NIMH Multimodal Brain Stimulation Speaker Series: Angel Peterchev, PhD
Today I will be talking about building TMS tools for selected neural engagement from the ground up. And I have parted this title into three lines for a reason. First, building; building, this means that the stuff that we're doing right now. So, a lot of the work that I'm going to show is work in progress. And I really intended this to be a -- kind of to share the excitement of some of the new technologies that we're developing and what they can offer us.
"Selecting neural engagements," selected target engagements that are a very important part of kind of the multi [indiscernible] studies and funding these days, how can we improve our tools to get more selected effect on the brain. And finally, "from the ground up" that [indiscernible] that we're looking and we're doing work at the very, I would say, low level; low level in the sense that we're really redesigning the devices which we use for TMS, the pulse generators, the [indiscernible] modeling at the level of individual neurons in the brain. So, we think of this as really building our understanding in technology for TMS from the ground up.
I want to give my disclosures. We work on technology, so we have pretty extensive collaborations with industry, including [indiscernible] companies that work on magnetic stimulation technology. We have received some research funding from them and we have worked with them to commercialize some technology. And also we have received extensive support from the NIH, which we're very grateful, and that has enabled the majority of the work that you are going to see. And we have received funding from other foundations and organizations as well.
And, importantly, I want to acknowledge the people who have done most of the work that you're going to see. And actually one of them is my former [indiscernible] Zhiyong Zeng, now at NIH, and great to see him. And there are many other lab members, but I just highlighted three of them who have been really the central effort behind what I am going to show you. Luis Gomez, Stefan Goetz, and Aman Aberra, and I also want to acknowledge my faculty collaborators, which include Dr. Lisanby and Bruce Luber, who are both at NIH now.
So, I won't spend much time introducing TMS. I think you are familiar with that, and I think most of the audience online is familiar as well. TMS is a technique that gives strong magnetic pulses generated by this TMS device that are sent to the brain, and there they induce an electric field which results in changes in neural activity. So, TMS is really a way of inducing electrical stimulation in the brain using electromagnetic conduction. So, it's good to keep in mind that TMS, fundamentally electrical stimulation, is just delivered in a way that's tolerable to the wake subject, the patient.
You're probably familiar also with the applications of TMS. Here I have a brief summary. There are three main categories, research applications from the standing brain functions; therapeutic applications in psychiatry and neurology, that are the list is expanding. We know that TMS has been approved for depression and also for pre-surgical cortical mapping. There's also proof for migraine, although one may argue that that's more like cranial nerve stimulation possibly. But the [indiscernible] many, many other disorders that are being explored, experimented in [inaudible] as well. And finally, [inaudible].
So, these applications, we are -- the applications can be as good as really our technology and understanding of mechanisms. So, what are the limitations and what can be done about it? Well, here I pulled an image from the web, from the Mayo Clinic, which is obviously [indiscernible] institution, but when you consider how they emphasize TMS, you can see some problems with this picture. The TMS coil [inaudible] and looks like a real TMS coil, but then it looks like it's missing pretty narrow beam of magnetic field into the brain. This target, the target is quite big and it's kind of a focal activation there.
So, fortunately, [indiscernible] TMS is far from what actually happens. And to optimize for the TMS, you have to have appreciation of real TMS rather than this idealized picture. So, [indiscernible] adding and correcting this figure to emphasize what actually happens. So, first of all, the strongest stimulation and activation [indiscernible] range [indiscernible]. You can characterize that area of activation with some [indiscernible] which gives you the [indiscernible] stimulation, but, of course, then you can feel that it's not the beam.
You can feel the rounded [indiscernible] is weaker, but it still, you know, covers some various neural populations that you may be affecting. Then we need the target and [indiscernible] surrounding tissues [indiscernible] they are neural populations that you may be affecting, right, and won't target just one of those. Then there is trans-synaptic activations, you know, if the brain is interconnected. So, when you stimulate the neurons, you stimulate fibers and tracks in other brain areas, and so on. And then, beyond the brain, actually in the scalp, you have nerves and muscles in the scalp, you stimulate those. The field in the scalp is stronger than in the brain.
Then the coil, because it's very strong energy in the coil, it also vibrates. So, you can actually feel the vibration [indiscernible] stimulation [indiscernible] the brain. And because of these vibrations, the coil used in TMS makes a pretty loud sound. So, you have vibration that's transmitted both through air and through the bone of the skull, and you hear that sound, bone conduction and air conduction sounds.
So, all of these kind of additional activation sites [inaudible] we have all these unintended effects that occur during the administration of a pulse that are beyond our ideal desire to maybe stimulate one of these neural populations [indiscernible]. So, they kind of look somewhat disappointing, and it is the result of various fundamental physical constraints of TMS, but not all of it is fundamental. And we have been trying to really rethink design of TMS of how we can really try to build TMS closer to an ideal TMS. And, in part, it's understanding how these processes work, and also eliminating some of these extraneous [indiscernible].
So, I have a list here for how I think a functional [indiscernible] TMS, which is what we want to accomplish, and factors that contribute. [Indiscernible] contributes [indiscernible] mostly by the coil. Then the temporal characteristics of TMS can also affect both how you recruit neural populations. Of course, the repetition parameters of pulse also can interact with the dynamics of larger populations and they have significance in what you affect and how. Then the neural connectivity can stimulate the superficial brain areas and then stimulate deeper areas [indiscernible] promising way to target deeper brain structures. It matters what the brain is doing while it stimulates, what the context is, and I think you should take more of that into account.
And finally there is this ancillary effects, auditory stimulation, somatosensory stimulation that are not fundamentally a part of the process. So, if we can kind of minimize them, right, we can have more selective stimulation process. So, this presentation, I'm going to give you really a kind of brief overview of some of these technologies that we're developing to address these, to improve TMS along these lines, that would be along the lines of spatial targeting related to coil design, and both shape targeting, [indiscernible] TMS waveform, and also effects of actually the sound of TMS.
So, first, talking with coil design. In this figure, this is result of this work [indiscernible] where we put together a list of most of the coils that can be published or [indiscernible] or proposed in TMS, and want to evaluate how focal and deep can TMS be at the same time; right? We really want something that's focal and also can go deep [indiscernible].
So, here, we stimulated in this case 61 coils. And these resulted in [indiscernible] quite frequently because it actually shows a very kind of definitive tradeoff between the spread of the field, that is the kind of [indiscernible] how much of the brain it stimulates, how much of the brain surface it stimulates, and then versus [indiscernible] stimulation, how deep the field penetrates.
You see there all these coils are pretty much lined along two curves. The lower one, which is the better one, right, we're given [indiscernible], this one corresponds with the conventional figure-eight-type coil [indiscernible]. And then you have -- when you have larger coils, you go up this curve, when you have smaller coils, you go down this curve, but you're always in this curve.
And all kinds of tricks that people have proposed didn't move them, they didn't budge from this curve. I mean, they may get worse than this curve, but not below this curve. So, I personally thought that that's it, that we should reach the limit here, but then we dubbed this new [indiscernible] in the lab with [indiscernible] whose PhD is in computational electromagnetics. And he's very bright and the kind of tools that he has in computational electromagnetics allow us to take a very different approach to coil design, which he has started working on in his PhD at University of Michigan. He's now continuing [indiscernible]. And it is not to play with -- come up with configurations and try them out, which is what people have done before, but actually computationally synthesize this coil so that it optimally meets our requirements, our specifications, which makes a lot more sense; right? If you're asking basically fundamentally what is the best thing to do for targeting [indiscernible] then you think how it may be, you know, coming up with your -- drawing your design.
Basic computers are better at electromagnetics than we are. They can, you know, take all of the relationships into account and our brains can't really do that. So, the way this approach works is you specify some requirements where this target is located, what should be the field direction there, what is the stimulation threshold, you can estimate that, and that's not, you know, something that's scalable, so it's not an issue. How strong can [indiscernible] field relative to that field, that's a constraint you put in, and how much energy we can practically use for the coil.
So, you can put all this constraint, then you choose what you want the flat coil or a curved coil or a cap coil. So, you choose kind of the overall shape of the coil, like [indiscernible], and you also choose what model you want to do the optimization. You can use a spherical model, which is what we used for -- with [indiscernible] for these evaluations, or you can use a realistic head model derived from MRIs. The first mention of MRI, so that's an interesting use of multi-modal techniques where here we actually use structure MRI to design coils for TMS.
So, whilst we make these choices, then the computer algorithm basically breaks down the surfaces into this spatial [indiscernible] modes, and then weights them in different combinations [indiscernible] combinations to meet in an optimal way these specifications. For example, optimum may be getting all these specs with a minimum stimulated volume of the brain, so with maximum focality. So [indiscernible] this group, and you run that on the super computer. It may take you a few hours or a day in a worst case, but, you know, that's no problem. Now we have super computers, everybody has one at this point at their campus.
And then [indiscernible] coil design. And I have here [indiscernible], you know, fresh from, like, as of last week what's going on. So, we are -- this may be important technology, so we're about to submit [indiscernible]. I'm not disclosing the actual coils we're getting, but hopefully we'll be publishing that soon. But I can show you on these curves what we are getting in terms of performance, and we think that this is the fundamental limit actually without constraint to the coil energy.
So, it looks like we can be [indiscernible] you get approximately [indiscernible] for the same time. So, the point is that we're given that we may be able to increase focality of measures by spread of the field by a factor of two, which I think is pretty good. I mean, I think it is meaningful [indiscernible] is -- you know, can be significant. And actually for different coils [indiscernible] almost like a [indiscernible] here.
So -- [inaudible]. [Indiscernible] fundamental limits or practical coils, I think we can get close to that, that's also on the practical end of the constraint. So, that's a little kind of peek into something exciting that we're working on and that may result in a different -- you know, a different category of coils [indiscernible] characteristics. And also recently the lab was [indiscernible] be speaking here at the NIH soon in this seminar. They have done some great work also in that direction of computational optimization of coils. So, really, this field is going in that direction right now, and apparently we can do some things, you know, better than most [indiscernible].
So, the next topic -- obviously, you know, the first topics I'm going to cover relatively quickly, and the third topic will be a bit more extensive. My next topic is on our efforts to build devices that are quieter. So, why is that an issue? If you have experienced TMS as an operator or as a subject, you know that it is loud. I mean, you have to wear ear plugs. There are, you know, safety guidelines [indiscernible]. We operate [indiscernible] you can hear often the device down the corridor, even with a closed door or in neighboring rooms.
If you quantify the noise, it can be over 100 decibels. At maximum [indiscernible] 120 decibels. So, that's loud. There are various comparisons, like it's a leaf blower or jet engine [indiscernible] many kind of comparisons, especially very loud sounds. It doesn't sound as loud because it's very brief pulse. It's never a very strong -- very strong sound.
So, it's not just a nuisance. It's actually a major cost [indiscernible] to TMS because what you do is you activate the auditory system synchronously with each pulse that is delivered to the cortex [indiscernible]. And if you look at what happens in the brain, and if you [indiscernible] multimodal, TMS [indiscernible] in this case, you can see on this picture, you stimulate the brain through the auditory perception.
On this figure here, you see -- again, TMS [indiscernible] intensity of pulse is delivered to frontal target, ranges from 100 percent of model threshold to -- sorry, from 80 percent to 100 percent to 120 percent of model threshold. And you can see that under the coil, below model threshold, you don't get much activation, it's 100 percent. You see some activation, and that is 100 percent [indiscernible] more activation, which is what you would expect from ideal TMS.
But look at what's going on auditory, cortex, brain stem, they're all lighting up, even at lower intensities, and that's because we have a loud [indiscernible] sound, and actually it's worse in the MRI scanner because we have the [indiscernible] forces, also the board reverberates, so it's like this, you know, reverberation chamber. So, it's really loud.
And when you look at this you can't really say we're focal. I mean, you can replicate these effects with [indiscernible], right, and you say, "Well, I'll have same effect," but it doesn't mean that you can just subtract them out. The brain is obviously [indiscernible] that it can't just be, you know, replicated something to subtracting that and saying, "Okay, that didn't have any effect, I'll just subtract it out." So, sound is a real issue.
And it can -- even auditory stimulation [indiscernible] can modulate the brain [indiscernible] stimulate somebody [indiscernible] there will be actually your module is already [indiscernible] lasting. It complicates lighting [ph] in studies, some obviously [indiscernible] some patients, maybe [indiscernible]. So, it's really something that we'd rather not have. And when you think about this, it's an acoustic field. It has nothing fundamentally, physically to do with the magnetic field. They don't have to come hand in hand; right? So, you should be able, theoretically, to design the TMS device that makes no sound simply by not allowing it to produce any mechanical energy, only having magnetic energy out of it. And that's what we are trying to do.
So, we call the approach quiet TMS, or qTMS. We have a lot of TMS with modified sounds. So, it's kind of -- but we also have acute [ph] TMS. So, it's a two-pronged approach. The first part of the approach, which is probably the less confusing part is that we are -- you think, oh, you just make a quieter coil. It's not just the coil, because the problem with conventional TMS is the energy of the pulse that is sent to the coil, the semi [ph] sinusoidal pulse, and it is standard in the -- around in the range of two to five kilohertz, which is smack into the back hearing range for humans; right?
[indiscernible] hearing frequencies up to, for young people, 20 kilohertz, for older people it kind of rolls down a bit, but that's smack into where we see its effects; right? So, we have to start -- well, so, one point is let's get out of this range. So, what we've proposed to do is move this pulse frequency from three kilohertz by a factor of ten to something like 30 kilohertz. So, then you're outside of this human hearing threshold. So, it's a good start, right, at least conceptually, to keep the energy in that frequency where you shouldn't hear much of it.
Now, why is that not trivial? It's not trivial because this means making the pulse -- the TMS pulse ten times briefer. We have ten times higher frequency. So, making the pulse ten times briefer because we have strength-duration curves in neurostimulation, it means that basically you have to increase the [indiscernible] of the pulse by a factor of ten, approximately. And that is technologically challenging. And actually, with conventional devices, making such a brief pulse is also challenging, but we have already technology that can do the briefness, but the amplitude is a challenge.
So, we're going to address that by building a device that produces these briefer pulses and uses higher volts [ph] and higher [indiscernible] just to get that factor of ten. You can split [indiscernible] proper coil design. Interestingly, the energy that's needed is about the same [indiscernible] 100 joules is an approximate threshold, that doesn't change. In fact, it's theoretically slightly lower for briefer pulses, even though the pulse is [indiscernible] because of the [indiscernible] lower energy.
And the second part of this two-pronged approach is to build a better coil. And the purpose of the coil reasonably is to make sure that the magnetic energy that is delivered to the coil is not converted to mechanical energy, or rather a minimum amount is converted, not only that, but the conversion is linear, that is the -- you don't have down-mixing of the frequencies, from high frequencies to low frequencies, because of mechanical vibrations out of the coil. So, you really -- most energy you end up emitting is still at this, you know, 30 kilohertz of and three kilohertz.
So, back -- the bunch of mechanical engineering, and it turns out that the best way to do it is have multiple layers within the coil. You have very steep, winding, and then you have these different layers, [indiscernible] and then another, like, stiff casing. This actually comes from literature on machine design for quiet machines, and with appropriate adaptation.
So, when we combine these two approaches, we [indiscernible] kind of combined significant improvement in the sound [indiscernible] by TMS. And that's an actively ongoing project that's funded by a brain initiative [indiscernible]. So, whenever [indiscernible] started, we already had demonstrated a scale down [indiscernible] use conventional voltages [indiscernible] one kilovolt and kind of scaled the [indiscernible] so we can see the benefits, even though it can't get to threshold, basically sub-threshold here. And with the new device, we have managed to go to threshold levels.
So, at the sub-threshold levels, we scaled everything. We scaled conventional [indiscernible] sub-threshold level, and now we're [indiscernible] ultimate sub-threshold level. So, I have this comparison matrix where in the -- this is the acoustic waveform that you record in the microphone coming out of the coil. So, here, on the upper left block you see the waveform coming -- the acoustic waveform coming from conventional [indiscernible] with a conventional maximum duration pulse of 300 microseconds.
Then, to the right, you see the conventional [indiscernible] a brief pulse, 45 microseconds, with the existing devices that we have specialized devices even we can't get 30 microseconds with them yet, but it's still much briefer; right? And then you see that the [indiscernible] waveform is reduced significantly.
Then at the bottom -- or in the middle row on the left, you see the quiet TMS coil that we developed with a conventional pulse, and its very weak signal. And then, finally, we combined the brief pulse with the coil. We have an even briefer and shorter pulse. And the plot here on the bottom right summarizes that. The loudness is given in decibels, and that's relative to the, essentially, auditory threshold. It's a way to scale that and compute that from the waveform. So, it's not from humans, that's from computing, that's from the acoustic waveform. So, you see the conventional devices and we see our [indiscernible] with all the features together into the blue bar.
So, you have 90 decibels, which is almost a factor of ten reduction in the loudness. So, that's already quite remarkable. And we're going for [indiscernible] to go for 40 decibels, at least [indiscernible] which is a factor of 100. Now I have here waveforms that I will play with. We figured out that you can only play them actually during live preview models. And I don't know how well they play over WebEx because the decibels [indiscernible] in the microphone. So, I apologize to the viewers on the WebEx, but I'm going to give a little demo.
So, this is conventional TMS sound. You may recognize [indiscernible] TMS. [Indiscernible] sounds like. Okay. So, now let's hear quite TMS. And that's for match -- it should be matched in terms of [indiscernible] threshold, the previous pulse. Does anybody hear it? You hear the faint clicking? Yeah, the young people -- I guess you're all young. When I have given presentations to [indiscernible] group meeting of older CEOs, they're like, "Where's the sound?" I'm like, "You can't hear the pen clicking?" They're like, "No." I'm like, "That's perfect," you know, you can be our subjects for the study. We're going to -- so, after we build the device, which we're now developing, you know, it's a pretty challenging engineering project, but we think it's doable, we'll test it in humans. We select [indiscernible] subjects. Maybe that was a mistake, so, but it should be ten times quieter than this, so, you know.
[Inaudible] surface stimulation or the sensory?
That's an interesting question. So, actually, yeah, so the question is does it interfere with the sensory stimulation in the [indiscernible]? And we have done some studies with briefer versus longer pulses, but we haven't done such brief pulses. And it is possible that sensation may go down based on theories to that effect. And even if that doesn't happen, I don't think it will go up basically, because the [indiscernible] of cranial nerves and nerve endings and [indiscernible] they tend to be generally a bit longer, or the same as cortical neurons. So, I don't expect it to get worse. It would be very interesting. I'm very curious to experience what that would feel like. And the only way we'll figure that out -- that question came up from the reviewers of the grant. And we said, well, this is what we can hypothesize, but ultimately we can't test that because we don't have a design that will know that by the end of this project.
And now I'm moving on to the last component of the presentation, which is the most extensive. The previous two were basically previews of this technology that we're developing to kind of whet your appetite. And I'm going to show you a few results that are a bit more kind of closer to publication. And that all has to do with -- or already published, but that all has to do with how we can leverage and understand [indiscernible] pulse waveforms in TMS and neurostimulation.
So, first of all, [indiscernible] background in TMS [indiscernible] that conventional TMS devices typically offer you one of two waveforms, or some of them offer you both. The typical waveforms are monophasic waveforms, where the pulse is this kind of monophasic cosine [ph] pulse, and biphasic, which is, again, a cosine pulse but has a full period of oscillation. And the biphasic pulses are the basis of essentially all repetitive TMS devices, including those used for therapeutic application.
Some devices [indiscernible], just the [indiscernible] devices, and I think maybe now some other ones, offer half-sine pulses, which is [indiscernible] pulses in the negative space. So, even with its limited amount of pulses available, there have been quite a few studies that have shown that the pulse waveform matters. And one very interesting result of this study, as shown here, is from Mark Sommer and colleagues, published more than ten years ago, where they compared these monophasic pulses, half-sine pulses and biphasic pulses. And they also changed -- this is [indiscernible] -- and they also changed the direction between flowing posterior-anterior, PA, so [indiscernible] flows from the back to the front, or the anterior-posterior, which is kind of going in the opposite direction.
And you can see here that the latency of the [indiscernible] potential, which is the finger twitch that you get when you stimulate [indiscernible] cortex, is approximately the same for all of them, around 24 milliseconds, except for the monophasic pulse when pointed from anterior to posterior. So, for that direction, the latency is significantly higher, by almost two milliseconds, and that's a big deal. Two milliseconds, that's about a couple of synaptic lengths, because going from one synapse takes about one millisecond.
So, it's very reasonable to interpret this as this pulse with -- the monophasic pulse with anterior-posterior orientation of the fields, if stimulating a different neural population than the other pulses, and the signal then has to travel through a couple more synapses to then get to the output neurons from [indiscernible] cortex and get to your finger. So, it's a sign of -- even though the field distribution is the same, just the pulse shape combined with the direction of the field provides [indiscernible] selectivity of what neural populations we're stimulating. And, in that sense, it provides you with some additional selectivity.
So, that's an interesting and potentially important additional degree of freedom that -- or degrees of freedom that we have really in controlling [indiscernible] in TMS. And -- yes, another question?
Just to probe on that, does it really provide us more selectivity? I mean, the latency was a ten-percent difference, but it was still [indiscernible]; right?
So, I wouldn't think about -- so, the question is does that really provide [indiscernible] activities, just the 20 or, you know, ten percent difference in the latency. This is -- you know, we should not think about it [indiscernible] I think the proper way to think is synaptic [indiscernible] because most of the reasons that [indiscernible] 24 milliseconds because it takes time to travel from your brain to your finger. So, that's -- you know, we're talking about stimulation cortex. And two milliseconds within the same cortical area, that's a lot. You're really stimulating -- you have to travel through two synapses. So, you're simulating something different, a different population in the cortex. So, it is significant. And actually that significance is really highlighted by studies by, again, by Mark Sommer and by some other groups that -- I have four kind of [indiscernible] here as examples of exciting results that show that that actually makes a big difference when you do your modulation with repetitive [indiscernible].
So, for example, in this study, the one on top, they stimulated [indiscernible] with monophasic pulses in either posterior-anterior or anterior-posterior, or biphasic pulses with one or the other direction. And when you see what happens to the [indiscernible], that's really fascinating. Five [indiscernible] excitatory frequency. So, the biphasic pulses that are conventionally used for [indiscernible] produced, you know, reasonable excitatory effects. [Indiscernible] approximately for either direction, and that makes sense because they're kind of bidirectional pulses. And that's what we normally get in studies [indiscernible], but now look at what happens with the monophasic pulses.
The posterior-anterior pulse gives you this massive excitatory factor, like almost five times the A to P [indiscernible] I mean, increased by a factor of five. When changing the direction with the monophasic, convert that to miles inhibition. So, the biphasic then appears to be almost, you know, the summation of the two, you have this intermediate effect. [Indiscernible] this case, by using monophasic pulses with appropriate orientation, you can produce stronger and more selective, in terms of what you're trying to get, [indiscernible] modulation effect.
And there I won't go into detail into these other studies, but basically pretty much any study that has looked at monophasic versus biphasic both seem to show that monophasic pulse, especially, like, a given direction of the monophasic pulse, produces stronger and/or longer-lasting neural modulatory effect. So, that, again, indirectly indicates that we're modulating the cortical circuits in a different way when we change the pulse waveform.
So, the first question is what's going on? Where do these differences come from? And the field of TMS, really we haven't had many good tools really interrogating cortex at the level of individual neurons and how they get activated by TMS. And I think now with the proliferation of exciting tools and imaging tools in neuroscience, I think there are lots of great opportunities to explore that. And one of the opportunities that's now afforded by computational methods that are available is actually the simulations, and that doesn't mean that [indiscernible] experimentation that I'm going to talk about this later, but let's see what is in the computation.
First of all, again, starting from a structural MRI, yet another use of multimodal techniques in TMS, if you take a structure MRI of the subject, you can then construct the 3D model of their head, including the different tissues that are relevant, like the brain, grey matter, white matter, [indiscernible], scalp. And now there are packages available online that you can actually use. So, it's very user-friendly at this point technique that should be more widely adopted, and I think that's starting to happen now because it's really [indiscernible] now.
So, with this techniques, we can create a 3D model of the head and we can simulate based on these electric [indiscernible]. And that has been done in the literature for a while now [indiscernible] but that doesn't tell you, for example, there will be no difference between the monophasic pulses going -- in fact, there will be no difference between biphasic or monophasic going in the direction [indiscernible]. All these different conditions that I've shown you will give exactly the same electric [indiscernible] map because the coil is the same place. So, this doesn't tell you much.
What really is different is how these field characteristics interact with the complexity of neural populations in the brain. So, we have to start modeling that to really get an insight into what is the mechanistic basis of the different effect of different pulse shapes. And that's what we set about to do.
So, in addition to making these realistic models, which is, again, now relatively mainstream, we set about to differentiate different layers in the cortex. So, you can see a cross-section here of our project. You have the layers from, you know, layer one, layer two, three, four, five, and six. And then we populate these layers with neural models that are pretty realistic. And you can see one rotating down here. And you can see what the brain looks like, at least for one layer, when it's populated. Obviously that's not the real density of neurons in the brain. It's about 140 times less than the actual density of neurons, but, you know, it can increase that density if you wish, if you have the computational resources, and already with intensity [indiscernible] gives a lot of detail into what happens in the cortex. So, if you think [indiscernible] neurons around it relatively well. And, again, you can, depending on your computational resources, you can [inaudible].
The neurons are based on recent -- relatively recent data studies [indiscernible] published from the Blue Brain Project in Europe, that's headed by Henry Markram. So, Markram's group did really elaborate work in [indiscernible] characterizing [indiscernible] cells, and then also documenting their morphology, and creating neuron models [indiscernible] stimulation package for neural stimulation. So, we took these neural models and we adapted them to humans, because these are from relatively juvenile [indiscernible]. And a lot of these [indiscernible] like, we're all mammals, but there are some differences.
For example, we have to add [indiscernible] and we have to scale the cells and, you know, there are other adjustments that are all done based on well-understood, basic either biophysics or neurobiology. You know, it's not like we're tweaking them to give us what we want for TMS. We're just building models based on what we know about how neurons are.
And then, with these populations, you can -- you know, you have [indiscernible] simulation. And then [indiscernible] that stimulate each of these neurons with local [indiscernible]. And then you can map activation with neural populations for each layer. And even within [indiscernible] you have different neural populations. So, here, you can see what we get out of these simulations, we can get these cortical maps where now for each layer -- we don't see the layers, just one layer, we'll see the layers later -- you can get the map of the threshold for activation; right? So, normally, in TMS, you do, like, the model threshold [indiscernible] you get one threshold that's your model threshold, but here we're getting -- for each neuron, we're getting a threshold. So, now it's a threshold map that we can do all kinds of things with.
This is the library of the neurons that we model within each layer. So, obviously, within each layer, you have more than one layer of neurons. We selected, for this particular simulation, for simplicity, we selected neurons with the lowest thresholds within layer. And you can add -- and we have other models that can be added. And these end up being the most important ones that you'll see are layers two, three, and four, and five, where in two, three, and five, you have pyramidal cells, and in four you have interneurons, which at the lower threshold are large basket cells.
So, let's see the first thing that we got out of this kind of quantitative results. Well, you can ask how do the thresholds for activating different layers compare against, you know, the other layers? And these results you see for the different layers, going from one all the way to six, and you see this is a bunch of different -- all these different, like, monophasic, half-sine, and biphasic in different directions, as we saw previously in the experimental study. We threw all of these in for fun. You can focus on, you know, one particular color. But you see that the trend is pretty clear.
And the trend is that layer five gets lowest threshold, followed not too far behind by layers four and three, too. So, layer one is fine, you know, we have these cells that are way up there. They probably are very likely to be activated. Layer six, especially given that it's [indiscernible] weaker, it's also probably likely to be activated. So, this seems to be the layer that we should be focusing on probably.
And I want to point out that those of you that actually appreciate the pulse amplitude in terms of amperes per microsecond, which is one which we should quantify pulse amplitude, for example, [indiscernible] give you that value measure of amplitude. These values are actually, like, the lowest value here, they're on the order of 60 percent of the conventional design output.
So, unlike some other studies with more simplified neural models in the literature, we have managed to get something that seems quite realistic in terms of thresholds, in terms of [indiscernible] thresholds, which is good; right? And we have to do [indiscernible] and a bunch of these kind adjustments that are appropriate, and that [indiscernible] threshold. So, we think that we have pretty good models. And now we can play with them.
So, we like -- one way in which we like to play with them is to see how they -- how they look compared to experimental studies; right? If you get no relationship, that makes sense in the experimental studies [indiscernible]. So, looking at some basic things like from the literature, from the studies that we've shown previously, and a couple from Sommer and [indiscernible], they measured the ratio of the model threshold for monophasic versus biphasic pulses. It's not that monophasic pulses have higher thresholds, and monophasic versus half-sine thresholds, and the threshold is approximately the same. So, this is experimental data here.
And this part is actually standard error, so it's actually very narrow [indiscernible] it's really the proper thing here would be to look at standard deviation because you want to seek out [indiscernible] across subjects, which would wider part. So, anyway, in red is a very simple model of the neural membrane that assumes the linear sine constant. It's a simple integrated [indiscernible] model that people have used, and we have used in the literature, to approximately model the effect. And you can see that that model is also in the right direction, but it really does not give you the full extent of the difference. It's actually quite bad.
Now, with our model, it gets much better. So, it's already within this [indiscernible] standard error bar. And this bar does vary depending on exactly how you define your region [indiscernible] cortex. This also [indiscernible] starting to get better estimates that match with what we see in the literature. And now I think it gives you a very -- some very interesting insight and importance of insight into that phenomenon that increased latency of the [indiscernible] potential that can be known for these empirical [indiscernible] monophasic pulses.
So, we can model that. What you see here is the cross-section of the hand knob, right, the hand knob in motor cortex. People like Mark [indiscernible] dedicated probably the last 30 years, you know, studying this area of [indiscernible] motor cortex. So, the hand knob is the area that [indiscernible]. And if you take a cross-section, you see this is the anterior direction, this is the posterior direction. So, this is the -- the hand knob is in the precentral gyrus. So, here, we have simulated the distinct difference or three different ways [indiscernible] with two directions for each. And, again, we can simulate the threshold for each layer in each location of neurons within the layer. You can see the 2D slide that come from that, you know, that same that's also the hand knob.
So, this pulse gave you the strange, long latency. So, what's happening there? So, notice what's happening in the gyral wall versus the gyral crown here; right? This is the gyral crown. This is the wall. And why this particular area [indiscernible]? These are the areas that control your finger [indiscernible] muscle that's recorded here. So, you can see that when you reverse the direction, you can go from posterior-anterior to anterior-posterior, the activation in the crown remains, by and large, of the same -- you know, approximately the same. But we have dramatic change in the activation in the wall of the gyrus. So, that's a very notable difference here when you change the direction. The field distribution is exactly the same. You're just changing the direction of the field. And that does not happen with the other pulses, which are more bidirectional; right? We see that [indiscernible] change, but still, you don't have that kind of what you consider this relative to [indiscernible] you don't have this dramatic effect anywhere else.
So, that's interesting. So, does that make sense in terms of neuroanatomy of the hand knob? And it turns out that it actually makes a lot of sense, and you can see that by looking at this beautiful figure from Bradalov [ph] and Fig [ph] in a paper from 2009. This is like a flattened image of the hand knob in monkey. And this area here, this is the wall of the gyrus. And this area here is the kind of lip of the -- or crown -- the lip of the crown and the crown of the gyrus. So, these dots here tell you where the neurons that directly control -- the output neurons that control the [indiscernible] finger area.
And then here, with micro-stimulation, they mapped -- stimulating wall neurons gives you the twitch in the [indiscernible] fingers; right? So, you see that the output neurons are here, but the other neurons in the crown, in the lip of the crown, they're still can "synaptically" activate the finger. Well, right, there we are. So, it seems that from what we have seen with the monophasic pulse in one case, when we stimulate the wall, we have more direct output, but when you stimulate the crown here, and the field is stronger here, we have indirect stimulation, which is transsynaptic. So, this beautiful picture that very well matches what this understanding of the [indiscernible]. So, this is an example of how these simulations can really, combined with other pieces of information, can help us with [indiscernible].
So, our time is quite limited. So, I'm going to briefly talk about a technology that we have developed called controllable pulse parameter TMS, cTMS, that allows us to shape the pulse waveform more dramatically than you can with conventional devices. So, it generates more rectangular pulses. You can control the pulse with the pulse direction [indiscernible] more expansive and continuous way than in conventional devices. And it is that it can allow you to explore [indiscernible] waveform more expansively.
And this is the family of systematic devices that we have developed. And now there is a commercial device called by [indiscernible] Research. And actually I just saw the device -- the TMS device that [indiscernible] at NIH. And so we have one if you want to play with one here. Other labs are [indiscernible] that as well. So, it's -- it gives a lot of [indiscernible] waveform [inaudible] that.
So, with that, you can do things like [indiscernible] duration curves which you can't do with conventional devices. You can vary the pulse with and you can get model thresholds of different pulse width. So, again, here, we ran our model, also with [indiscernible] pulse width. And the red curve is the [indiscernible] duration curve that we got from the model, and the black curve is what we got from the experiment.
Now, the overall amplitude we individualized, so we normalized so that we can see the curvature. And the curvature matched really well between the model and the experiment. We have also run our, at Duke, our first study with [indiscernible] stimulation. We did some pulses. One of them is conventional and three of them are cTMS pulse. So, I won't go into a lot of details. The point is that the -- we got different amounts of inhibition across the pulses when we changed the pulse shape.
So, again, TMS pulses are doing something. And when you look at, again, [indiscernible] here, we got the pulses that, again, have more unidirectional characteristics, we have them in different directions. And, again, when you simulate them, you see this phenomenon in our recruitment -- differential recruitment of the gyral crown versus the gyral wall. So, basically the same kind of phenomena that, again, results in this difference in latency when we measure these pulses.
And, interestingly, these are biphasic magnetic pulses which produced more unidirectional [indiscernible] because you can really play with the [indiscernible] of the magnetic field, which is probably [indiscernible] I don't have time to go on that. And also we played with whether we predict well with thresholds that are -- that we got for these different pulses.
So, again, compared to the simple model that we have previously used in other [indiscernible], which is the linear membrane, which is in red, you can see that our models, more realistic models, which are the green bars, are closer to the actual data, which is in blue, compared to the red [indiscernible] model. Generally, within the standard deviation of the results. So, again, it seems that we are having a tool that's better at predicting what's going on. And, interestingly, these results are [indiscernible] exactly how you define your region of what neurons [indiscernible]. So, there's more subtlety there that we have to really explore carefully.
And finally, just a quick advertisement, we're not just playing with models, we're also doing actual recordings of actual neurons in the brains of monkeys [indiscernible]. During TMS, we have specialized equipment for that. And we have these beautiful pictures of what happens when you deliver a TMS pulse is different intensities; right? It's a stronger pulse if you go up. And then you can see how obviously slowing down you don't have much of a response.
And for the TMS pulse, times zero, when you have [indiscernible], you have this, you know, rapid firing immediately afterwards, and then a silence period, which is actually known for also electromyography and [indiscernible]. And we can look at -- we can identify the point of different neural elements, axons and inhibitor and excitatory neurons. So, it really gives us an exciting window into what actually happens, and, indeed, that we're coupling that with our modeling to develop better models.
So, in conclusion, even though my talk is probably not the usual thing [indiscernible] multi model techniques where you combine TMS [indiscernible] or combine TMS with EEG and, you know, use that to study the brain, I can show you some less kind of obvious ways in which imaging and electrophysiology are now getting more and more important use together. And one important factor is using MRI as a technique that gives us understanding of the structure of the brain, and then you can use this structure in sophisticated models that can allow us to understand better the mechanisms of TMS.
We're also then developing tools, like [indiscernible] TMS and more focal TMS, that can be used in MRI scans, or with EEG. And both of these can have fewer pulse kind of artifacts and also ancillary effects and wanted effects in brain stimulation during stimulation and imaging. So, for example, quiet TMS makes a pretty dramatic difference in both TMS-fMRI and TMS-EEG, since the auditory component is major in both of these. And in addition to that, we can kind of elaborate MRI-derived models with populating the neurons that are actually [indiscernible] derived from stimulation [indiscernible]. And we can also do TMS with neuron-level electrophysiology to understand what actually goes in the brain and use that for our model. So, thank you for your attention, and I'll be happy to answer a question or two.
So, if any of the participants have questions, you could put them in the chat window. I wanted to ask how does the modeling the findings of the crown versus the wall [inaudible] greater, what direction [inaudible]?
The question, that would be quite a wide leap to, you know, say anything about that yet. We are also doing work on getting cortical [indiscernible] models, where you, you know, try to wire everything. And then you have to add, you know, synaptic [indiscernible] to that. Then we can start to really understand that better. So, we're not there yet. So, we can't -- you know, I'm not even trying to explain that other than say, look, we are starting by stimulating different neural types in different [indiscernible]; right? You can see how that then ultimately leads to different modulatory effects. But exactly how that happens, we'll need more circuit-level models. And, again, that's something that we're interested in and working on.
The comparison you did with the simple linear, the realistic and then published, was that just to the published normative data or is that using those -- the actual MRIs from those published studies?
So, this was basically [indiscernible] comparison of model thresholds. And then you take the -- so, just to repeat the question, the question is whether we have these -- when we have comparison with these simplified models and measurement data, and our more complex models for thresholds. We're mostly looking at ratios because, in terms of the exact absolute threshold, we still, you know, it can be off by some percentage. So, that kind of just means that you can practice to tweak, you know, in the model. So, what's more important to us right now is to look at the ratios because it shows different -- and there we normalize what we -- what we did is we normalized something. For example, we normalized the stimulation for one pulse, and then we look at the differences for the other pulse, which is equivalent to looking at ratios.
Sure, but you're comparing to the published normative data; right? You're not looking at each individual MRI and then reconstructing that, and then looking at that ratio for that --
Okay. Yes. No.
Okay. All right.
We're not looking at individual MRIs, no. This is -- yeah, this is, right now, it's just one [indiscernible] model, one [indiscernible] model. That's why I'm saying that the proper error parts in the data is the standard deviation, which gives you the variation across subjects because its past subject was not even part of that study. So, still, you know, it has to -- again, we're building from the ground up; right? [Indiscernible] the neurons, and then, you know, and then somebody can start putting them in individual head models. That's actually straight forward. It's just a matter of acquiring the MRI, you know, from the same subject that you did the experiments with, and then populating it and doing the results of these calculations. So, they should be doable [indiscernible].
I'll add a little bit more. So, you're showing this [indiscernible] have you looked at how your prediction for the layers firing [ph] compare with the EFIS [ph] data? Obviously you would need a multi-head model with that, but.
So, we are -- with election [indiscernible] unlike the simulations that take a long time to select data. So, you see here that we have 287 neurons. It is basically 287 work dates, plus, of this study. You know, it's very, very complicated. So, we are very interested in trying to get information of actually where exactly the electrode [ph] is. And you can -- if you sacrifice the animal, you can tell that -- these are monkeys that are used for many years, so it's not like it's one experiment and you sacrifice the animal. So, it's a bit tricky to establish where exactly you are. And we're looking at techniques that can allow us to identify more specific [indiscernible] and then we should be able to get this information.
And also we're using these rather simple but robust [indiscernible] single channel. And from that we've taken analysis. You can extract up to, like, three neurons at best by interpreting the amperes and waveforms, but you can use also, you know, more complex electrodes that have multiple contacts, for example, you know, [indiscernible] getting multiple contacts across layers. And that also may be an approach going forward.
You know, the issue here is that the throughput of this study is quite challenging to get high throughput, but we have a nice funding. So, we're doing modules on the scale of the [indiscernible] so we can run more conditions and solve more questions at the same time. And this is the [indiscernible] work I want to mention is done in the lab with my collaborator, Mark Sommer. I believe there was a question online.
There is one. It says, "How sensitive do you think these results are to gyral geometry, and if you have validated these simulations using more than one model?"
So, no, we have not delivered multiple models at present, although that would be really straight forward. And so, in terms of sensitivity, we have not evaluated that quantitatively. Qualitatively, I would think that, yes, it would have an effect, especially at some level, but that may be more at the level of electric field rather than, you know, the neural -- because you still have -- unless you have some really unusual subjects with some differences like thickness of cortical layer, you'd expect that that structure is quite [indiscernible]. So, then most of the difference will result from the cortical geometry that effects electric fields. So, it's not just the kind of the gyrus, but it would be more like the gyrus and surrounding sulci and its particular type, and the overall geometry of the conducted layers that you have around it. So, that can be -- that can be explored.
I think that the main effect would be from the [indiscernible] for example, in the hand knob, from the fact that people have very different hand knobs. And in some MRIs, some subjects, you don't even exactly see a knob, you know, at least on the surface. It's kind of hard to identify. So, I think this variation can result in differences. And it would be very interesting, indeed, as Dave [ph] here was suggesting, can we do that individual MRI. We can do that, and that would be very interesting study to correlate actual experiments with stimulations in the same subject and explore that effect.