All right. Next up, Dr. Andrew Koh, who's going to talk to us about closed loop DBS system for essential tremor. Thanks. So, yeah, I'm going to talk about one of our implemented systems for this. Thanks for inviting me to talk. I appreciate the chance to do this. I don't really have any disclosures. So, we looked at treating essential tremor with a closed loop system. Don't need to tell anybody in here that much about essential tremor, but just for background, it's the most common movement disorder. It affects about 10 times as many people as Parkinson's, and it's usually characterized by an uncontrollable rhythmic movement during voluntary action, and you usually don't see it too much at rest, and that kind of comes a little bit more important when we're looking at strategies for treating this. To kind of harken back to our previous discussion about ablation versus neuromodulation, every time I talk about any of my patients, George Ogerman always brings up the fact that he's been looking at the same target since the 60s, but the VIM thalamus are just above it or just below it. All of these sites have been shown to work and work about as well as DBS, but I like this picture because it's from before Lipidopa, which I always find entertaining, but that early work with thalamotomy kind of leads to the dogma of DBS, which is when we're doing the stimulation, we're mimicking an ablative procedure, but in a reversible fashion, and in terms of targets, we usually target VIM, VOA, VOP, or sometimes preliminiscal radiations, but this dogma is really not totally true. On the other hand, it works pretty well. So currently, we set up a set of parameters for stimulation, treat people, it works pretty well. It's simple, and mainly the downside is that if you want to make significant adjustments, you have to do this with a clinician. It really doesn't do much to what we think of as restoring basal ganglia function, so it's not really responding to any behavior from the patient, and it's essentially a static therapy in this dynamic environment, and I think increasingly people are thinking that we can do a bit better than that, trying to do that by responding to patient physiology. In terms of essential tremor, I think there's two sort of main problems to overcome. One is what to use for control signal. Unlike with Parkinson's, we don't have a beta band increase in STN, and then the second problem is looking at stimulation parameters and what to do with them. So in looking at this, again, we want a surrogate marker for the benefit of DBS, so this could be a brain signal or a wearable sensor or some sensor that's standing in for the effect of DBS, and then the rest of this diagram here is basically just describing a control loop where you can take this data, generate a reference signal, get a state estimation for what's going on with the patient, look up desirable features that we want to see in the signal, come up with an error signal, send that to the controller, try to get the reference signal and the error signal, I'm sorry, the reference signal and the state signal to be closer to each other. But basically this is a neurophysiology problem or an engineering problem, so one we want to look up or figure out some sort of marker for DBS that's working or a state that requires more DBS, and then figuring out a system to implement the responsive stimulation. We started a closed loop trial a few years back. We're using the Activa PC Plus S system, but this was just basically our aims, is one, to try and reduce power usage by delivering DBS only on demand, and our goal was about 20 percent power reduction because it seemed pretty easily accomplished, and you need about 10 percent to make up for the sensing. In terms of the biosignals we're using, we have wearable sensors, inertial sensors, a smart watch that patients can wear, surface EMG, and we also implanted cortical electrodes to look into the possibility for using a neural signal, either for the intent to move or in more of like a BCI fashion, letting patients try to adjust their DBS voluntarily. In terms of what we're looking at, just how much tremor they have and how much power it takes to get to that degree of tremor control. So this is the system, I think again, I don't, everybody in this room probably knows what this is, but the Activa PC Plus S is a Medtronic IPG that allows sensing. It can stream data to an external computer, and it also has a fairly limited computing power on board so you can do linear discriminant analysis, something simple like that. The rest of the signal includes the Nexus controller, which is basically a communication device, and that does allow you to stream data to a laptop or, we started using a phone because we thought it was cute, but it's really not necessary. And we implanted a Resume 2 electrode over motor cortex for an internal sensor. This is what I was trying to fix earlier. We have approved for five subjects, so far we have four implanted. This is data from three, but you can almost see the depth electrodes here in VIM. And we were targeting basically aiming to straddle central sulcus with two leads over motor and two leads over S1 for this four contact cortical electrode. This is sort of the paradigm that we were using, so the main thing to take away here is that we're streaming out electrocorticography data. At the same time, we're gathering EMG and movement data and working on developing this part of the process here where we're doing signals processing and modeling behavior for the control algorithm. Our initial experimental tasks have been fairly straightforward. One, we wanted sort of a prompted tremor-provoking movement that we could repeat and kind of have a consistent time course. So we chose about 10 seconds of holding posture for these patients. And then we would run them through the FTM tremor rating scale just to get sort of clinical objective findings that people could relate to. And we had those blinded clinical raters doing our video assessments. So when we first started this, we looked first for potential biomarkers to use, maybe like a cortical tremor signal or something like that. But to sort of calibrate the system, we first used these kinematic sensors to trigger stimulation in different ways. So here, this is just our inertial tremor estimate. So again, looking at the smartwatch, we just bandpass filter at about the patient's tremor frequency. And we're having them repeat a posture-holding task here. But without stimulation, you can see every time they move, they have a fair bit of tremor. With their open loop sort of clinical settings, pretty good tremor control, you will see these little spikes here. And that's basically artifact from initiating movement. We can't really filter all of that out, but essentially no tremor. We looked at using this actual tremor estimate to proportionally deliver DBS. And so depending on how much tremor we were seeing with the smartwatch, we start to turn the DBS on, turn it off when we're not seeing tremor. And you can see that the stimulation delivered is going on and off. And we are seeing a fair bit of tremor. The last approach we did was just using the surface EMG to detect movement onset. And so as soon as they started to move, they're on. We turn the DBS on. And you can see the stimulation pattern here is much more regular and kind of periodic, which is when they're doing the task. The takeaway point from all of this was that there's going to be a significant tradeoff if you're trying to do DBS based on their symptomatology. So using the inertial tremor estimate to trigger stimulation, you have about 40% more tremor than if you just left the system on all the time. On the other hand, you save about 80% of the power. So it's an awful lot more, and you're not delivering the full stimulation. Using the EMG-triggered stimulation, you pretty much control their tremor. You get about 90% coverage, and you only use about half of the power. And that's basically because our task was they're moving half the time. So takeaway point, both of these are, there's some tradeoffs between how much symptoms you have and how much DBS you're delivering. Interesting, if you ask their patients, they actually, with the tremor, the proportional tremor stimulation, they actually don't notice that they're having that tremor. They feel like things are just as easy to do, but they're doing about 40% worse. So that's just some of our conclusions from this first part. And again, this kind of motivated us more to look at doing this closed loop, not being as complicated. Instead of trying to find a tremor marker like cortical EMG coherence or anything like that, we just thought, you know what, we'll turn this on when the patient moves, and we'll just kind of mimic the EMG-controlled paradigm. So we did that using the electrocorticography strip. Again, as everybody knows, we're looking at voltage changes. There's frequency information in that that reflects overt and imagined movement. And again, if you look at sort of the low frequency bands, alpha and beta, you get this desynchronization and lowering of spectral power with movement. In the high gamma band, you'll see this broadband increase when you are moving. And you can localize that to different parts of the brain, and basically you can use this to turn the DBS on and off when the patient's moving their arm. So this is the spectrogram that we get from the Activa PC Plus S device. You can extract the beta band power. And basically, if you set a threshold and the beta band drops below that, you turn the stimulation on. If it goes above the threshold, you can turn it back off again. And that seemed like a great idea. And then we found that if you do effective stimulation for tremor control, you actually change the spectra. And I know Phil Starr's group wrote stuff about this a while ago, but what we noticed is this pattern that you see with effective stimulation in the red here, it looks a lot like the desynchronization that you see when you move. And so that does complicate things. You actually have two different states that you have to look at. So DBS on versus DBS off and movement or not movement. And there's different ways to deal with this. One is you can just look at the thresholds and manually tune it. There's usually enough difference that if you set the threshold for each individual patient, you can get good results. It just seemed kind of displeasing aesthetically. So we looked at using some machine learning techniques and it's a pretty, it's not too complicated. We don't need thousands of samples. You need about 10 movements for each state that you're in. So DBS on, DBS off, moving, not moving, not moving, and about 30 seconds. You can train these algorithms to figure out when the patient's doing what. And the thing that was nice is that you can do this with a fairly straightforward linear algorithm. So we can use the Nexus-E, the Activa device to actually do that. So we came up with some patient-specific models. And again, I put this up just to show that the spectra for these patients are pretty different. Even just here, their stimulation's off. They're not moving. Their frequency peaks are pretty different. With stimulation on, again, you see changes in different parts of the frequency band. So it doesn't end up just being a beta desynchronization that we're detecting, but we are catching some changes in alpha and theta as well. In terms of how well it works, this is sort of from one of our early sessions. What you'll see is obviously the patient will start to move. You get an average velocity from the smartwatch here. There'll be a red line that is the classifier detecting whether or not they're moving. The stimulation amplitude will show up here. And I think there's just a spectrogram at the bottom. So as he moves, you'll see the hand shaking a little bit. The DBS will turn on. And it steadies. It correlates very well with the stimulation. And as he puts his arm down, everything goes back off again. And so in terms of the parameters that we were using, we were actually using a fairly wide window here for averaging, because we wanted to make sure it worked at first. But we can shorten that, so we can make it respond quicker than what you were seeing. This is sort of a composite of a couple of other subjects. Again, where we're looking at, I think this is actually a Fontalosa Marin test. So we're having him do finger-to-nose with different hands, different tasks. And the blue line is the stimulation. And this is the classifier detecting whether or not that motor cortex is involved. In terms of how well this works, I think the box to look at here is the stimulation sensitivity, which is a metric we kind of came up with on ourselves. But basically, what we're looking at is, do we miss any movement with stimulation? Because false positives are OK. You can give them a little extra stimulation. We're not wanting to penalize ourselves for that. If you do, the classifiers are down around 60%, 70%. But just in terms of whether or not you're missing any movement to deliver stimulation, we really don't. In terms of how much you're reducing voltage, again, for prompted movement tasks where they're moving about half the time, you save about half the power. To make it a little more visible in terms of what the difference between these two techniques is, these are sort of four to six months post-op with stimulation off Archimedes spirals. And these are patients with open loop versus closed loop. And I think the only one where you can really see a difference is down here, where I think it took a little while for the stem to kick in. So we had a little bit of trouble at the start. But doing this dynamic on-off actually helps just as much as the open loop with the spiral draw. In terms of the rest of their Fontelos and Marin scales, we, again, videoed these exams, had three neurologists blinded to the conditions look at it. And even with that initial tremor that you see when they first start moving, they actually rated the closed loop better for all the patients than the open loop. And it wasn't statistically significant, but just as a... And I thought that was interesting, that even with that initial slight tremor, the start of initiating movements, that these scores were identical. So in conclusion, we can use this beta desynchronization as a control signal. The blinded rating looks like it's just as effective, at least with this small sample size. Something that did come up is that these on-off type voltage changes, people always ask, how much parathesers do people get? And what was interesting is what we saw, that it's really quite variable. So one, you can change the slew rates. So if you have it take one to two seconds to ramp up, for a lot of people that's enough. If you switch to a bipolar configuration, you can ramp a lot faster. But we do have some patients where you're turning it on over seconds rather than hundreds of milliseconds. And these are the folks who did most of the hard work, and I just wanted to thank them. It's really great, Albert. Thank you very much. Any questions from the audience? Yeah. I mean, that's part of the point of doing this, was just to say that, look, if you can get your eight hours of saving, you've extended your battery life by one, two years, right, for a regular PC plus S. Again, you can make these tradeoffs, and I wasn't really emphasizing that too much, but if you wanted to, you can tighten your windows and try to get it more sensitive. We were more aiming for, look, we just want it to be on if somebody's moving, for sure. And so we do accept a little bit of extra stimulation. So that's why it's a 47, 40% saving, as opposed to 50-50, which would be like if it was 100% accurate. Yes. Okay, same question. You're right, for an essential driver, it really doesn't hurt to do stimulation, but for some other disorders, it wouldn't. Can you teach us how much extra stimulation you typically need? And utilizing the smartwatch as well as the grid, how far do you think you would be able to milk this by putting it across the fire to make it highly sensitive? I think, again, you know, using the smartwatch, if we do have a really sensitive tremor detector, you can save, again, I think that proportional stimulation is about, you can save about 80%. The downside is you get less tremor control. But again, subjectively speaking, when we're talking, I mean, part of this project is we have an embedded neuroethicist asking these questions. You know, how do you feel about this? Do you think it's, do you notice the paresthesias? You know, does this make you feel more like a robot? Things like that. But most of our subjects actually don't mind having a little more tremor. And two out of the three that we've had in-depth conversations with like the paresthesias, they're not so severe, but they say, you know what, it actually makes me feel like it's going to start working. So I feel more steady and it's kind of reassuring, which, you know, again, this is a small set of patients, but I think you have a lot of variables that you can play with in terms of trying to save power. But again, I'd be happy if we can really make sure we turn it off at night, you know, that that's a third of the power saved. Rather naive question, but are all these patients got bilateral? No, I usually do unilateral. And for this, if I do bilateral, I would stage. And so all of these are unilateral at this point. Yep. So actually, I didn't talk much about this, but we do have a marker for, you know, you can see a marker for successful stimulation, which you could potentially leverage into a control signal. We do see this drop in alpha theta that we don't see if you stimulate at some place that doesn't relieve the tremor. You also get a peak at about half the stimulation frequency if you're controlling their tremor, which I've not been able to find an intra-op and I'm not sure if it's because of stun effect or what have you, but with these chronic patients, all of them have had it so far. And so, and this is over M1, you know, M1, S1, kind of a bipolar recording. Could you look at premotor or other sites? Absolutely. I think so. It's just, this is what we have. So I won't say that we're, you know, we're not looking in the wrong place, but we do have some other indications that this could be useful. Thank you. Thanks.

closed loop deep brain stimulation

essential tremor

wearable sensors

surface electromyography

Activa PC Plus S system