Hi, my name is Dan Cleary. I'm a neurosurgery resident at the University of California, San Diego. I'm going to talk to you about some of the research that I've been doing on ultra-high-density microgrid recordings and their application to studying human language. To start, I have one disclosure. I'm a startup founder. However, I've taken no payment or disbursements, and the work of that has no relation to the research I'm presenting today. I have two objectives to this talk. The first objective is to highlight development of a novel, non-destructive neural interface device that we've been working on here at UCSD. The second objective is to show the application of this device towards the study of human language processing. So the fundamental question I want to get at is, on what spatial scale should we be interfacing with the human brain? One of the core computational structures of the brain are the cortical columns. So these are small areas of the cortex, about millimeter in diameter, where each of the neurons within that area of the brain generally respond to the same type of stimulation that every other neuron within that column responds. We know that these columns exist in rats, cats, non-human primates, and various other species. However, in humans, we don't have a lot of direct evidence for them because they're functionally defined, not anatomically defined, and so they have to be defined based on response properties. To help us study cortical columns here at UCSD, we've developed a novel high-density neural interface device. This device is a series of electrode contacts on a very, very small scale. In this case, what we're looking at is an array of 64 by 2 electrodes, so two columns of 64 electrodes for a total of 128 contacts. Each of these contacts is about 30 micrometers in diameter, and they're separated by about 50 micrometers. So the entirety of the electrode, with all 128 contacts, covers about 3 millimeters of surface. And the goal here is to look at the very fine structures, computational structures of the brain. For comparison, on the right is one of our novel electrodes, and on the left is a standard clinical electrode. So all 128 of our contacts will fit onto a single contact of a standard clinical electrode. We've come up with an experimental design for studying human language using these electrodes. So the hypothesis is that we can distinguish single columns in the spirotemporal gyrus by using phonemes. As a reminder, phonemes are one of the fundamental structures of language, such as a combination of a single consonant and a single vowel. We recruited patients with dominant hemisphere pathology who were undergoing a weight craniotomy that required mapping. After the clinical mapping, we recorded from the spirotemporal gyrus and presented the patient with three different classes of phonemic stimulation. In the first class, we presented them with real words such as say or boy. In the second class, we presented them with nonsense words such as bah or gah. In the third class, we presented them with amplitude-matched noise. In summary so far, we've recorded from a total of 13 patients, and we've seen task-related signal modulation of the spirotemporal gyrus in seven of those patients. I'm going to highlight the results from one of our cases here. In this case, the patient had pathology marked by the red dot, and the patient had a recording using one of our research electrodes on the blue dot. Following the craniotomy, the patient would undergo standard clinical mapping, followed by recording with our research electrodes. In the figure on the left, you can see that we have a clinical grid placed over the brain. While this grid is in place, the patient would undergo a number of tasks designed to identify areas such as Wernicke and motor cortex. Following identification of those areas, we would then place our clinical, our research electrode, over areas identified as being involved in speech that are also part of spirotemporal gyrus. Here, I'm going to show some of the responses that we saw to the auditory stimuli that we presented. So in this case, we're looking at a gamma band activity. So above about 70 hertz to about 190 hertz that has been filtered for amplitude only. So in this case, at time zero on the x-axis is the start of the presentation of the auditory stimuli. What we're looking at here in yellow are the responses to noise stimuli. In comparison, when we look at the responses to nonsense words, so these are sound like real words but aren't actually real words, we see that that elicits a very strong response from the patient in spirotemporal gyrus. And for last comparison, if we look at the responses to words, we see that that also elicits a response from the patient. And again, these are responses recorded from superior temporal gyrus. So this is a single electrode, a single channel. And so what we want to look at as well are the spatial properties of the responses. So here in the figure on the left, we're looking at the spatial responses across all the channels in a single across all the channels. On the y-axis are the channels according to distance. And on the x-axis are the time of the stimuli, time relative to the stimuli presentation. So what we're looking at in this first figure are the responses to noise. So you can see in the top half, the top spatial half of this electrode, there's only a very weak response to noise stimuli, where in the bottom half of this electrode, the bottom spatial half, you can see that there is some slightly stronger response to noise stimuli. For comparison, again, if we look at the responses to nonsense stimuli, we can see that there is a much stronger response in the top spatial half of the electrode, whereas in the bottom spatial half of the electrode, there's still just a weak response. And last for comparison, if we look at the response to words, we can see that there is a moderate response to real words in the top half of the electrode. But in the bottom spatial half of the electrode, there's only a very nonspecific weaker response. So what we're seeing here is that in the bottom half of the electrode, this area of the brain has a somewhat nonspecific response to all types of auditory stimuli, where in the top half of the electrode, it's more selective for real language related stimuli and doesn't respond very well to noise. If we look at the boundaries, trying to look at the boundaries, we can look at the spatial distribution of the amplitude of the responses by the stimulus class. So in the top figure, top part of this figure, we see the responses to noise. And so there's a stronger response in one half of the electrode, followed by a drop off and a weak response. And in contrast, if we look at words and nonsense, we see that there's a weaker response followed by a stronger response. So what we're looking at are boundaries of response properties over space. So in summary, I've highlighted development of novel surface microelectrodes. I've shown that they're ideal for nondestructive high density cortical recordings. In application of these electrodes, I've shown that we can find this fine boundaries of speech discrimination in superior temporal gyrus. In this case, the boundaries indicate that the individual areas are somewhere around one millimeter in diameter. However, this is still an ongoing work. There's still a lot of analysis to do, especially looking at the responses to individual phonemes, looking at further in-depth spectral responses, looking at boundary analyses. And then one of the things we're actively working on are multi-thousand recordings up from 128. This work is part of a much larger collaboration involving multiple institutions and many contributors. I need to specifically thank Dr. Eric Holgren here at UCSD, who oversees this research, and Dr. Shadi Daya, who oversees the production of electrodes. Thank you.

neurosurgery resident

ultra-high-density microgrid recordings

human language

cortical columns

task-related signal modulation