 Diolch i chi, dyma'r haf y dyma'r bod yn ystyried. Rwy'n gŵr i'r bwrw yn Llyfrgell i Cylligeu Hymraeg. Roeddwn i'n gweithio bod yn ystod cyllidegrintoch yn ddegion cyllidegrin i Llyfrgell i Cyllidegrin Cysur. Llyfrgell i Cyllidegrin Cysur eisiau yn ddiolch y Cymraeg, ond mae'n cael ei gael cymaint o'r ei reio. Mae sydd wedi cael ei gymryd neu'r cyllidegrin Cysur argymrydau. Cyllidegrin Cysur eisiau eisiau yn dda gydag ychydig o'r ffwrdd gwrthig ychydig o gyfrifetio'r ffwrdd gwrthig yw'r ffordd gwrthig. Ond mae'r ffordd iddyn nhw'n mynd i wneud y cyfrifetio'r tîmau o'r dŵr y sefydlion yn y cyfrifetio. Yn y cyfrifetio, mae'n mynd i'n mynd i bwrdd 65 miliwn ar y cyfrifetio. Mae'r gwrthig yn ychydig yn y bwyr. Mae'r gwrthig oedd y r�ie yng Nghymru ond mae'n gynhyrchu'r gwrthig yn y cyfrifetio. the developed as well as less developed world and seizure control that's a number that you usually see is not effective in 30% of patients and even in the 70% of patients where it does work the medication side effects often are a big problem or a big challenge to patients and so the improvement of patient treatment and their care it really is needed to improve their quality of life and I just want to highlight I'll show you how a seizure looks like so that you get a bit of a better feel what I'm talking about when I say it's actually a disturbing disease that really wants our attention from research so this is a patient that's I think 16 or 17 years old in this in this video and he's just coloring his book in at the minute but in the second you'll see him freeze about now and then the seizure starts in a second and you see this head so this involuntary head movement including then the arm movement and the sound and then the convulsions etc I'll I think I'll not let this run the whole basically this lasts for about a minute until the patient then stops and essentially is in a dazed state and so that's how a seizure could look like different seizures can look slightly differently it's just an illustrative video and how clinicians then diagnose seizures is usually by using EEG so I don't think I need to explain to this audience what an EEG is basically measure the potentials from the brain and I've got here I've just used the mouse actually I've got here a trace of a of a seizure actually in a patient and in this particular example is from this electrode that you see some abnormal EEG activity and abnormal in this case refers to these spike and wave kind of activities and then experience clinician will tell you okay that's probably suspected seizure activity okay so the from a research perspective then that gets right into the the topic that I want to talk about today is how when you look at the EEG or in the in the intracranial case the ECOG how do you decide what is a seizure how do they look like how do they start just morphologically from just the observation of the um electrographic signature and you can actually um or in the past people have actually tried to categorize um how this morphology looks and the the consensus I would say although there's not it's a bit of a battle in the field but the consensus is essentially that over 50% of seizures start like this um in the way that I've highlighted it so the you would call this segment a normal um background activity and then here you see some higher um slightly higher so this higher frequency oscillations that are fairly low in their amplitude building up some sort of abnormal activity until it gets released into a higher amplitude oscillations which you would definitely call epileptic spikes or epileptic spike waves in this case so this seizure in starts with a low amplitude um fast oscillations so that's what most people call these types of seizures um but there are other types of onsets that you can observe and I've broadly categorized them into um what I call high amplitude slower oscillations so essentially the amplitude of onset is a lot higher than what we observe here and the oscillatory nature so the frequency of it is slower so this usually is above alpha so uh so where people set the cutoff is is dependent on the clinician but um usually people say over 10 15 hertz is what you would call this low amplitude uh fast activity and then this is usually in the range of um below 8 hertz um sort of activity and that happens in 25 percent of focal seizures so just to quickly summarize um these different patterns so the different patterns can be identified and can be categorized and most commonly what we see is this low amplitude fast activity and the sempicon most common is this higher amplitude activity of lower frequency and just from a scientific perspective of course you can ask how can these different patterns be explained because there are clearly categories and why do we see in some patients one pattern and in some patients the other pattern and we took essentially a computational modeling approach to try and explain this phenomenon uh the model that I'm going to talk about is actually a previously published model and it's actually also available on model dv if anybody wants to look it up when I get my mouse uh the model id number is there and the model is essentially using the idea that um the computational unit of the cortex is a a cortical column and uh we basically take a lot of these cortical columns and concatenate them to form a cortical sheet and so this is illustrated here we have a column here and then we concatenate them into form a cortical sheet um each column in this case is modeled uh very simplistically by uh Wilson Cowan unit and this was in this case um actually sufficient to capture the dynamics that we wanted to capture um you this this model obviously also needs uh connectivity so we've been talking about that a lot and this connectivity was um are we basically drew from the existing literature in terms of what this mesoscopic connectivity is I may just draw your attention also to the scale of this so we assume essentially each uh column to be of 50 micrometers and we concatenate them to form about um the cortical sheet of seven and a half millimeters by seven and a half millimeters so this is really talking about mesoscopic connectivity here and what we know about it or what most people would agree with is that there are feedforward projections from excitatory to excitatory populations which is like in the in the distance dependent manner so that's what this is showing so this is illustrating the connection from one column so in colored in red to and its projection its target columns colored in black so the red is projecting to the black dots in this in this case um in the feedforward excitation projection um equally there are also feedforward inhibition projections so excitatory populations projecting to inhibitory populations and it's thought to be that the projections are actually um working in a similar manner but additionally there is a um or there's evidence for another type of um mesoscopic connectivity which is this um patchy remote connectivity um it's been termed in the literature which is essentially that um this red dot can actually also project to slightly further away um columns but um they usually form patches like this and so all these types of connectivity are included in this model and we're basically trying to model this uh cortical patches in terms of its connectivity as realistically as possible and the types of outputs you get from this model um I'll just show you some examples um at this so this is a normal background normal background it's background activity um so there's not much activity going on you can just see some background fluctuations so it's because that's driven by noise the system but you can actually also find a more coherent oscillatory state in the model which we set or which we equal to the seizure state due to various properties mainly the oscillatory component of it and because in the model we can actually find these to this this background activity and seizure activity we then analyse the model in terms of different parameters and drew conclusions but oh that so that's in the previous paper but in this paper we actually want to focus on how the how the seizure started how the seizure patterns actually start and the way we've done it is basically by taking this idea that seizures actually start from tiny little areas so called micro seizure clusters so these are shown in black here essentially I have these little clusters of mini columns that actually behave slightly abnormally in that they show seizure activity or they can spontaneously show seizure activity and we have a couple of them and they can show seizure activity independently but they're embedded in this healthy surrounding tissue meaning this tissue normally you know if you don't provoke it it won't show it won't behave abnormally so that's the model and using this model of course you can already imagine there's a lot of parameters to to adjust or to play with and to actually get a seizure and just show you an example seizure that you can get out of this model so above you will recognise this clinical recording again of the low amplitude fast activity that we showed at the beginning and you can actually find a similar activity in the model as well so this trace the simulated ecoc recording I've essentially taken the average of this cortical patch that I've been simulating and essentially pretended that there is an electrode above that patch and then that's the recording that I obtain and you'll actually find fairly similar activity in the model as well in terms of the onset pattern you can see at the beginning that you have some fairly low amplitude fast activity which evolves then into these higher amplitude more coherent oscillatory activities you can then of course because we have the information look at what is actually happening spatially temporarily so what is happening in these different areas on this patch and you'll see so this red line marks where this t equals zero is and you'll see that over time essentially slowly some of these micro domains become active show seizure activity and that that is essentially this phase where you see this low amplitude fast oscillations and then they kind of they form little pockets which then coalesce in terms of seizure activity to then recruit the entire cortical sheet into a into a seizure which is in this period here essentially okay so that's good we can essentially find some some correspondence in the model of the onset pattern that we after can we find the other high amplitude onset pattern as well and yes the answer is we can and basically it looks something it can look something like this and if you look at the spatial temporal evolution of this seizure again you'll see that it's actually very different so again the red line marks t equals zero and you'll see that the evolution of the full recruitment of the sheet happens a lot faster and within 0.4 or within the first 400 milliseconds essentially the seizure has already spread to about a quarter of this cortical sheet which is why we see this high amplitude onset essentially so in the model essentially we can find different types of onset patterns as well and then we were interested in okay so what are the model parameters actually determining these different onset patterns and I've performed a this is a parameter scan essentially I'll walk you through the different axes slowly so the the first axis I want to highlight is this number of subclusters well that is essentially if you remember we essentially had these micro seizure domains that were scattered in this healthy or embedded in this healthy tissue and the number of subclusters essentially refers to how many of so you know how many of them are scattered in the sheet whether it's one big one or several or if they are divided into several different subclusters and the percentage of mini columns of micro seizure activity is essentially the percentage of the black versus the white so how much healthy tissue do we have how much abnormal tissue do we have so these are the two parameters that we're scanning and this third parameter that I'm scanning is essentially the excitability level of the surround so this this healthy tissue here that I've said how excitable is this and you'll see in a bit what I mean by that essentially it regulates or this parameter in the Wilson Cowan essentially regulates the background input level or the excitability level of the Wilson Cowan oscillator okay so with these three parameters we can you know have a look at the parameter space scan and you can visually already I should tell you what actually the dots are so the dots are the dot size so the bigger the dot the higher the amplitude of the seizure onset and the darker the color the higher the amplitude of the seizure onset as well so visually you can already see that the high amplitude onset seizure seemed to be located in a area of parameter space that's on the high end of this surround excitability and actually we can show this a little bit better even basically what I've done then is to collapse all those data points that you've seen previously just on this one parameter this surround excitability and you can see that the seizure amplitude essentially drops with this parameter quite clearly and you can actually look at the onset frequency as well and you'll see that there is a distinct jump here between the higher excitability levels and the lower excitability levels in terms of the onset frequency okay so that's interesting basically it's it's telling us that this surround excitability seems to play a crucial role in determining what the seizure onset pattern is and to actually put a cherry on the cake basically you can actually analyse the model and look at dynamically essentially what the model is doing and what you find is that in this highlighted red region here you'll find that the model is by stable whereas in this white region is mono stable so what I mean by that by stable in this dynamical systems concept basically means that the surrounding this healthy surrounding that we that we deemed healthy actually has a coexisting seizure state already and it only needs some provocation to move into this coexisting seizure state whereas in the mono stable state there is actually only the normal state and you have to work really hard to really create that second seizure state so you have to provoke it very hard to actually make it essentially create the seizure state from the input whereas here it's already pre-existing so that's something so that kind of from a modelling perspective makes sense basically meaning that if your surround is already ready to take up the seizure activity because it's basically just sitting then you only need the provocation it makes sense that this that this onset patterns are very high amplitude because they rapidly can generalise over this whole sheet and whereas in this case you kind of still have to build up and invade the surrounding and then hence that's generating this low amplitude fast activity okay so that's all nice from a modelling perspective you know we're happy as modelers um but it actually gives a very um it gives a model prediction as well which is in which can be put back into the clinical context and that is that it actually makes a prediction on surgical outcomes so what do I mean by surgical outcome in these in these seizure patients in these focal seizure patients what happens with them is that if medications fail or in particular cases where where clinicians are fairly sure where the seizures are coming from they're actually proposing to remove the piece of tissue that's generating the seizure activity and so we're essentially um doing epilepsy surgery the surgery success rates these days is um depending on what paper you read ranges between 50 and 70 percent and so there is in a in some patients the surgery is not working so well and others is working better and in our model actually or from our model we can actually make a prediction regarding what onset pattern would predict a better surgical outcome so in the low onset amplitude a sort of low amplitude onset pattern um we would essentially so we would essentially say that because the seizure kind of starts from these little micro seizure activity but it needs a long time to recruit the surrounding tissue um to become this full seizure activity we predict that if you were we if we were to remove a piece of the tissue then maybe we can cut out this essentially in this simulation it was actually this big cluster here that in the first place started the invasion into the surrounding tissue and hence cast this full recruitment if we remove that cluster of micro seizure activity we can actually stop the seizure onset in a way and although there are still other seizure clusters around they don't recruit anymore because they are not in the because they're not um either not spatially close enough to recruit or to recruit the tissue between them or they're just not enough of them so essentially the prediction here would be that if you see a lot low amplitude onset you would say that they probably will have a better chance of success because the surrounding tissue the healthy tissue is actually healthy because there's no coexisting seizure state but in the high amplitude onset pattern because from the model you know we we predict that it's actually the surrounding tissue that is already probably impaired and has got a higher excitability level and probably a coexisting seizure state already we would say that the surge risk assess is probably not so good because after you remove um a piece of tissue in the in the high amplitude setting the although this healthy surround although this surround is basically the surround can still be provoked to have a seizure from other seizure micro seizure cause and that's essentially illustrated here basically that even after the surgery although the seizure starts slightly later compared to this original simulation you can still see that a small provocation essentially starts off a full seizure recruitment again um this prediction is um it comes from the model but we can go back to the literature and actually see if there is some evidence for it and although the evidence is not 100% clear cut because the I think it's a problem with actually identifying what patterns uh what people actually associate with actual what mythology um despite that there is I think some good hints that that's actually the case so um I've highlighted four papers here and essentially all of them say um so this uh low voltage fast activity and rhythmic sinusoidal waves were associated with a favorable outcome more often than the other three patterns the other three patterns were in this case higher amplitude onset patterns um this paper here equally fast focal activity at onset was associated with a favorable post surgical outcome um here they distinguished it by frequencies but also that the higher frequencies so higher than 8 hertz indicated better surgical outcome here this is a slightly less clear cut um where it says basically that the low voltage fast activity but also higher amplitude beta spikes predict a better surgical outcome than the other types um so we can see that there are there are definitely indications going in the direction that this um that the onset pattern is actually a predictor for the surgical outcome further consolidating the idea that are coming coming from the modelling that this um that's essentially the surround excitability will play a crucial role in determining your surgical success there's actually a one piece of direct evidence as well for what we're proposing in the model and that is um from Inazmeral and they essentially tested the excitability level in different onset patterns and what they show is um so in this is the low voltage fast so the the one that where we predict a better surgical outcome and where we predict the surround is actually less excitable um they provoked or they stimulated um the the cortex of these patients and here you see essentially the stimulation response um the one to focus on is the solid line so that's the stimulation response in this case in the low voltage fast and they repeated that for the high amplitude um spiking pattern as well in this patient and um the response is this one and essentially they argue that the overall response um if you were to put an integral over this response is higher than this and the amplitude in general also higher so that's what they say in the in the results basically with repetitive spiking group amplitudes of the um stimulation response is higher than in the paroxysmol fast group so basically saying that if you were to test the excitability levels they find that this high amplitude a case gave higher excitability levels than this which is exactly what the model is saying with the surround excitability as well okay just to conclude this um the different onset patterns um different onset patterns can be identified and focus seizures and categorized and our model suggests that the onset pattern might be associated with fundamentally different settings of the surround excitability and I just want to emphasize here actually it's not the way that these micro seizures clusters are organized or how the seizure starts by itself but it's actually what we think is the surrounding healthy tissue that's determining one case against the other and this notion that we're proposing is also supported by some evidence from the literature either by direct stimulation experiment or in terms of the surgical outcomes in the two patterns and because I was told I have a bit more time I thought I'd squeeze in an outlook um so the outlook is is essentially asking the question so what now if if that was true what we're predicting in the model then we're essentially saying that you know in some in some patients with the low amplitude faster activity they will be fine because if we treat them with um with surgery then they should you know if we really get the focus then they should be okay they should recover but what about these patients that actually have the um high amplitude um oscillations because we're saying that that this around excitability is increased meaning that even what we think is the healthy tissue might be impaired so there are essentially two things that I would like to propose in that context and um one of them is to actually test and track these excitability level changes in the surrounding tissue or in general in the in the cortex and I've done some simulations in this case um where I've changed the again this parameter here that essentially determines the surround excitability and I've simulated the corresponding eeg for that and you can see that essentially despite these changes in excitability levels there's not much change in the actual underlying eeg um which is exactly the scenario that we actually have with seizure patients you know when we just look at that background eeg we don't actually know what the excitability levels are that underlies you know what state the what stays there in um but with repetitive stimulation we might be able to get at that it's this same idea that's that I've shown you before in the clinical context where you essentially stimulate the cortex and measure the amplitude of the response and that should hopefully tell you something about how provocable or excitable this cortex is and if you do that and measure the stimulation responses you can actually see that it gives you in the simulation um a profile like this and you can actually correlate these two so essentially you can plot the um excitability level here and you can plot the average response amplitude here and you see a fairly tight correlation between the two so it's just in a model uh demonstrating demonstrating the proof of principle that we can possibly actually get at these excitability levels and very nicely actually recently uh Christian Meiser published a paper in PNAS um showing exactly that point and that you might even be able to track these excitability levels passively without the stimulation so that's I think one way forward um because if we can actually track these excitability levels there might actually be a way to control them and that's essentially um the next point that I'm trying to show in this outlook what we can in terms of what we can do um the so if imagine that you can track excitability levels and essentially almost predict uh when patients are more likely to have seizures you can imagine a close loop control device um that interacts with the either the excitability levels themselves or with the seizures and I've just done a simulation of the principle essentially um that I envisage so here um I'll show basically some background activity and then about halfway through this video um there will be a seizure provoked on this uh cortical sheet like there and the seizures would spread and recruit the entire sheet so that would be how the seizure would be happening without any intervention but if we could imagine that we actually could track this rise in excitability level leading up to this seizure and interact you know and know when to interact with that um you'll see this in the second video this is with close loop control essentially and what you'll see is actually you'll not see very much because it's controlled almost immediately by this close loop control device oops I don't know if um I'll just show that again it's about halfway through you see this activity popping up but immediately being controlled by the um by it's essentially a three by three um micro stimulation device and then the seizure is controlled and you can actually move on and you know basically the patient wouldn't have any symptoms or hopefully wouldn't have any symptoms in this case um so this is essentially the outlook I wanted to show that um with these um tracking and intervention of with these excitability levels we can maybe actually get a handle also on the second group of patients who um have these high amplitude kinds of um onset pattern okay that brings me really to my end um so acknowledge my um collaborators in Newcastle as well as um at Columbia UCL and Kings College and as a final slide I just want to draw some attention to a conference that we're organising in Newcastle on computational neurology and I think that's very relevant for this audience um the abstract submission has opened and uh it will the conference itself will run 20 years and 21st of February next year and yeah we'll welcome you all to Newcastle please register well please submit an abstract and then register thank you very much for listening I have a question in my opinion uh the model the wc model can permit you see these two states because the bavocation of this model permit have the oscillation of fixed point and by changing the pyramid you can observe the state the dynamic state of the the network can jump from the uh fixed point driven by a noise so you can see this so-called small amplitude to the oscillation state this low frequency so my question is however you as you as we know the system is a periodic attack disease so that means after a period of time it can go back to normal by itself so how can you explain it since the pyramid already jumped from the fixed point to oscillation how do we jump back it seems a back and forth the pyramid in your model so how can you explain that yeah good question um it's actually a slightly different field of research that um I've not touched on here at all because I've was more more focusing on the clinical aspect um yeah it's very good question um essentially if you look at it from a dynamical systems perspective how can you conceptualize it and you're very right um one group of people say it's a bavocation through um you know whatever this hop hop for a homo clinic and you essentially go through this background state which is a fixed point to a oscillatory state um limit cycle and then the parameter driving that is some sort of rain state change that can you know drive you into a seizure and then they say okay the seizure would then terminate through this parameter changing back again and that's a reasonable proposal I think um but I think so far nobody has been able to actually show what this parameter is how we can control it etc it might be this excitability parameter it might be something completely independent um but there are alternative proposals out there actually as well um there are okay one very prominent one of of course is that from lopistasilver where they say actually it's not a bavocation it's a bias stability essentially these two states coexist as exactly exactly what I was showing in that model um as as one of the mechanisms to distinguish the high and low amplitude onsets so in the bias stability so that's bias stability in a spatial temporal context what lopistasilver meant was simply that this through this simple coexistence um through noise on the system you can actually just get um while random jumps to the oscillatory state and random jumps back so you don't actually need a parameter change or you need is a noise that drives that and the onset and offset would be perfectly determined by the basing to buy this coexistence and how far they are in phase space that explanation also has got its problems when you test it in a clinical context and there are other proposals out there such as excitability and it being actually a um uh turbulence-esque um attractor but I think all of these concepts can explain some proportions and some phenomena but I think it's a as a general question it's not an easy one to answer and don't think the definite answers out there yet. When surgeons take the tissue out do you then test it in the lab to see if you can detect any signs of the uh the sorts of mechanisms you're talking about? Sorry say that again. When the surgeons if the surgeons actually take tissue out to try and cure it can you then test that tissue to see if there's a direct evidence of your of your mechanisms? Yeah yeah that's a very good question um and there are so in Newcastle actually we have the lab facilities to take out the um so to take the samples out and test them in various ways electrophysiologically as well as um by other chemical means um the the the test of performing it of um high amplitude versus low amplitude onset has not been performed to my knowledge ever yet in real life um it's it's probably worth doing um but the other thing or the other problem that comes with it is as soon as you actually take these tissues out of their embedded environments you might be actually destroying some of these excitability um so if it's locally embedded in the tissue yes but if these excitability levels are actually driven by either the network or some inputs from the network then it might be more difficult to track so in a way it's it's there's it's not an easy experiment to directly address what I'm proposing in the model hope that answers your question um so how long time does this seizures take in in the clinic or how do they stop and how how do you think are they jumping back then to the original stable state or what is happening okay um from a so first the clinical side um the seizures usually last seconds to minutes um there are some there are some unfortunate patients which have something called status epilepticus where the seizure is thought to last for hours um although I don't think they are true I don't think it's the same seizure state as you see in the patient's way it's only lasts for seconds but so that's on the question of duration but that also makes the picture that we are trying to model actually slightly complicated you know what is it that we're trying to capture in on because the timescale of these seizures can actually range from seconds to hours so it's not quite clear so in the from a modeling perspective how the seizure stops I would say I would give the same answer as actually previously I don't think there is a definite answer out there as to what the mechanism is a lot of mechanisms have been proposed either it's a parameter change back or it's this by stability where you use through noise can go back to the seizure state um there are ways actually to test for these different scenarios more or less um and I think depending on the seizure and depending on the patient it might be entirely different mechanisms as well so yeah unfortunately it's a very handway the answer but I think the I think this actually emphasizes something from the talk before that actually precision medicine here is is is required we can't actually just say oh yeah seizures in general start with this and end with this is actually a family of diseases and it's actually I think quite a heterogeneous group and that's why we're observing these different surgical outcomes as well you know we can't just treat everybody was just yeah my opinion but yes okay another question up here thanks very nice work thank you very much quick questions about the the model first do you have any delay in these connections or an instantaneous and then second question as if you do the rapid recruitment do you require prior um basically I think it was all for whatever you had a background activity or could you actually do this rapid recruitment from basically zero is there is a requirement of prior isolations or not there's quite often a quite different recruitment in in in such models okay the just to clarify actually I think that I didn't explain that very well the the background state isn't actually in the dynamical systems sense it's actually fixed point with noise that's why it looks slightly oscillatory it's around yeah so it looks slightly oscillatory but irregular but it's actually just a noise driven node in that case the and to answer your question about the delays the simulations I've shown you were actually without delays I've repeated the experiments with delays it doesn't actually change this bifurcation points for the bi-stability and the other points at all and the results are essentially qualitatively the same okay thank you okay thank you very much so