain isn't really normally shaped. So how do we find that in this kind of person? So we created an algorithm and what it does is it looks at different areas of the cortex and it, the machine learning knows whether a voxel belongs to Broca's area or not. And so what it does ultimately is it's capable of determining the boundaries in an individual person and where it excels is when there's stuff that's not normal. So what it can basically do is it can locate things that behave like brokers area. And as you can see, it can map it to a new area when we first came up with this. Um I basically put the worst brain I had in my entire collection. This is a patient we did not re operate on but they, because I trained every resident in the or everyone in the building to order AD T I on everyone. We had AD T I scan on this person. So you can see that there is a large part of the frontal lobe has been removed, the insula, if you look up in the upper left has shifted almost a centimeter and a half interiorly into the space. But despite that and, and the recurrent tumor, the machine learning knows how to find the basal ganglia, it knows what's missing in the cortex. So it's very robust. And that lets us have a serious discussion about a patient like this, right? So this is a butterfly glioma. It is a bad tumor, right? Everyone agrees on that. And when I was a resident, this was considered an inoperable tumor. And the reason it was considered inoperable was that if you charged in and took it out, the patient was often left in bad condition. But if you follow the natural history of the disease, this is a very bad prognosis without surgery. But basically, patients are usually in bad shape within a month or two after diagnosis. And the question is why is this so bad because we cut the corpus callosum all the time. We, we've cut it for decades and the consequences are not catastrophic. So why is it so bad to remove that tumor? Well, part of it is, is this is not really a corpus callosum tumor. It's a tumor that happens to be in the corpus callosum from the parietal lobe. So that means that it has networks like language and motor that's growing around. But even more important a network called the default mode network or DMN is actually almost 360 by a tumor like this because it, it, it sits in the singular gyrus, the corpus callosum wraps around the singular gyrus. And as you learn, uh we weren't talking about the DMN when I first, I had never heard of this until I started studying how to take these tumors out. But the neuro scientists have been talking about this extensively for almost 20 years. And when you tell them that we didn't even know this existed until about 2018, everyone's shocked. And part of it is because the DMNM modulates a lot of higher cognitive functions. It's really important for uh uh internal cognition. It's abnormal and lots of mental health issues. And the reason people began taking psychedelics seriously as a as a therapy for depression was that they decouple the DMN and let it reset. It's like a defibrillation almost. And once there is evidence that it did that people start taking it seriously because we know how much the DMN is a part of many mental illnesses. So what we can start doing is thinking instead of going into the tumor and just sucking it out and hoping that we stop before something bad happens is actually making deliberative cuts next to networks. We also can look at a patient like this where in the past, we would just say this looks really bad and not just it doesn't make it good. We still know this is bad, but why is it bad? What's bad is first, it's wrapped around the fault mode network. So she transected it on one side, the A I can find that. But the A I also found that this tiny rim of brain that we would have normally thought couldn't possibly have anything in. It actually does. It has a network called the salience network and the fibers are actually being displayed around it. This is a solution that only a machine can come up with humans can't. So what are these things doing? Why are they important? Well, the first way to think about it and this is a, a simplified model of a much more complicated process, but it's helpful for getting, putting your foot into this. So ultimately, if you think about what the higher networks are doing is they're allocating resources in other networks, they're determining how does the language system, talk to the visual system? What are you going to focus on and what states do you need to get the brain into. And ultimately, there's kind of an internal mental state, that's the default mode network. There's an external state that's the central executive network. And a third network called the SAN network, which I just showed you, which is kind of like the gear shift. It's determining who's do, who's controlling the microphone, so to speak. And so we spent a lot of time getting granular in ways that the neuroscientists weren't because they're not gonna go have to treat a patient with this. So it doesn't really matter if it's here or here or here. And for us, it does matter, right? So using that analysis, we map the default mode network. You can see that it's an anterior cingulate, posterior cingulate and lateral parietal network. It's bilateral though though it not uh most people, about 70% of people won't tolerate a unilateral injury or at least won't tolerate it very well. Um That's anecdotal, but I can promise you that other people in the world have seen that same anecdote. Um Salience network is a middle singular anterior insular network. It has a track that, that no one knew about it until 2008 connecting it to. It's, it's two parts called the frontal Aslan track. And the central executive network is kind of a multipart network that extends around the brain. It's actually the network that develops between the beginning of adolescence and late and late or middle adulthood. So, between 13 and 30. And again, you can imagine what that's doing. If it's the main thing that's developing and during your maturation process, on the right side, it's heavily linked to, to a lot of the behavioral and, and judgment issues people get. So if you look at what the control networks do, the evidence on this is extensive, um Any of these things could be a grand rounds topic by themselves. Um But ultimately, they do a lot of the things that we care about. And part of it is if you think about higher function, and I'll explain this in a little more detail to what they're doing is aligning the brain to get into the complicated states ne necessary to navigate real life. But we can start to basically look at what cognitive syndromes or, or clinical syndromes that we kind of understand a little bit that are roughly localizable. We don't know exactly where and understand why. So, one of the things that we've known, I at least learned about it in medical school was that there's something in the medial frontal lobe that if you injure it, particularly on both sides. But even on one side, people can become abulic or have a kinetic mutism. So we kind of know this is being bifrontal, that's the slang term for it. But when you say bifrontal by what exactly because it kind of matters if you have tumors in both frontal lobes or you're gonna be crossing both sides or, or what are we doing wrong in patients or what, what when someone has an ac a stroke, what exactly is causing them to not do well. Well, now we can start to grasp this because if you understand that on the, on the medial hemisphere and only the medial hemisphere, these are actually connected in the sagittal plane, they don't talk outward. Usually that you have the default mode network where you think to yourself, the Salience network which shuts that off and turns on the active system. And half of the S MA being part of the Salience network, the supplementary motor area that this is a system that gets you out of your own world and start to move. And again, um beyond just being conjectured, this is a paper out of Harvard and PNAS. They found when they looked at the network lesions of akinetic mutism that it matches that pretty well. Um And what's interesting is that I talked to Mike Fox when we, when we right around the time he published this right before he put it out the door. And I told him that when we were doing awake brain surgery, every time we touch something in the singular Gyrus, the patients just shut off almost. Uh I got videos of if you're interested and he said it's right here, isn't it? And I said, how did you know that? And he showed me this paper that he was about to publish. So it's interesting when two different people independently come to a similar conclusion using a di uh different lines of evidence. But the bottom line is we start calling this the initiation a access because that's what it's doing. It's initiating movement. There's actually two other attention networks. One, this one is called the dorsal attention network. I won't go too much into it because again, it can be its own grand rounds talk. But essentially the way to think about it is, it's how the frontal eye field talks to the interet sulcus and how that talks to the visual system, it keeps your eyes focused on a task and that doesn't sound as important as it really is because what we know when you damage this network is that people's IQ tends to go down in a somewhat linear fashion to the extent of the injury. So if you've ever seen somebody with a glioblastoma or injury in the motor cortex, who kind of looks worse than you would just of hemiparesis, understand that you can always make someone worse by even if someone's hemiplegic by continuing to operate. Finally, um I remember medical school learning about neglect, having spatial neglect and it was always just it's in the right parietal lobe, but it turns out this is why um so we ultimately, we again, we could go into a lot of detail of why these parts are in here. But essentially, we now have a pretty good understanding of how to avoid this problem. So, one of the things about thinking about this in surgery, it's great to know these things. But how do we think about how to do an operation differently? And when I was a resident, people talked about skull based surgery and it's upbeat your chest. It's the hardest stuff and it is difficult. Right? We're not gonna lie about that. But there's one advantage that you have in skull based surgery and it's that everything is visible to you. So there's no 14th cranial nerve that we're gonna find out in 20 si 26 that we need to change our operation. It's you kind of know what the, the playing field is and you can see it. Whereas when we talk about the cerebral cortex, it's very easy to go in there and with the suction or AUSA and remove the tumor, but you don't see most of the anatomy. So where's the DMN in here? Because it's actually on the screen? And the problem with it is that it, it's a really, really complex and largely invisible anatomy. So it seems simple, but to do a good job, it's really hard meaning getting a good tumor resection and leaving somebody in the best possible state because how do you get a sign where to stop? There's just not that much visible anatomy. Um I wrote a book on this several years ago. I always put that in there. Um when I went to Korea about a month ago, I almost everyone in the audience had me sign the book. So uh it, it is, it has. So it's at least sold there. But the point is what we're trying to get across in the book was that in reality with these tumors because of the nature of the tumor gliomas integrate into synapses, whatever you think you're doing at some point in the operation, you're removing brain, it may be not very functional, it may even be seizing. But you have to make a decision of what you're willing to sacrifice to get a good resection. So yeah, there might be an area of necrosis or something else or something that's just purely tumor. But at the borders, if you're really actually moving the goalpost, then you're gonna have to make a decision of when to stop and you don't have good anatomy. So we started thinking about how do you make cuts around networks to do this as safely as possible? Now, it doesn't mean that it's always a good idea to remove the whole frontal temporal lobe. But this is if you had to do it, here's the best way to do it. And we started thinking about how to automate this. Um And again, I've gone over this with now, had probably hundreds of people at this point. And what we learn is a couple of facts when you have having the networks on the screen that you actually learn the first is that most people taking or train and take the shortest path or go down the souls. That's sometimes a good idea. In this case, it was not, it takes you right through language areas and taking a longer path is better. Um Also I can, if you, people think that trans social approaches are always better, I can show you situations that's not true. Um But more interestingly and this is something that I people have often ask is what they think is. Ok. Well, we gotta get these tumors out. This is really bad. Um And so what's the big deal if you cause some memory issues? And again, that's easy to say about other people's memory. But the other thing about it that it's important is that that's usually not what you're talking about because most people back off when you start becoming unfamiliar with the risk where you know that there's something bad somewhere and actually having clear defined boundaries makes you more confident. And so for example, when you look at doing uh this medial frontal disconnection and you see a tumor like this, the way I was originally trained to do this was we'd start on the outside and just slowly shave until we got scared and stop. Ok. Now we're pretty good. But the reality is there's probably a more efficient way to tackle this tumor. And what it is is that you think about it as a series of disconnections. So you disconnect that from the motor system, the default mode and the central executive and notice that this is not a biopsy. That's not what we're talking about. This is a aggressive operation, right? But a couple of things to note. So there is tumor left. Look at the lower, right. That's the arrow that's in the basal ganglia and the basal forebrain. If you remove that, then please publish your results. But I don't. Um uh uh the other thing is that you see this a little thin rind of, of default mode. That's all it takes to keep it connected to itself because remember it's connected anterior posterior, it's not connected laterally. And that means that we basically minimize the dysfunction. This patient went back to teaching and still alive eight years later. That's a good balance. So it isn't that, that m more tumors better or that's, that's absurd. And we don't all know if you think about this. It's also not that you back off and leave huge amounts of tumor to save a tiny function that we make these decisions based on facts and rational thought and value judgments that align with what the patients willing to tolerate. So, networks are important. I'm gonna dive into a little bit. Um This, I think this is fascinating and it's also important to know why are networks important. And the basic gist of what I'm about to go through is that they are the organizing principle of the human brain. This is how the brain does our cognitive functions. So what I start with is a concept called brain states. OK. And we're gonna work backwards to how does the wiring of the brain serve those brain states and where networks fit into this? So a brain state is a configuration of what is on and what is off at a specific time point. So let's just think about a real world scenario. So let's say I see someone coming and I have different things that are on. How do I feel about this? What's the context? Who's where, who are these people? What, what visual impulses, what auditory impulses? Well, there's not a part of the brain that we know of, that just gets all that information and integrates it. It's actually a bunch of simultaneous events. OK? That's a state and you have a different state for part two of this, this vignette and so on. So the brain is always going through its states over and over again. And in underlying this, some areas are probably turning each other up or down, changing the gains on connectivity, um turning off and on certain cortical columns, sending different rhythms. There's a lot of things happening. OK. But at the basic level from a systems Neuroscience standpoint, basically that state is the representation of reality at different parts of the brain. And ultimately, your brain is gonna cycle through these states through about every, every moment of your life right now. How do we know that that's actually how the brain works. Well, this is a paper of Japan that I wish I wrote but I didn't. And what they did was an incredible amount of work where they basically put together a, I don't know, like a gauntlet of F MRI studies where they induced 20 the 24 derivative human emotions in a scanner and studied how people's brains what was on it at the point. You were feeling this emotion interestingly, it's fairly consistent. So they actually gave me the raw data because I I wanted to be served on this. But what they found is that again, this is the map here. This is actually the same scheme we're using and given on what is on or off at a given time point. You can deduce what emotion someone is probably feeling. Now a couple of things first is if you're really paying attention, this is like the default mode, central executive salience, that's what's actually the red area is the visual system is here. So you can see that and that's not, doesn't really play much of a role. Why? Because you could be feeling emotions, seeing different things or doing different visual tasks or not even having a visual task at all. But ultimately, you can actually break this down in its distinct states for every emotion. So couple of things, it's bilateral, it's diffuse and it's very complex, but those are states that's how they work. And so we can represent states based on um uh uh we, we can abstract them down to a graph, right? What's, what's on and off at a given time point. OK. And if you look, think about this, the brain can do a lot of things. There's lots of different combinations when you have 360 areas with different configurations. So that makes sense that we could represent a lot of different realities with that. But some of this stuff for a reason, my slide keeps screwing up on this, but there's some states that are implausible. So having the brain all the way on, all the way off in sync is really not gonna happen. And there's some states that could happen that are plausible, they just don't happen for whatever reason. So ultimately, you have states that actually happened in the real world. And when we do surgery in the brain, what we do is we make that state impossible. OK? Or in a set of states, the more brain surgery we do, the more states that start to disappear. So we basically go in and we blow up a set of states. Now does that matter? Well, probably not in some cases if we're talking about is opening your mouth and that's the area. Unless I take out the mouth, motor cortex, you don't see that disappear. But if we take the same scenario between these two. And let's say this is not getting angry in the first place. And the other one is calming yourself down, the context of that mouth opening is different. And so what I want to get across with that is that if we do inter operative testing, how can we possibly test every possible context? It is impossible. So we actually have to use better tools to figure that out. What can we get away with and what can we not? Now this is complex. But ultimately, what we did a study in this a while ago, we said, what are the regions that we have called eloquent, like the motor cortex of brokers area? And it turns out you can measure how many different areas does that area talk to? And the areas that are eloquent are the ones that most of the brain wants to talk to. They have the most connections. So the brain stem is number one and motor cortex is high on there, the visual system. So what we, what we actually mean is that when we do a surgery in an eloquent area or you have a stroke in an eloquent area, you actually eliminate a lot of states at once. And that's why they cause deficits that are obvious because you lose a lot of different ways the brain works. Now, if you start breaking this down, you can get the brain into circuits which are different configurations of states, different circuitry that tends to be part of a, a lot of different states. Um I won't go too far into this but uh it's actually where the N I Mh is taking the field of psychiatry over the next decade or so. And what they're ultimately doing talking about is the following. This is where it gets interesting. So instead of thinking about depression or other mental health issues as a disease, a common disease, it's a set of different circuit problems. So it's not m unifocal, it's multifocal. And what I mean by a circuit is, let's say I have a circuit in the brain that does three. Where probability is your assessment that if I do something will something good happen? Well, it turns out that there's places in the brain that decide that right? And they could be part of a lot of different states depending on the situation. If they're not really, if this circuit always gives you a negative response. Oh, life is terrible. Nothing good ever is gonna happen. I don't know why I even try. Right? But let's say, for example, it always gives you a positive response like, yeah, do it. Well, you're manic. That's what manic is. It means even you're doing things that don't even necessarily make any sense. So we've mapped circuits, we mapped them in a lot of detail. Um And ultimately, we've used machine learning to do this. Um What we've really tried to say is, this is really complicated and it's not really up for if I just give you the spreadsheet and say, hey, why I go for it? Um It, we need computers to help us do this, but we don't need computers to make judgment calls or goals or decisions because the challenge with machine learning is it, it doesn't do very well on extreme outliers. And we all know that the teachers walk on our leg all the time who just don't fit the book and the machine will really hard to train it on something it's never seen. So ultimately, we don't need a I to see that there's a brain tumor there. It's nice, but it's not helpful. It doesn't solve a problem just like we don't need it to tell us that someone's schizophrenic or that someone who can't move their arm is in, it has a poor state of health. We need it to actually find useful inputs. And so we basically built our machine learning to answer questions and built a big hyper cluster in the interest of time. I won't go into how cool this really is, but it's able to process about a million MRI S an hour if we wanted. And this is actually the coolest slide in the talk in my opinion. So what this is doing all these dots is we uh we tell the machine learning to ask a question. This is what we wanna know, go out and find everyone in the database who fits these criteria and try every possible machine learning approach to f find the best fit answer to that question. So all those dots are this thing running hundreds and hundreds of different approaches. And on the far right, is it replacing the model till it gets the most accurate model? So it's brute forcing every conceivable solution in about 35 minutes. But the we, the question that we wanna know sometimes is OK if I have someone who's hallucinating where in the brain is abnormal, OK? Because we know that's gonna be complicated because we can do things about that. Theoretically, we put electrodes in, we can do T MS, we can do other things. And what we get is a read out. This is where, where those things happen. So this is pretty interesting. The machine learning here told us that hallucination is in the auditory cortex and the visual system. That's not surprising. So that's good to hear. But it even found an area in the insula that when you think that the government has implanted something in your abdomen, that's the somatic concern issue. It actually it lights up and it turns out that area is actually synchronized with the stomach. If you look through an electro gastro gram at CWMB MRI, that is the only area that's synced with the stomach. So it's pretty cool. The machine learning pulled that out. So we built um large language models to help people work through this. Um So you can see where this is ultimately going, we know that no one is going to memorize all of that stuff. But what we want to do is how do we translate what the patient says into circuits so that people can actually work with it. And I won't go diving into that too much. But the whole point of it is is that ultimately there's a symptom behind all these circuits, the brain is really complex. We need tools to help us quickly get to answers. But coming back where networks come into this, what ha what a network ultimately is is that these two areas of the brain are co fring together because they're doing something in common. So what that means is if I look at these two waveforms every time point I put a dot on here and I determine how often if A is firing is B firing and if the number is high, these areas are network and that's fairly consistent between people. But why, what's the point of that? Like, what's the brain telling us? So the first that's telling us is that these are the states that I use most often or the parts of states. So they're really useful configurations of the brain because they're doing certain higher cognitive functions that we use all the time. The other interesting thing is if you take a map of all the networks of the brain you get a map that looks like this, it looks kind of chaotic, a couple of things that are interesting. Um If you look at the motor cortex, it's actually not one network, it's two face is actually a different network. But the other question is you see these kind of like areas all over the place and, and this seems like almost irreducibly complex and if I thought it was for one point, but it's actually not. So there's actually an order to this. And I'm gonna show you it um as quick as I can because it's it is complicated. But I think fascinating. So if you look at the at the hemispheres again, a couple of things show up. So first, most of the basal ganglia is not doing a motor function. So if you look at that green area at the butane at the bottom right there, that's the only motor part, the rest of the caudate and striatum is doing higher cognitive functions. So it was the cerebellum actually only this part is the motor area. The rest of this is cognitive, but most of the hemisphere, the other thing is someone pointed this out to me at a meeting one time and I I became interested in it. Uh and this is the wrong color, but you can see that these little triangle angle, arrow shaped things on the medial hemisphere kind of look like a segmental patterning. So we became really interested in the idea that this might be similar to the spinal cord because it comes from the same neural tube. We also noticed that there are these triads, brown, red and yellow. It's EMN language and central executive for those who are keeping tabs at home and that's repeated over and over, it's repeated in the cerebellum, it's repeated in the basal gang. They, so why, why is this? So uh pattern? Well, it turns out that if you look at uh the hemisphere, we can actually roll the clock back, we can roll the clock back. And for some reason, this video, there we go to that point when, when you're early embryo logic stage, all the animals are relatively similar in the mammal, mammalian hierarchy and start to look at what was the brain like as a neural tube. So we did that and you get a map that looks sort of like that, that's half the tube. And what you find is there are these, those triads that they pointed out are aligned and reproduced along and rostrocaudal axis. And what what we then did sort of like this? See that's the, that's the diagram you see this is the four triads and they're all symmetric around the motor cortex. So we basically then took every animal in in back to fish figured out. When did these areas first show up? And what we found is that humans have four of these triads, macaques have two and cats have none. That's actually the big difference. So what I tell people is a, this is kind of an organizing principle of the cerebral cortex. B if you basically damage these networks, you're taking people back the evolutionary chain. That is the main thing that mammals did. And if you wanna make someone into a cat, this is a great way to do it. But also it explains why the cerebellum looks like it does. It's actually just tracing triad number two, right, which is actually the macaque triad. This is the human one up here. So needless to say networks are important and this is the pa paper I referenced earlier. This is actually a paper out of the Barrow that was published a month and a half ago and they looked at 364 Caver Nomas. So this is 20 years of data, multiple surgeons um and what they looked at was not subtle cognitive issues, but whose modified rank and score got worse. And we didn't ask the machine learning, you can see these are all the approaches they took. So it covers the whole hemisphere and again, modified ranking if you're not familiar with it isn't cog a cognitive score, it's disability, it's a six point disability scale. So ultimately, if you look at these areas, these are the areas that the A I said these are the problems. Well, you can see a few trends. That's DMN, that's Salience, that's DMN. That's dorsal attention network right there. And in fact, this explains 9 95% of their complications. Ok. So why they're not going through mo motor cortex? They're not going through Broca's area. This doesn't say that that's good. This is the areas that trained surgeons are screwing up at. The only other one is, you know, going through visual cortex. So what it shows is that again preserving networks is critical. It's not just the subtle cognitive issue, this is this ability. These are the people who don't go back to work. And finally, how do those tracks come into this? So we know that there's a really rich architecture of cerebral cortex level. We know that there's again, that's the arch of fasciculus just to show you how complicated this really is and how much time we've spent diving into this. But what is the architecture doing? And so let's again, the triads are the main thing that happened in mammals. And what else is different between mammals and every other non mammal. Well, so ultimately, the everyone below us has a four level layer cortex and we have a six. So what happened is we have these two inner layers, those are the new layers. And what those layers are doing is that the outer layers are parallel in local processing for the most part and the inner layers are designed to make long range connections. Well, that makes sense if you're trying to coordinate activity over vast distances, get your brain into a lot of states and do higher cognitive functions. That's the whole point of our brains. Why we replaced all of our allocortex with Neocortex, which is what happened between birds and, and reptiles in us. And ultimately, it's because brain networks are important for getting the brain into all the states necessary to represent reality. What's the point? So it raises this question of what does the hippocampus do? Why is a structure like the mammary body so critical for memory? Well, what is the hippocampus is doing? Is it taking the states and putting them in sequential order? So you can memorize that, order it, then fires it around the forex, the mammary bodies and then out to the singular Gyrus where those states are rehearsed over and over again to write them into memory. So destroy the DMN, destroy the mammo bodies. This is what what happens to this is what they're doing. And that's why the mammo bodies despite not being widely connected to the brain are so important because this is a way to quickly distribute because in order to get the brain into a state repeatedly, you gotta move really quickly. So finally, I'll touch on this really quick, I think because I think this is cool. Um Interventional neuro rehabilitation is a concept we came up with uh we didn't invent T MS or doing even doing this for stroke. But it's the concept of if someone has a surgery, a brain surgery instead of just sending them off to rehab, we actually take a more active role. We get imaging, we look at their brain. We try to figure out how can we speed up the recovery? You can see these are patients who had glioma surgery. We've actually posted another paper after this since then. It's bigger. But ultimately, we found that most of these patients get some improvement sometimes pretty dramatically. This is someone who went from, uh, 05 to four or five in a week, that's pretty fast. Um This is actually an RCT of the same stuff out of Germany. What they found people, we used to have to make the caveat that we don't know whether these patients would have improved on their own. This is a number needed to treat of two class one evidence showing that if you have a hemiparesis after a glioma and the tract is not transected, the number needed to treat to permanently improve someone at three months is two. So these patients actually don't come together over time, they diverge and I'll just show what that looks like the whole process. So this is a patient that my partner in Sydney operated on. He was a young guy, wanted to be very, very aggressive, willing to take a deficit and got one Semitic left leg is mostly paralyzed. He has a little bit of abduction. This is the T MS coil, it delivers 100 and 50 pulses, a second to the brain. We've neuron navigated it. The question is, where do you put that? Because if you move it a little bit, it gives the pulse to a different part of the brain. And the other question is why is he so paralyzed? His corticospinal tract is intact? What you can see here is this is connectivity on a resting state F MRI. So just pause it real quick. The patient lays there in the scanner for eight minutes. The brain cycles through all of its states and you look at how correlated all these areas are with each other. The seed comparison is in the motor cortex on the, on the on the left and right side, you'll see that and correlated areas are in red. So we expect this whole screen to be basically a light red because the motor system should all correlate with other parts of the motor system. And you'll see that he actually doesn't do that. So if you spin it around here, if you look at the supplementary motor area on the, on the injured side, that's the right side. It's not correlated, that's not normal and he doesn't have any other reason. The tract is fine, we didn't cut it. So we gave stimulation to that area. That's after uh about two treatments, we do it multiple times a day in combination with physical therapy. And that's after five days, she went surfing about a week and a half after this. So maybe he would have got better on his own. We don't know, we can't run that experiment in the DeLorean. But what we can say is that, that's pretty fast and that means he's getting into, into chemotherapy faster. He's not picking up ad VT, his quality of life is better and the side effects of T MS are very minimal, the risk is very low. And so, uh I've been recently told that the, the group that did that previous trial has done a bigger trial with MD Anderson hasn't been released yet, but apparently it's gonna probably be FDA cleared within about a year or two. So given that this was something that nobody was talking about five years ago, it's made a lot of progress towards basically being able to do this. And this is the kind of power when we begin to really understand what's happening in that machine, be able to take actionable insights of what we can really do. And, and ultimately, you can see where this can go over the long term as we begin to understand the cognitive system a lot better. So, thanks everyone. Published July 17, 2024 Created by
BROADCASTMED