false
Catalog
Ultra High-Density Neural Interfaces for Monitorin ...
Ultra High-Density Neural Interfaces for Monitorin ...
Ultra High-Density Neural Interfaces for Monitoring and BCI
Back to course
[Please upgrade your browser to play this video content]
Video Transcription
Well, welcome to the afternoon concurrent session. This session is about intracortical high-density systems for neural recording and stimulation uses and challenges. And we have a great lineup. My name is Ahmed Raslan. I'm a neurosurgeon at Oregon Health and Science University. And here's Kelly Collins. She's one of my partners also. Kelly? I'm a pediatric neurosurgeon at Oregon Health and Science University. Yeah. So our first speaker is Dr. Sid Keshe. And Sid Keshe is a professor of neurology at MGH. He is a well-known researcher in the area of cortical physiology and brain-computer interface. He has collaborators throughout the nation, multiple publications, very impactful. And he's going to talk about intracortical high-density system uses and challenges. All right. Good. Good afternoon, everybody. Thank you for coming and deciding to spend time with us talking about intracortical arrays instead of doing like the teapot ride or whatever was your other option for the afternoon. My goal here is to start with an acknowledgment that intracranial recordings have been around for a long time. And they've provided incredible insight into how the brain works, both from a pathology point of view as well as from a basic science point of view. Until recently, and recently keeps changing in my mind as I get older, but, you know, 10 years, 20 years, depending on exactly how you want to measure it, more so in the last five years, those technologies for basically grids and strips haven't really changed for a very long time. But that's actually starting to change quite a bit. We now have a lot more capability, a lot more options for us. On this slide, I'm just showing a few different kinds of technologies that are out there. I'll talk about some of them in just a little bit more detail in just a second. And then Shadi and Daniel and others in this session will probably talk even more so in more detail. So we're now at a stage which is sort of fitting for Walt Disney. He has this famous thing that if you could dream it, you could do it. We're now at the phase where we're no longer really dreaming it, we're doing it. We have the engineering capabilities and the manufacturing capabilities to actually get the kind of devices that put us into a very different territory than we were before. So we're going to talk a little bit about some of those use cases for those devices and some of the challenges that come along with them. So I'll give a brief overview of what I would mean by high density and afterwards we could sort of discuss what that could mean or does it really matter. I'll review some of the high density systems that are out there and what they could do and then go into use examples for both research and clinically, giving examples for some of the work that's out there. And let me preface this with an apology. I'll show work. Most of it will be related to groups that I am in or collaborate with, with some exceptions. And I apologize if I'm not covering your favorite report in this domain. There's too much to cover. I'm just giving snippets to give a feel for it. I'll talk then about some of the challenges that I see in this field and a little bit of summary and thoughts for the future, hopefully leading into the rest of the talks in this session and providing some discussion fodder. So clinical and research devices used to record from the human brain now vary quite a bit in type, in scale, in form factor. And I'm not going to go through every little part of this slide, but over on the left-hand side, over here we have depth electrode type arrays, which range from macro contacts that you can get from any of a number of manufacturers, PMT, Adtech, Dixie, ones that have micro contacts on them, the now famous Benkey-Fried system that has a spray of micro wires that come out the tip, other versions of micro, here's a Dixie, which has a positionable microsystem. So that could give us high density and intracortical applications. We have the Utah Array type electrode, NeuroPixels and so on, that have high density, relatively high density down a shaft to go into the gray matter. We have thin film technologies, and again, you'll hear a lot about this from Shadi, that give us very high resolution on the cortical surface, and the same technology can also be used for various depth electrode types. And all of these can be used in various different situations. I'm showing a Utah Array set up here in a grid case for an epilepsy patient, thin film in the operating room during mapping, NeuroPixels in the operating room as well during a mapping case. And these different form factors allow us to sample lots of different areas of cortex, subcortical regions, subcortical nuclei, in all sorts of different ways. So there's lots of different ways that we can get to high density, ultra high density recordings of the nervous system. What does that really allow us to do? Well, there's a number of features. At the cortical surface, we get high resolution field potentials, that is, local activity at that surface, which is likely representing primarily synaptic activity. We also can get unit and unit-like activity from the cortical surface. That's a relatively new feature that's only been explored in a couple of papers, and what it actually indicates and what it means remains to be seen. But these microcontact built arrays, for various reasons, give us excellent response in the high frequency range, so we can get gamma activity, high gamma activity, very high activity that might be pathological, et cetera, all with these kinds of devices at high resolution. If we look at what these devices could do in the cortex or subcortical structures, we can also get high fidelity recordings of local field potentials, but now in a much smaller area and therefore much more localized. We could, of course, get the high frequency activity, the gamma and above that we get with the surface, and we could get true multi- and single-unit activity at different cortical depths and in subcortical structures, so really getting us down to the level of single-neuron action, which is where we want to be for a lot of physiological questions. And finally, we could also do high-precision stimulation as well through many of these devices, and this could be used in cortex, which is where I'm focusing most all of this talk really, but it could also be used elsewhere in spinal cord, for example, for neuromodulation, for a variety of different neuropsychiatric diseases, pain, sensory restoration, neuropsychiatric problems, et cetera. So there's a lot of potential for both high-fidelity recording and high-fidelity neuromodulation. So let me just give you a couple of examples of where this has been used and for what reason. So in the basic and translational areas, we've seen human single-unit activity being examined in a wide range of cognitive tasks, sleep, anesthesia, et cetera. Some of the most famous of that work comes from Itzhak Fried and the folks who came out of his lag, Florian Moorman and Kuroga, Gabrielle Kreiman, et cetera. There's a whole group of them that have been incredibly productive over the years looking at really fundamental aspects of cognitive activity in patients. We also have work with Utah arrays, and we've been a proponent of that along with Cathy Chavone and Zaglug down at NIH, using Utah arrays in patients with epilepsy to look at very fundamental questions in cognition and elsewhere. We also have medium and higher-density recordings that have been used for human speech analysis, and most of this work comes from Eddie Chang's lab, where they've used it very successfully to look at phoneme encoding from motor speech and, as I'll show in another slide, moving toward restoration of speech function. And these fundamental questions can also really get to basic, basic neurophysiology. So on this left-hand side, I'm showing a panel from work by Mila Halgren using laminar microelectrode arrays. These are a laminar microsystem developed by Istvan Ubert in Hungary a number of years ago that give you precise recordings of current source density and multiunits and single units throughout cortical depth, and she used that to analyze alpha patterns and where it's coming from, how it can be generated. Work here by Dan Cleary, who's probably here somewhere, I hope. Raise your hand. There he is. There he is. Okay. Using thin-film microelectrode arrays and asking fundamental questions about, what is the unit area of cortex that actually carries information? Is there a module in the cortex that's the right scale? I'll get back to this question in a little bit. So these are really fundamental kinds of questions that we can answer. We could start moving toward more translational questions as well, and this is work from Eddie's group using the neuropixel in the operating room to look at phonemic information representation in single cells in the lateral temporal lobe, and there are in superior temporal gyrus neurons that respond preferentially to different phonemes in different ways, and in fact, if you look at different cortical depths, you can find different signatures as well. So this is work showing, as we expected to some extent, that there's an incredible richness of information at this scale that can certainly be harnessed for lots of purposes. Our group did a very similar related study, but now in frontal cortex, in patients getting a DBS electrode, again with neuropixels, again showing that we can get single cell responses to phonemes in very specific ways that suggest that at that level, at the single neuron level, we have tremendous information that can then be harnessed at a later date for BCI applications and other related techniques. So what are the possible clinical uses for these new electrodes? And I should say that right now, none of these are in clinical use per se. That is, we're not, you know, prescribing them to our patients. I sort of think of this as three main domains in which we will see likely use of these devices. The first is in the acute setting, in the operating room, where we can place these in a variety of different surgical situations and get information that's necessary for that surgery or necessary for further treatment of that patient later on. We could, of course, use them in the semi-chronic environment where we're putting electrodes in for 7 days, 10 days, 20 days, particularly for mapping epilepsy, although groups are starting to do that also for mapping neuropsychiatric disease, particularly depression. So we might be able to use them in that environment to get more information. And then finally, we would think that we could start using them chronically as well for a lot of the same kinds of reasons, and particularly in the domain of brain-computer interfaces. So where is that right now? So first of all, in the brain-computer interface realm, there's increasing evidence that having that intracranial higher and higher resolution permits higher decoding. It lets us have more degrees of freedom in whatever it is that we're decoding. And I think Daniel will talk about this in a couple minutes, hopefully. We could also see this in lots of examples in, for example, movement or cursor control or speech prostheses or speech decoding, and I'll show a very quick example for that in the next slide. We also could think we could use these very high-density recordings to give us more information about pathology in epilepsy, in pain, for example, in tumors, and I'll show an example of that in a minute, in psychiatric disorders, and really explore the use of this in that OR setting for high-resolution mapping. So let me see if this is going. So this is a paper from the BrainGate group, and as a disclosure, I am part of the BrainGate group, using the Utah array, or multiple Utahs array, to allow speech decoding at very high fidelity, and hopefully. So this gentleman has a number of arrays in, has, you know, we are decoding information and getting very rapid decoding and high-fidelity output, so we could now restore speech. So it translates, reads it, thinks about it, and it translates it out, and so we're getting to a phase, a point now, where we could get highly accurate, high-fidelity, high-throughput speech decoding in a way that actually is clinically useful, and this patient and other patients who are in the BrainGate clinical trial are actually using this on a daily basis, not just in research sessions, but actually for use, for daily use. So in a sense, we're already there at the stage where we're starting to use ultra-high-density approaches for clinical purposes. We could also use this for other pathology, pathological investigations. This is work from my lab, done mostly by Angelique Polk, who's a junior faculty in our group and incredibly active in this area, using Utah array recordings in one hand, thin film electrodes in another, and neuropixels in a third, to look at epileptiform-type activity. And this work is just showing that, at the level of this scale, we see all sorts of activity that is totally not present or not clear on the macroscopic view that we get with your typical grade or your typical depth electrode, showing micro-seizures, onset of activity that you wouldn't otherwise see, and other features that we might be able to harness to better delineate seizure-onset zones or better delineate ways of actually controlling seizures. Finally, Angelique has also been looking at the role of these microphysiological measures in delineating tumor boundaries and also tumor invasion mechanisms. So the idea here is to look at the border of a tumor at very high resolution, look at that physiology and see whether you can harness that physiology to understand where is the border that needs to be resected, and why is it that the tumor is able to spread. And so we're using neuropixels as well as thin film electrode recordings over and around the tumor to map that out better and see whether we really have a biomarker there that could be used surgically. So challenges. These systems are really great in many ways, but there's a cost. So first of all, I'll go through this quickly because we talk about it a lot, but of all the challenges, it's the least interesting in a way, and that's the regulatory challenge. There are lots of regulatory steps in this process. All these different entities are involved, the FDA, your local IRB, the NIH is highly involved now, other funders might be involved, and this interacts with where are you doing it? Is it in the OR? Are you a commercial entity? Are you an academic entity? Are you some sort of mix of that? Who is this for? Who is it being used for under what circumstances? And is this research? Is this clinical? What stage? Is this industrialized? Is it not? Et cetera. All these factors come into play as you sort of go through this regulatory space, which is challenging for a couple reasons. It's just hard to do. It takes time and effort. And it's changing. It's a dynamic landscape right now, making it a little bit challenging to figure out exactly what you're supposed to do when. So lots of people at all these different institutions and levels are certainly happy to try and help you through it. It takes some effort at this stage. But there are more fundamental questions, I think, that are interesting from a clinical and a scientific point of view. And this is a big one. I know Shadi will talk about this later on. There's a fundamental design question. There is some tradeoff between altered high density and location versus coverage at this point in time. We can't cover everything all of the time, at least not yet. So what do you want to do? Do you want to have a high density in a small area for some reason? Or do you want to have small contacts that are high fidelity recordings, but do you want to have them spread out? And to figure that out, you largely need to know, what is the scale of the activity that's going to give you the most information for your particular purpose? And that's a really challenging question, because there's just not enough known about the fundamental neuroscience, as in Dan Cleary's paper, where he's looking at, what is the fundamental unit underlying language and speech processing? What are the fundamental units of whatever it is that you're trying to go after from a clinical point of view? There's also a challenge of too much data, which is not something I ever thought I would say. I always thought, you know, more data better. I always want more data. But we're getting to a point where this is starting to become a challenge. We've got now hundreds to thousands of channels. We're sampling them at 20k, 30k. That's a lot of data coming out at a given time. Not only are there engineering challenges for moving that data around, but then you could get terabytes of data from a particular experiment or recording. What do you do with it? You have to store it. That's not a huge problem. It's mostly a money problem, honestly. It's a time problem to some extent. You have to process that data. That's still a problem, even with all of our GPUs and everything. There's still real issues there. And there's a what do I do with it issue. So just clinically speaking, we've been exploring, as an example, we've been exploring the use of high-density thin films in the epilepsy domain, let's say 1,000 channels. That's about five to ten times the number of channels that we do in our typical intracranial cases now. I can tell you the epilepsy fellows who read those studies are already fairly overwhelmed by having 128 channels to read. There is no way that they're going to read 1,000 channels of EDD. That's just not possible clinically. So now we have to come up with solutions that allow them to read that data or to process it in a way that's clinically useful. That's not a small thing to do in order to move that actually into the clinically useful fashion. There's also this larger issue that combines a lot of this of what device, when, when are we going to deploy it, and how do we know it's the right device, right? So we're going to need to figure that out. And for each of these devices, we're going to need algorithms to actually use the data and make something useful out of it. And how do we know what it's going to do with that? Well, we have multiple possible devices. I have multiple different thin films now at different densities. Which is the best for a given need? And how do I know that? And how do I test that it's different, right? As everybody in this room probably knows, doing research, and particularly clinical research in the surgical domain, is very hard. It's not like you can do very large numbers quickly and compare them in lots of different ways. It's really a challenge to do that. We have to come up with new ways to try and figure that out so one can figure out what are the best devices to be using in a different environment. And then there's the challenge of trying to get all this to happen. So the way that we do it is we have a really big team or sets of teams or collaborations, and I'm partially using this as an acknowledgment slide. All the people on this slide and on the next slide I'm going to show have contributed to this work over time. We need all this to get it done. We need the diversity of expertise, the neurosurgeon, the neurologist, the engineer, et cetera. We need all these different people to actually get it done. Now that's a power in a way, because you're getting all that input, all that expertise, all that different perspective. But that has to be managed, it has to be set up, et cetera. And I'm going to take that opportunity to thank, these are not in all, many of the people we have worked with in just the last couple of years on these kinds of projects in order to get things done. And especially the folks who actually do the research for us, which are the patients. We really have to pay attention to what they need and their participation in this. And that's not trivial either. So finally, I want to emphasize two points here, one of which is partial solutions to this. And I think this is an important issue. I'm a big believer that one of the ways this field can really move forward is by being more open and sharing more as much as possible. And that means sharing regulatory information, so we're spending less time on that particular challenge. Sharing techniques and codes, so we're not constantly reinventing analytic techniques and other necessary features that might not be that interesting in and of themselves, or are in fact a big advance, but now could be applied for other folks. And can we share the data, can we actually get data out there so other people could look at it and move everything forward faster? So I'm going to put in a plug for that. And then finally, just to summarize, we're getting to a point now we've got really high resolution systems for brain recordings and stimulation that give us a very different perspective and get us to sort of fundamentals of neurophysiology in the human brain. I think these are going to have substantial impact. I think they're already starting to have impact, and that's going to just grow over the next few years in a whole bunch of different domains. Of course it's really getting us to the fundamental level of how the brain works, but there are challenges in using this, setting it up, advancing it, moving it into the clinical sphere, which I think are easily, not easily, but they are overcomable in many ways and will let us really go forward. And I hope in the next few talks you'll see actual applications in various different ways of where we're using this. Thank you. So Sid, why don't you sit here, we'll do the questions after the first three. So we're going to go through the first three talks, and then we'll hold all the questions for after our first three presenters are done. Coming up next, I'd like to introduce Dr. Shadi Daye, who is a professor of electrical and computer engineering at UC San Diego. He is a pioneer of multi-thousand channel thin film electrodes used in human brain recordings, and is an awardee of the NIH Director's New Innovator Award, NSF Career Award, ISCS Young Scientist Award, and was a J.R. Oppenheimer Fellow at Los Alamos, and he's here to give us a more detailed engineering perspective of high-density thin film growths. Thank you so much for the introduction. It's great to be here. Good afternoon, everyone, and Sid, thank you for giving this overview on progress, challenges in the whole ecosystem of brain recording and modulation. I'm going to hit more on the question of what resolution and what coverage we need and how do we address that. We are all here because we are interested in understanding the brain better. We know that the tools that we have today give us a blurred vision when we want to record, measure, modulate the brain, and that there is a need for a new generation of technology that will allow us to look at the local dynamics, and that requires very high resolution, and that resolution to be extended throughout the region that we are measuring for the brain, possibly throughout the brain, and that requires high channel count. Now, going by the Mountcastle view of the organization in the brain, we have about 150,000 cortical columns that are continuously predicting incoming information to predict, to make sense of this information and to predict the next action. And going by the Nyquist frequency of picking up that information, that means we need about 300,000 electrodes to map the brain. If you want to map even finer features like directionality, then we multiply that by a factor of two or four, so we need about 600,000 to 1.2 million. If you want to record macro columns, that decode features, there are about two million of these, so we need four million electrodes, but there are fundamental limits. We don't have access to all of these cortical columns, so it's not really practical to map the whole brain. Also, when we measure the brain, we take out energy from it, so there is energy deprivation with any modality that we use. And that modality also is that measurement, it changes the brain state as we are doing it. Different regions probably need different density of recording electrodes. So what is the density? It really depends on what type of performance we need for a given application. And the best thing to do is to have a high density that offers wide coverage and try to figure out which regions of the brain would decode activity in units that are packed in high density and which regions that need lower density. So we go again to Dan Cleary's modules for language processing. This is what Sid talked about. The average module size is about 1.8 millimeters in diameter and they have about 200 micron short transitions in between these modules as we see on the far left. So the smallest module that we have measured here is 1.2 millimeters in diameter and therefore we need about 600 micrometer pitch in order to record all the units without crosstalk. But there might be smaller units, this is for language processing only, and there might be smaller units that would require even higher density. And so we started looking at this problem in a number of animal species with high density electrodes and today I'll report our most recent findings in this area. This is a very well known experiment, the whisker barrel experiment that has very good organization in the brain so we can test high channel count electrodes on rats, on mice with this whisker barrel experiment. Air puffs are applied to individual whiskers and that evokes activity in barrels that are limited in size. So across all of these rats right here we find that there are modules, these are high gamma map modules corresponding to each whiskers that extend about half a millimeter in extent. When doing post-mortem histology we found a one-to-one correlation between these functional units and the high gamma map units. More recently we packed 1,000 channel grids in one by one millimeter squared so we have another 1,000 in a sparse density with a contact-to-contact spacing of 150 micron, that's what pitch means. And then in the center we have the high density where we have 31 micron contact-to-contact spacing. And this uses the platinum nanorod contacts that have low impedance and allows us to scale this to a few microns without suffering from noise issues. This is placed similarly in the rat whisker barrel experiment. Put it over the barrel cortex, these are recording from the 2,000 channels. We see responses, we see some traveling responses or dynamics on the surface of the brain as I'll show you in the next few slides. So the way we do this is we take several tons of trials, in this case 60 trials, we look at the trial average, we do high gamma filtering, take the Hilbert magnitude, look at the maximum in the Hilbert transfer magnitude and then identify within a few standard deviations the most responsive regions for that specific stimulation that we are doing. And so by doing that we can now localize the barrels for these different labeled whiskers and each pixel here is 31 microns so we can see that there are transitions that are pixelated even at the 31 micrometer spatial scale. Also here we mark the region where the electrode was implanted and then we did post-mortem histology and again we have one-to-one agreement with the histology. Now if we had a lower resolution grids to sample this activity we start to lose the boundaries that we were able to see with the high density grids. Now going a step further to decode directionality, we can do that by doing the air puff from different orientations on the whiskers and this different orientation encodes within the whisker barrel different regions that are going to be responsive. So we did here three directions and you can see as the response is played that for some directions the response starts in the top left whereas for other responses in the bottom right and we identify mainly three zones that we suspect that these are macro columns. We couldn't verify that by histology, it's only by high gamma activity at the moment. Now again if we down sample this data to the 150 micron pitch, we can still find a global responsive region but it would be hard to identify the macro columns and therefore it would be hard to identify the dimensionality. Now what are these recordings that we do from the surface? Are they really from within the cortex? So we had an electrode that has a thin depth penetrating electrode in it that penetrates the cortex and this sits on the surface. We carried out the same mapping approach and as we simulate different whiskers, we see the movement of the response as we saw in the previous slide but for specific whiskers particularly here C3, a little bit D3, we see that the response is stronger only at the C3, C2 region. And so we suspect that this depth was in between the two cortical or barrel columns of C2 and C3 and this activity is coming from within the cortex itself. Now we went to looking for columns in a higher animal model, in a larger animal model than the pig. This is a 4,000 channel grid placed on the somatosensory cortex while we are also delivering air pulses on the pig's snout. We see also somatotopic organization that follows the topography of the snout, very well arranged over the somatosensory cortex. Now to look at higher, with higher resolution at this activity, we packed 3,000 channels in one by three millimeters squared at the center of the grid as you will see here. This is also with the 31 micron pitch, the system that it was used for and then we were delivering air pulse to individual small vibrocy or hairs on the pig's snout. And we wanted to look for the smallest possible response basically from the surface of the cortex in a different animal model. We have online visualization that will allow us to align this to the responsive regions. And now here is a snapshot of the responses. We see that there are small dynamics that appear in the one by three millimeters squared in addition to the large module that we have observed for the somatotopic organization in the previous slides. If we didn't have this resolution, if we don't have sub-millimeter scale electrode, we would have missed that response. So this is a snapshot of how it looks like and then a little bit more dynamics of the responses that we observed within this small box. And so then we did 25 simulation points going from the top left to the right. These are the spots that we isolate for these specific simulations. And as we go down spatially, the spots move slightly upwards. But we could not isolate with clarity columns as we have isolated in the whisker barrel experiment. And lastly, also we wanted to show that this surface activity stems from a cortical column or from beneath the surface of the brain. So we inserted an SEG electrode. I'm sorry, so we inserted an SEG electrode as we see here in the lower left under the grid while we are delivering simulations to the snout. And we see responses on both the surface and the depth electrode in the same regions. The depth electrode was slanted to hit only within two millimeters from the surface. And so if we look at autocorrelation with spacing from these experiments, we see that even at the 31 micrometer limit for different frequency bands, we don't have a one correlation or any correlation between the contacts. So really, even at the 31 micrometer, we have intimate contact with the brain. There is no separation, no CSF shunting. And of course, we have good electrodes. We can isolate different activity between nearby channels. So where do we go from here? Well, this is to put this experiment in context that we've done for language mapping earlier in Dan's work. The area was 0.3 by 1.2 centimeter squared, 1.3 centimeter squared is very small. The language processing happens across five different modules that we know of today. If we want to cover all of these modules and we want to retain the same pitch of 200 microns, we need to build a grid over six by six centimeters squared that has 100,000 channel electrodes. So that is the next step for us in the lab is to build this technology and to develop its software visualization analysis and so on. This work was done mainly by Ji-Hwan Lee who led the FAB and the animal experiments as well as Tara for instrumentation and analysis. And it builds on collaborations across many labs. Some of them we saw in Sid's presentation include Sid's lab, Dr. Aslan's lab, and many others at UCFD, at MGH, and at OHSU. And we are funded by the NIH to carry out this work. So thank you for your attention. Look forward to discussions. Thank you. Okay, thank you, Dr. Daye. Next, I'd like to introduce Dr. Daniel Kramer, who's a surgeon scientist at the University of Colorado Anschutz. He is a stereotactic and functional neurosurgeon who treats movement disorders, epilepsy, and facial pain, and co-founded the Neural Engineering Research and Design of Colorado Lab where he uses high-density intracranial recordings to study executive function and movement disorders. Thanks so much. So my area that I'm gonna talk about is use of some of these devices across VCI. So broadly speaking, very, very broadly speaking, brain-computer interface is a brain with a computer and the interface. And I think oftentimes we, since there's been a lot of focus on primarily motor, we forget that this really started in visual and auditory implants. And when we think about it that broadly, the applications are very wide-reaching and largely unexplored. So nearly every sensory system is still at sort of the early stages. And then executive function, emotion, memory, there's the things that people have tried in various settings, but that we haven't really quite applied some of these new technologies and some of the concepts that we have from brain-computer interface for. And then even within motor, there's still a lot to be worked out. Classically, a problem across many of these systems is stopping, holding, and selecting. This in the bottom here is some very nice neural responses to movement. And so there's pre-movement activity and then movement activity. And then you can see it gets pretty messy at the end there. So there's still a lot of those signals that are just probably not being represented in the places we're recording. Thankfully for us to make some of these advances is a lot of new devices, which we've just been hearing about extensively. And these are really some amazing technologies and they give us high-density recordings. And high-density recordings for pretty much everybody in this room sounds great because more equals more. The more neurons we can get, the more areas we can cover, the better at least we think we can do. As a reminder, there's 80 billion neurons are estimated anyway. And so quite literally so far, we are just scratching the surface. But also these different technologies have specific features and these specific features give us expanded possibilities. And so I just wanna go through some of these and talk about sort of where they have their potential best uses across not just motor BCI, but really any applications that we can sort of think of in the future. So the first one and the most straightforward is just higher density coverage. So the classic has been Utah array with 96 channels as sort of the max and we can put several of them in, but there's companies now working on very high-density arrays as you can see here. I always think this kind of looks like a cat hairbrush, but it had the very high channel counts. And this is the most straightforward, but gives us these high neural populations, which in recent years has been very clearly sort of, if not the way the brain communicates, closer to the way the brain communicates, which is that you can reduce all of these high density into different neural dimensions and those dimensions, low dimensional spaces are really what's covering the activity to execute some sort of action, usually motor action, but really anything. It's also of the least clear importance, which there's a trade-off of extra degrees of freedom, but there's also overfitting that happens. And there's an open question of how many neurons at least in the same region improves performance. There's been some of the early work that sort of said that it saturates around 50 neurons, although that's sort of under assumption that you're in the same population. So if you can get into other subpopulations into as we've heard about sort of different micro columns and things like that, there might not be a saturation. But obviously if you're putting more high density into the same regions, you might run into that problem. But there's some very clear applications for this. The first and foremost would be on anything that's somatotopically organized. So in the bottom left there is some of the early work in Utah arrays for somatosensory restoration with stimulation. And you can see there's a pretty small area that gets covered on the hand as you can expect from these very small arrays. And so having an expansion of that would give you sort of more concentration within that area. And then of course, things like the somatosensory cortex are not just one area. So there's areas for proprioception and tactile feedback and things like that. And so getting sort of a higher local coverage might be able to give you access to some of those different features. Next conceptual approach is broader coverage. This of course, you can use Utah arrays or really whatever. And this is definitely the approach that Neuralink is taking, which is that you can cover with high density, maybe the same density arrays, but multiple different brain areas. And this for sure is gonna guarantee that you're gonna get higher subpopulations, which should give you higher signal to noise ratio. And you'd be able to use it across different aspects of whatever you're trying to decode. And also this gets at fundamentally how the brain works, which is network communication. We're sort of lucky that motor and sensory are so densely communicating and they really focus on sort of one activity, but most of the brain does not work like that and is involved in very dense network communication. And so you're able to sample from there. And then you can also sample from some of these domain general regions, such as the DLPFC or the SMG, things that are involved in just about everything as pretty much anything that we've done, we're able to decode pretty effectively in the SMG. And so you can get a lot out of a small area. This is great for the classic closed loop application. So the M1 to S1 closed loop, where you have movements, you receive feedback, you send that feedback up and stimulate an S1. But of course you can do that across lots of different potential applications. And so here you can actually, by having broader coverage, use distributed and combined biomarkers. For instance, perhaps you have motor activity in one region, but you're also getting error signals from another region that whether or not that's something that's directly related to the motor error is still something that can be used across all of BCI. We don't have to really know what we're decoding. We just have to be consistently right in decoding. And then this is great, of course, for the sort of holy grail of some of these highly distributed processes like executive function, attention, memory, which are clearly across multiple brain regions. This is some of our early work that we're doing with decoding executive function task, a Stroop task. I won't get into too much detail, but you can see a cascading of temporal activity across multiple different brain regions. And unsurprisingly, the DLPFC and SMG are sort of give the best decoding strength across this executive function task. Next up is these neuropixels or really any linear array that's got a high density and is tiled in their arrangement because that lets you sample across the layers. And of course, these layers are doing something different. Sort of at least our current iteration of what's happening across these right now is that you're getting cortical to cortical and sensory processing communication that tends to take place in the superficial layers. And then likely action selection and subcortical communication in the deep layers. And this is in the bottom there is some decoding of the different regions based on LFP. And so it gives you this layer specificity. And this is a sort of early potential application for this would be to get at some of the most difficult things, which is telling the difference between preparatory activity or internal manipulation from actual action selection. If that's taking place at different layers, then you can start to get at that. And be able to move in your decoding from thinking of moving, preparing to move to actually moving, which if you're really gonna use this in the real world is critical. These linear arrays also let us get at sulci. So 60 to 70% of the cortical surface is in the sulci. And most of the technologies that we have right now have to sit on top of the surface for lots of different reasons. But something like a Neuropixel, you can actually send along the gray matter down a sulcus. You know, a clear application for this is of course, famously true primary motor and the majority of somatosensory cortex are actually in the depth of the sulcus. You can also get at these deep structures. So iMac, the company that makes Neuropixels is working on a four centimeter Neuropixels probe, which gives me palpitations about thinking about putting that in. But if we can get it in there, then we can get subcortical recordings from all of the critical structures that are beneath the surface, hippocampus, striatum, STN, even something I don't wanna say as simple as, but as the VIM. This is some of our work on sort of functionally what the VIM is doing. And we get these very nice tiling in between the black lines there of neural activity around stopping. And below that is some mouse work from the deep nuclei of the cerebellum, which really shows that probably things like online correction and stop signals are coming from these subcortical regions. Then cortical surface, we just heard a lot about this. So these are the thin film technologies. And in addition to some of these somatosensory applications, there's also traveling wave applications in memory and recall, some really interesting work there. And then getting at this microcircuit function, the brain is sort of, the way we see it now is sort of high redundancy, which is good for us and good for decoding, but there's very likely to be some closely related, but different function of some of these microcolumns. And then obviously this is ideal for anything that's somatotopically organized, such as motor cortex or somatosensory, where even if we could get the coverage, we can't actually implant those into the surface without destroying it. So laying something that has great coverage and is long and thin is sort of our ideal methodology. And then, I kind of want to get into finally the combined approach. A lot of these different companies, as I'm sure you've heard about, are making some of these different devices. But really if we're going to get successful real world use of BCI, we probably need to do combinations of these. So I always think about this as sort of the Terminator and there's lots of things you have to do in the real world. When we think about a lot of BCI applications, really amazing ones, oftentimes we have to plug them into a highly constrained environment. But if we want to send people out into the world, we can't really do that. They can't plug into a computer or a word decoding system. They need to sort of go out and interact with everything out there. So you need to, there's multiple things you need to do. First, you need to select something for manipulation using attention essentially. Then there's an act that is involved in mental manipulation or action planning, whether that's conscious or subconscious. Then there's an action execution. And then there's feedback to that, including online feedback and then post-event feedback. And this is going to take, I mean, I'm not saying that we're potentially ever going to do this, but if we're ever going to get close, we're going to need a combination of approaches. So we need multiple brain regions, including some of these domain general areas, layer specificity, soul size, stimulation, all of these things are going to be critical. So I just want to kind of end on some of the needs as we have all these great technologies, but there's still a lot of open questions and just some very logistical things that we're going to need to do to be able to implement them. So the first need is we have some of the things like thin films that are great for decoding across some of these somatotopically organized areas, but we need some way to get stimulation back into them. On the top was the Utah rays, in the middle was using a, what's called mini-ECoG, so just a sort of tighter space clinical ECoG, but we can stimulate through that, and we got really nice coverage of the hand through that. And obviously we get amazing coverage with these thin films, but we need to be able to stimulate back if we're gonna use these in a closed loop way. We also need structural stability for these electrodes, for chronic recording. I mean, the obvious example is Neuropixels, which are amazing, but have not yet been designed in a way that we can sort of safely leave them inside of a human for a long period of time. People do do this in animals, but we don't really have any clear method of how we're gonna do this, certainly for humans. And this, but this does apply across a lot of these systems. They haven't been tested chronically with the exception of Utah rays and a few others. Even some of these newer systems that are very similar to Utah rays just really don't have the data yet. And then same thing, real world data transfer. So we've got a bunch of issues that we need to deal with. Some of these are probably simple, but they have not been solved yet. So heat dissipation from very high density recordings, battery life, you know, there are rechargeable systems, but we don't really have a good sense of how much this would use to have someone go out and use this in day-to-day life. And then data structure and data storage, how we'd be able to, you know, interpret this data across large swaths of patients and be able to use that on a day-to-day basis for them. Okay. Thank you. Thank you, Dan. Can you have a seat here? All right, so I'm gonna take five minutes for questions because we have 10 minutes at the end. We'll just break them into five after the press. So this first segment was about high density electrodes and the various form factors, the intracortical, which is mainly Neuropixel. It's not the only one. There's Neuralink electrodes, of course, but we couldn't get them to talk on a conference like this. There's also surface high density, which we talked about the platinum nanorod, the micro ECOG. There's also other form factor from precision neuroscience. There's a lot of other speakers sitting in the audience that could answer questions. Dr. Peter Conrad here, Dr. Dan Cleary. So we'll take one or two questions about high density electrodes before we go to the abstract sessions. Please. Hi, I have a question about the regular 48 and the 10 to the 22. Is that traded? What's the question? Yeah, particularly for the 48, you know, based upon if the like, you break into the four and then it's not necessarily the two again. Yeah. So did you hear the question? Everybody heard the questions? The question is about regulatory and ethical issues with penetrating intracortical electrodes, especially when it comes to stimulation. So just to clarify what the question is, right? We have electrodes that we put in as many people in this room do for their living probably, you know, for clinically indicated reasons, right? I think you're probably asking what about these more research electrodes like Neuropixels, for example? So all of the groups that I know of that are doing them, and so I'm going to speak mostly for myself, are doing them under IRB protocols. In some cases, we have non-significant risk designations from the FDA, or we have an IDE for doing this. In all of the cases, we're putting the electrodes into regions that are going to be instrumented one way or another. So either they're going to be resected tissue because they're coming into the operating room to get a tumor removed or a focus of epilepsy removed, or in the case of the deep brain stimulation, they're about to have a cannula placed and it goes into that same trajectory. So we're not putting them into tissue that is just going to be left in as a safety measure, basically. From a stimulation point of view, our group has not yet started to stimulate through those kind of electrodes, but I don't see there being any real problem with it. We have a lot of data with Shadi could talk to and others could talk to about the safety factors that go into stimulation from a point of view of causing tissue damage. So we know what those parameters are and we could stay well within them to avoid any injury. I don't know if that answers what you're getting at. Yep. Dan, and then Chad. Yeah, I can tell you regulatory issues, the things the FDA is concentrating on, whether there's these new technologies and sort of being able to use them, assuming that there is a significant risk designation, sort of what they want to see. It's biomechanical, so is this going to break? Is this going to, piece is going to come off of this, things like that, biocompatibility and sterility. So can, in the sterility is sort of, can you sterilize this safely, especially when you have some of these technologies with lots of essentially nooks and crannies that you need to show that this is able to be successfully sterilized so you can use it in there. And then of course, biocompatibility, it's interacting with CSF, which the FDA sort of assumes is to be kind of the highest risk in blood as well. So it's kind of the highest risk possibility. So those are the main things that they want to see. From the stimulation standpoint, I would echo the same thing that Sid did. We clinically put lots of current into people through DBS, through mapping, things like that. And so actually a lot of the things that we're using, Utah rays and stuff, where we put these small amounts of stimulation, we call it micro-stimulation because it's much, much smaller amounts. So for the most part, it should be safe, but there's always issues of kind of heating and things like that that you also have to answer for. Shadi? So, Sid. I just want to say that again, this is one of the points that Daniel raised. Again, that's why I think a lot of this kind of thing should be shared as much as possible. It actually is not cheap from an academician's point of view to get sterility testing done, to get biocompatibility testing done. If I've done that for device X and now someone's making or planning to use device X plus something, and maybe got a little bit different to it, as much as possible, I'd like to see them use what I did so they don't have to completely redo that and we know we're in the same regimen. Now, that's not always easy to do and there's lots of silos and blah, blah, blah. But I think that as a field, as a community, we could be doing better about that kind of thing so that this moves along faster and safer. So, Shadi, your response and then. Yeah, just an added comment. I think that the question has been answered well by colleagues here, but over the long run, these types of electrodes will cause less damage in the brain because of their smaller form factor, their smaller volume. And most of the technologies that we are working on, I'm doing also depth electrodes. We're going through the FDA, IDE, or 510-CARE-OUT to get approvals before we use them in humans. Excellent. Okay. One more question and go. I have a question here and we need to talk about how this will help. What I told you is that I'm in a part of the field. I'm in the field for the only concerns and what that will give us. The other one is the same. Let's say if you take a simple example, a prediction on this, having predicted 140,000 volumes, what I've done with it, and then what is missing from the electrodes and what is important to that for, and then get rid of the number, what will be good to have? We need more content. Yes. I think the first answer is it depends on the budget, how many electrodes we can put in that area. The question is, you know, any new thoughts on how many electrodes per unit area ultimately we need to put? What's the size of the cortical column and what is the spacing between these contacts? And this is truly still a subject of research because different areas of the brain have different organizations. For organizations that we know that there is somatotopic columnar organization, these columns are at about one millimeter in extension. And if you just want to pick up if there is activity in that column or not, that is, let us say a certain muscle is being contracted or not, without information of how hard the muscle is being contracted, you know, how much percentage of the muscle is being contracted, then you need two contacts per column. But if you need further, for example, information about the strength, then you need more contacts so that you have higher dimensionality in the decoding. We are seeing columns or macro columns that are as small as 100 micron right now from the surface of the brain. And in humans, we haven't done the very high density recording yet. We are getting FDA approval to be able to do that. But in smaller animals, we're seeing that even at sub 100 micrometers, we have boundaries that could be delineated from the surface. Excellent. I think we're going to take a pause on the questions. We have five minutes for the end and we'll go to the abstract session. We'll start by, you can see, by Brian Chung. He is a student, I assume, from Brian Lee Lab. He's going to talk about beta AEG power modulation in the human orbitofrontal cortex during a go-no-go reaching arm task, so. All right, honored to be here today and share some of our work. So the title of this project is beta AEG power modulation in the human orbitofrontal cortex during a go-no-go reaching arm task. So the orbitofrontal cortex is traditionally known for its key role in processing emotion and in the representation of reward value. However, there seems to be some emerging evidence that the OFC could also play a role in motor related activities, such as motor inhibition and impulsivity. Beta band neural activity within the 13 through 30 frequency range is thought to represent a city-state variable that promotes some type of motor status quo. So in other words, increases in beta band power or beta band synchronization is thought to reflect the maintenance of a motor state, while decreases in beta band power or beta band desynchronizations are thought to reflect a transition to another motor state. Our lab previously examined beta band power modulation in quantum mechanics and in corticolimbic brain areas, such as the hippocampus and the amygdala during motor execution and inhibition. And so we sought to use a similar experimental paradigm of a go-no-go center-out arm reaching task to characterize beta band power modulation in the human OFC. And so for this experiment, we had eight adult participants. They were diagnosed with refractory epilepsy and they underwent stereo AEG implantations with electrocontacts in the OFC for seizure monitoring. And we asked these participants to perform 64 trials of a direct reach go-no-go task with their right arm. And so this is our task. We elected to use the go-no-go task to isolate the effect of movement execution or go trials and inhibition no-go trials on beta band modulation within the OFC. And so our task involved three separate phases. The first was the inter-trial interval in which we asked the patient to wait for the fixation dot. The next phase was the fixation phase in which the patient points their right arm to the fixation dot and fixes their gaze to it. And then the third phase was either a response movement execution or go phase in which the patient would move their right arm to that target, that gray target dot, or a no-go phase in which the patient would just maintain their arms posture. And you can see a clip of a patient performing this task here, the go phase at the top, no-go at the bottom. And so for our analyses, we recorded beta band neural activity using stereo EEG. We calculated the power spectral densities through a multi-taper spectral analysis, and we performed some statistical hypothesis testing to compare the response and fixation phases. And so when we did those comparisons, we identified significant beta band power modulation and go and no-go conditions. And so on the top row here are power spectral density plots representing the frequency domain. The bottom row here are spectrograms representing the time domain. And what we hope you can appreciate is that there's a decrease in beta band power in the response phase of movement execution trials while there's an increase in beta band power in the response phase of movement inhibition trials. And so what we observed was that beta power differentially modulates in go versus no-go conditions. And so the left here is a power spectral density plot just overlaying the responses, the go response and the no-go response. And over here is average beta power with fixation, go, and no-go. What we hope you can appreciate is there's this differential modulation in beta band power for these two conditions. However, we did identify in a couple of patients that there was some variability in the degree and the direction of beta band power modulation. So on the left, in one patient during go trials, they exhibited an increase in beta band power. And on the right, another patient during no-go trials demonstrated a decrease in beta band power. And so to conclude, to our knowledge, this is our first report of stereo EEG beta band modulation within the human OFC during movement execution and inhibition. Beta band power modulation in the OFC may contribute to motor processing. However, the degree to which it does is still unknown and not yet clear. And finally, there seems to be some variability that exists within this signal. We think some of this variability is attributed to the underlying disease and there's literature to support that. We also think that it may, some of it, the medications the patients were on could also contribute to this, which we did not control for. And so for future directions, we're curious to better characterize the relationship of beta band modulation in the OFC to the rest of the motor network using coherence analyses. And finally, we want to better characterize the functional significance of beta band modulation within the OFC. And so I'd like to conclude with acknowledgments. None of this work would have been possible without our patients who graciously volunteer their time and collaborate with us in this work. I'd also like to thank Dr. Brian Lee for his mentorship throughout medical school and research, and then Roberto for his contributions to this project. And then finally, the rest of the USC Neurosurgery Department for the support. Thank you. This time, we'll take a question. Okay. So any questions? I do have a question. I think you said, yeah. So what's the cut time latency of this beta suppression? It seems to be suppression. Yes. And when it relates to action, is it precedes the action during the action? And is it site-specific or it's bilateral? Yeah, this is a great question. So I didn't include that analysis here, but we've performed a laterality analysis. We saw it bilaterally within the OFC. In terms of the latency of the response for the beta band suppression, we see it as the patient is in. So for inhibition, if the patient doesn't move their arm, we see it within a few seconds of the time start of the response. When the patient does move their arm, we also see it within a few seconds. I was just wondering if you could say a little bit more about why you would expect the OFC to be playing a role here, particularly in the inhibition. Is there some particular reason why you think the OFC would be part of the circuitry necessary for motor control in this environment? Yeah, no, that's an excellent question. I think the intention behind this study was really to, I guess, explore other deeper cortical structures. And so we started with hippocampus and amygdala, and we saw this pattern of beta band modulation. And the reason we chose those two subcortical structures originally is because there seemed to be some literature saying they're structurally connected, and there was some fMRI studies that showed some motor-related activity. And so we wanted to look at another cortical limbic structure and see if there's a relationship between these structures. And so that's kind of why we want to do these coherent studies next to better elucidate. Thank you. Thank you. Thank you, Ryan. Okay. And I'll introduce our next speaker, Dr. Jan Vesper, who is a professor of stereotactic and functional neurosurgery in Dusseldorf. And he is going to be speaking to us about real-world outcomes with DBS systems awake versus asleep placement for Parkinson's disease. Thank you very much for this nice introduction, and thank you for the invitation. And I want to briefly introduce you to a sub-analysis of an ongoing registry regarding real-world outcomes with directional deep brain stimulation in Parkinson's disease. We were ... That is an out-of-U.S., more or less European, registry focusing on Parkinson's disease, and which are typically conducted in patients while they are awake. However, there's an ongoing need and interest to also operate patients under general anesthesia due to their preference and also to advances in imaging and also in new physiological technologies. And those patients were operated with a directional deep brain stimulation, enabling us also to shift the electrical field and with those directional leads. And we compared out of this registry a cohort with awake and with one under sleep conditions. So the idea of this registry is quite simple, to include patients with Parkinson's disease eligible for deep brain stimulation surgery. And we had two PIs, one friend and colleague, a neurologist from Kiel, Germany, and myself as a PI for neurosurgery, and to understand what is a long-term outcome in this real-world cohort after one to three years. Meanwhile, we do have roughly 1,000 patients included for the one-year outcome. So that's what you can see here. And 700 were operated on the wake and roughly 300 under sleep conditions. And interestingly, this not-matched group is quite nicely comparable in terms of their baseline characteristics regarding the age, gender, and especially regarding the motor situation prior to surgery, the off conditions, and also duration of the disease. And the PDQ39, all those three conditions were quite comparable, so that even if it is not a match group, I thought that this might be comparable in terms of outcome. So the outcome for deep brain stimulation usually is measured by UPDS Part 3. That's the most important thing, and what you can see is that they significantly improved after this is a one-year result, but there's no significant difference, actually no difference among the groups, sleep or awake surgery. And this is also true for the PDQ39, improvement of quality of life, which significantly improved in both groups, but not among the groups. And so there, in conclusion, we could prove that we have a good outcome in both groups regarding motor function, quality of life, and this is also true, and Dr. Chou did this for the global impression of change. And with those patients who received lead-steering DBS procedure under awake and asleep conditions, it is under general anesthesia. However, general anesthesia doesn't mean that they're sleeping all the ways. There are conditions like in our center where we reduce the level of sedation in a way so that the patient's still asleep but is able to undergo microelectrode recording or also motor potential recording, which are very robust also in patients who were fully asleep. However, that's still a limitation. No RCTs existing so far comparing asleep versus awake, and that will be anyway difficult to run such a trial in the future. So a real-world data might be convincing for the future to get a higher market penetration in general for deep brain stimulation. Thank you. Yeah. Thank you for this question. I know that they are different to your registry or your studies you're running last year or presented last year. These were just STN patients first, and second, we had 20 centers in this registry. Exactly. Welcome. Can I just do a quick follow-up to that? Was each site generally an awake site or on a sleep site? Because that's been my experience. Basically, yes. That is interesting within the last year. So there are sites who committed themselves to 100% sleep, that's our site, other way around, centers who still stick for the awake, asking the colleagues why they stick for awake. So we had a similar discussion on the WSSFN a couple of months ago, and it's more or less for academical reasons, to tell the truth. The microelectrode recording were mostly done, performed with the awake surgery sites. And the apleep surgery sites used either motor walk potential and microelectrode recording as we did, but mostly just imaging confirmation. Can I follow up on this question? State again. I think this comparison, just for the audience, they're not actually trying to measure the effect of anesthesia on outcome. This is not awake versus sleep. The question here is that physiologic-guided DBS versus image-guided DBS. Exactly. So you won't be able to compare this as well, because there are sites who performing imaging adaptation because of the new imaging modalities, so nicer MRIs, but also running microelectrode or so awoke potential recordings. So it would be hard to compare anyway. Yeah, I wonder if the registry gets bigger, you may want to have like four components, awake with physiologic, awake with imaging, sleep with physiologic, sleep with imaging, because that gets the answer. Right. But my, let's say my expectation is there will be no difference at any point. So, that would be even more harder to compare surgeons with experience than without. Any more comments on this topic? All right, I think the last, but not least, I think we have the... Ghassan Makul, who is an MD-PhD student at Vanderbilt University, and he's going to be speaking to us about suppressive motifs in peri-ictal functional networks. From Dr. Engelot's lab. Hello, everyone. I'm very grateful to be sharing the stage with all the speakers. So, I'm Ghassan Makul, I'm in my second year of my graduate studies in the BN lab under the supervision of Dr. Engelot, and today we'll be discussing some work which follows up on previous studies of network signatures that mark the seizure-onset zone and its relationship to broader areas in the brain. So, I think it'd be fair to set up some background here. So, in prior work, we found some electrographic evidence, followed by some stimulation evidence that may have supported the idea that the seizure-onset zone experienced tonic inhibition in the inter-ictal period, so the period between seizures. And so, we then coined this as the inter-ictal suppression hypothesis, and then this work was primarily led by Graham Johnson. And so, the natural question is, if there's a resting state network configuration which could be responsible for inhibiting seizure-onset zone activity, is a relaxation of this network state then responsible for the permission of seizure activity? Or are any of these network features at all responsible for mitigating seizure activity? So, we wanted to extend some of these network analyses into the peri-ictal period and discover if this network motif held, broke, or had any role in the seizure period. So, I think it's worth calling out how immense this data curation effort was, so I wanted to also give a shout-out to Derek Doss and Graham Johnson for what I hear was the summer of data before I arrived into the lab, where they manually curated a lot of these early SEG recordings, and I kind of came in and finished off the curation of the data set for this project as well. But this is a lot of manual auditing, labeling of the peri-ictal periods, and scrubbing of data, ensuring that we're free of artifacts, ensuring that an FIS seizure has actually been tested for impaired awareness, so on and so forth. And so, because of their efforts, I was able to analyze 75 patients and aggregate across 686 seizures, so I just wanted to briefly shout that out right there. And so, the methods of this paper, or this investigation here, which is a preprint on BioRxiv, start with partial directed coherence. We can kind of get into this a bit. This is more explicitly discussed in the actual inter-ictal suppression hypothesis paper, but I'll call it a few qualities of partial directed coherence. It's a phase lag measure, so if you look here at this example seizure onset zone, partial directed coherence would pick up on the perturbation in this lead traveling across these as a function of time, right? So, anything like muscle-like movement or instantaneous perturbations would not be picked up as connectivity here. On top of that, partial directed coherence does decompose across frequency bands, but we found across several bootstrapping methods and analyses that the signal is quite global, possibly implicating some broadband activity. So, we restricted our analysis in the original ISH to, I think, just alpha and gamma, and here we'll just stick to alpha. These patients are coming into the Vanderbilt, the VUMC, for their workup of epilepsy here for medically refractory epilepsy, and so they have stereo EEG leads that correspond to the various regions of their brain, but a priori we do not know if these are seizure onset zones, which I have here in this example labeled as red, if they're propagative zones, which are channels which pick up the first 10 seconds of seizure activity but are not responsible for initiating the seizure, or if they are what we deem as non-involved zones, at least from the reports. We take these recordings, and then we apply partial directed coherence as a network measure to them, and we have these matrices which span the periods that I've defined here as the interictal period, which is up to hours away from the most recent seizure, preictal period, which is up to the minute before seizure, the early ictal and the late ictal period, which are the first half and the second half of the seizure, and then the postictal period, which is the immediate 10 minutes after seizure. So one thing that I think was important to also mention here is that we had to align across seizures of different length. So what you'll be seeing are network trajectories and summaries of these connectivities, but know that the first 10 seconds of network activity is preserved across all seizures. So all 686 seizures, 10 seconds, are aligned in time. The last 10 seconds are aligned in time. The middle window has some aggregation periods, but I think it's important to note that there's no loss of data across all these seizures. The other thing is we'll be looking at inward-outward connectivity metrics, but we'll mostly move to the net connectivity. So if I'm talking to you, that is me outwardly connected to you. If you're talking to me, I'm receiving that connection. So as that conceptual framework, we'll just move right into the results. So the first thing we notice here is for inward connectivity on the top, you'll notice that I have the region segregated by a band, or the area segregated by region. So the SOZ is in red here, and you're seeing 95% confidence interval plots of the inward connectivity on the top, and then outward connectivity on the bottom. So just sticking with the SOZ here, in the intractal period, we see pretty much stable activity, possibly a shift right before the preictal period, but really not much of a preictal warning sign here, at least not one that was statistically significant. So interestingly, we see a sharp rise in inward connectivity of the SOZ, a precipitous fall, and then a postictal depression, or a network collapse. So in the postictal period, we do often observe changes and even suppression in certain activities, but this network collapsing entirely was quite surprising to us. Another thing that we found quite surprising was that there's no preictal warning sign. So in net connectivity measures, we found that we could distinguish the SOZ across pretty much all the periods, and that it seems that when aggregating across these windows, that the SOZ does enjoy the most inward communication during the periictal period, particularly in the early-ictal and late-ictal periods. What we wanted to ask here, then, was the following questions. If outward connectivity is universally dropping, well, what's talking to this SOZ more? That's the first thing. Second of all, how does this relate to the spread of seizures? So next, we examined these same plots, these same network trajectories, in the context of seizure activity, and then restricted our perspective to only the non-involved zone. So you'll know if you've dealt with epilepsy data that most of the sampled regions are the non-involved zone, the areas that don't cause seizures. And so this drop in outward connectivity was likely explained by the fact that the non-involved zone was not talking to itself as much, as in normal brain activity halted, and the non-involved zone was trafficking substantial amounts of connections towards the SOZ, which is what we're seeing here in this plot. In this plot, I've also gone ahead and z-scored all network measures toward the zero, so we removed some of the ISH signal here to get a pure apples-to-apples comparison across all regions. And what we do see is a concomitant increase in the non-involved zone to seizure-onset zone communication, followed by a drop in NIZ-to-NIZ activity, which is, again, the part of your brain which is not perturbed or initiating epilepsy as much. So as we then correlate that with seizure spread, we thought we would use a measure of seizure activity that we found in the literature for beta power. So we ran a beta detector algorithm, a beta power detection algorithm, over the same contacts and channels and correlated these network configurations with changes in beta power. What I find most interesting here is this plateau in beta power at the beginning of the seizure period, and then this sharp increase. So just using some second-order differencing, we get a rough estimate of the point of inflection of this plot, and then we can correlate that here with this network configuration. What I'm seeing is that as the network collapses here, or as this inward connectivity towards the SOZ diminishes, so does the beta power increase. We interpret that to mean that the SOZ is probably receiving some sort of reactionary suppression from the broader network as it initiates seizure activity, but there is likely a push-pull antagonism between these two regions, and ultimately seizures win, thus the spread of seizure activity. The next plots here are a way for us to kind of do a statistical gut check here. So we're doing a repeated measures correlation here to see if these network trajectories do correlate with the changes in beta power corrected for the multiple measures that we have over time. And it seems like, especially in the late-ictal and post-ictal period, you can say that deviations from normal physiologic beta powers do correlate with decreases in the network speaking to the SOZ, thus potentially saying there's less inhibition of the SOZ during the peak of seizure spread in the seizure period. So this is putting it all together here. Beta power increasing the most as inward connectivity to the SOZ decreases, and then this is across the subject level too here. We take this to mean that the suppressive motif which we identified in the ISH is likely not enough to prevent a seizure activity, but is probably a component of the push-and-pull between seizures spread and seizure evolution. We also look at this to mean that it's possible that seizure activity could be mitigated by increasing high inward connectivity to the SOZ, and that it's possible that there are more than just non-involved nodes, that zones that are not responsible for seizure initiation may be differentially responsible for holding off on the spread of seizure activity. And we would like to explore some of the network properties of these non-involved nodes that could potentially be targets for neuromodulation in future studies. I just want to thank everyone in the BN Lab, and particularly Price for his help on this project, and Dr. Engelhoff for his guidance and mentorship. Did you look at other frequency bands besides beta? Yeah, so we look at gamma, too. When you do the PDC plot, you see the whole band of activity. Yeah. There's a lot of broadband shifts. And we also summarize in kind of like a rough and dirty integral of the bands of interest. So we kind of kept them all. Yeah. This is really interesting. Lots of questions. A couple of them, just real quickly. The way you're using the measure here, what if there's no connectivity? I didn't quite get what it was. What would no connectivity look like? Yeah. Well, how would you think? Some contacts aren't connected, right? Yeah, you would have a zero value for that. So how is that incorporated? Because I'm wondering mostly about the post-dictal response, where you sort of have a quiet signal. It's flat-ish often, post-dictally. Yeah. What does that look like with the way you're using the measure here? Right. That could be some of the limitations of partial directed coherence itself. So concurrent to the publication of this preprint, Bartolome and Cayabas put out a paper that talks about the peri-ictal changes in synchronization. And they used a non-linear measure of basically like synchrony. And they found results that show disconnection to prelude in the seizure period. And that would not be properly captured with partial directed coherence. And getting back to this beta response, if I'm following it correctly, and forgive me if I'm not, the non-involved zone is changing in power, though, right? Yeah. Which is fine. I'm all for that. But what's the definition of non-involved in that case? Right. In other words, is it blowing? Is that still non-involved? Yes, yes. So how are you defining those terms? Right. My usage of the word non-involved zone comes from the readout of the EMU report. So only focusing on seizure initiation, I think, is insufficient to describe these networks. So the clinician looked at it and said that there isn't seizure here. Seizure doesn't start here. Doesn't start? Oh, OK. Yes, yes. But it might be there at those points. It will spread there. I think that's the case, yeah. All right, so non-onset zone. It's non-onset zone. Yeah, yeah. Got it, got it. Cool, thank you. Yeah, no problem. Thank you for the questions. Any other questions? Well, if there's no other question, we'll give back five minutes. So we'll conclude the session. Thank you. Thank you very much. This was a great session.
Video Summary
The session focused on intracortical high-density systems for neural recording and stimulation, highlighting their uses and challenges. Dr. Ahmed Raslan and Kelly Collins introduced the session, noting the involvement of experts in the field, including Dr. Sid Keshe and other contributors discussing advanced technologies. Dr. Keshe emphasized new technologies allowing enhanced neural recordings and stimulation, changing how the brain is understood in terms of physiology and neurology. The session covered topics like the high-resolution capabilities of current devices, comparing advancements with past technologies like grids and strips. Dr. Keshe's talk detailed the use cases and challenges associated with these technologies, discussing intracortical arrays and their crucial role in expanding our understanding and application of neural interface technologies.<br /><br />Dr. Shadi Daye from UC San Diego provided insights on crafting multi-thousand channel thin film electrodes, detailing how these innovations support high-density recordings that are necessary for extended brain coverage. Dr. Daniel Kramer from the University of Colorado discussed applications of these technologies in brain-computer interfaces, emphasizing their role in sensorimotor restoration and executive functional studies.<br /><br />The session also included short abstracts from other researchers. Brian Chung presented on beta EEG power modulation within the orbitofrontal cortex, finding differential beta-band power modulation between movement execution and inhibition phases. Dr. Jan Vesper presented findings from a registry comparing awake versus asleep DBS placement in Parkinson’s patients, noting similar motor outcomes regardless of surgical conditions. Ghassan Makul discussed seizure network dynamics, finding network interactions that suggest suppression motifs associated with seizure phases, presenting potential targets for neuromodulation.<br /><br />The session concluded with discussions that emphasized regulatory challenges, ethical concerns, and the need for collaborative efforts to advance and apply these technologies effectively in clinical settings.
Keywords
intracortical systems
neural recording
neural stimulation
high-density systems
brain-computer interfaces
multi-thousand channel electrodes
sensorimotor restoration
beta EEG power modulation
deep brain stimulation
seizure network dynamics
neuromodulation
regulatory challenges
×
Please select your language
1
English