 Ευχαριστώ για την εξαιτήρική εξετήρική. Είναι η πρώτη φορά που μεταξύγει την εξαιτήρική μου στην εξαιτήρική εξαιτήρική εξατήρική εξαιτήρική. Απέτυχα για το μου σκέφτημα, one of the things that I am working apart from in general brain connectivity and muscle learning and how we can extract real-world connectivity by a micro-system. Many various disorders and disease. I am working on how we can define one connectivity estimator that can quantify έναν τελευταίο τελευταία που υπάρχει στη χειμωμένη χώρη με αγγελμακνήτρας. Και πώς μπορούμε να εγκοπωρήσουμε δυσκολετικές εμφρικουσίες, εμφρικουσίες, εμφρικουσίες κόπλοι, με ανοίκομα και εμφρικουσίες κόπλοι. Και αυτό θα είναι το πρώτο στιγμό μου πώς να εγκοπωρήσουμε όλα αυτά τα εξηγησία σε ένα σύγκολο αιστηματία. Ποια εγκονοκτύρι. Στην my opinion, this is more informative and more interesting to study brain connectivity compared to resonance activation. It's more informative practically and also it can be interpreted much more easier using many other physiological models. Αν λαγκλή activation detection connectivity can give you much more information about different conditions in cognitive neuroscience and what changes in brain connection during disorder and disease. There are many types of connectivity analysis based, particularly based on fMRI, network based using DCM or structural equation modeling or structural vector regression, but my approach would be functional connectivity b-variate using symbolic transfer entropy. There are numerous connectivity estimators and that can change the interpretation of results, like mutual formation coherence, imagine body of coherence, phase like index, granular causality, and many others to estimate also inter-frequency capping like phase locking value, modulation index presented by TOR and GLM by VINWINK. The main issue is that even though two or more groups of people are working on the same data set using Meg resting state, the same for an Alzheimer's disease, using the same social glasses algorithm, then it's very difficult to incorporate and to find consistent results. This is a main issue in brain connectivity analysis using neuro-electromagnetic recordings. The second thing is that I'm working already with Babel's work based on a metal arithmetic task and it's very informative. It's how to integrate all the types of interaction into a single graph and this is very informative and it is not just a technical issue but it's something that can be interpreted. Here is my first attempt to integrate both inter-inter-frequency coupling. This is a metal arithmetic task where I estimate the phase locking value, phase locking communication within theta, frequency in front of our areas, within paraccipital size in alpha 2, cross-frequency coupling you can see with green and then with insert of the cross-frequency coupling. On the right you can see with color how I integrate in a fax on connectivity graph with different types of interaction. This gives you a much more higher discriminated power but also a much more interesting result in terms of interpretation and connected cognitive neuroscience with different types of interaction. A symbolic transferator was first presented by Staniek in an epileptic study where I just demonstrated that it can detect the hub that propagates the seizure and demonstrated that very significant and very robust noise. The whole approach is, the symbolic transferator is based on transfer entropy on information theoretic approach but you must transform the symbolic time series in the symbolic sequences. We used the delay version of symbolic transferator which is the same but using different time blocks and basically made the symbolic transfer entropy between two times A and B and B and A and we take the difference and if it's positive then A drives B or if it's negative then B drives A. To apply a shallow gate analysis you must sort of, you must first, you must soft link one of the two time series symbolic sequences and then you can estimate symbolic transfer entropy in 1000 to 3000 times and then you create a distribution and you assign a pay value between those two time series. This is just a demonstration between FP1, EG signal and P7 in parietal, in front of the 3D hyperexpital alpha 2 and in C you can see how delay symbolic transfer performs in different time delay samples. You can see with black crosses the significant level using this shallow gate analysis and indeed you can see if you symbolize each time series independently and this is the most significant but technique that we can see in literature. The dependence symbolization for each time series is that you must define a better dimension time delay for your time series and then you reconstruct the space for each time series. For example if you choose a better dimension equals to one then you order this every sample of the time delay across the three dimensions and this gives you three factorial equals to six possible combination of this ordering and then you transform your initial time series into symbolic time series using symbols from one up to six that are related with this ordering. But this is problematic because it works independently for each time series. We propose the neural gas algorithm, this is a famous Lorentz system that you reconstruct both time series into the same space and then you find common symbols between the two time series. This is a demonstration in the meta arithmetic test of five levels starting from summation of two digit, two single digit numbers up to five which is the summation of two, three digit numbers and definitely you can see the drive of frontal theta to paraxipital alpha 2 incorrect trials compared to wrong. Using this delay symbolic transfer entropy we succeed to find a non-time lag equals to zero in between left and right the frontal areas compared to right paraxipital alpha 2 which is one of the issues that discriminate correct and wrong. So this is the first demonstration of this symbolic transfer entropy so both the strength and also the time matters to discriminate correct versus wrong trials. I applied symbolic transfer entropy in an open database provided by Michael Kanada, 87 subjects from 18 up to 64, 65 using a electronic system, a source localized using LCMV and using the 90 AL atlas. I defined symbolic transfer entropy, delta symbolic transfer entropy using this equation in order to describe the dominant, the preferential flow from anterior to posterior. If the delta symbolic transfer entropy rates between 0.5 and 1 or from posterior to anterior between 0.5 and also if a posterior anterior index average of posterior anterior rise in order to find the preferred information flow. In order to assign a significant value for posterior anterior index I solved the average delta symbolic transfer value across posterior anterior rise. I repeated 10,000 times and then I create distribution in order to assign the pay value to this posterior anterior index. The whole approach using this symbolic transfer entropy was realized independent for phase and amplitude. Using six predefined frequency bands I demonstrate that in all frequency bands with exception of theta in both amplitude and phase we can see in A amplitude in B you can see anterior to posterior. Posterior to anterior information flow in both domains and across frequency bands and this is a very significant result even in a trusting state which is important. The posterior anterior index was very significant. You can see positive values. This is the person that shows how as an index of the significant of how posterior drives anterior for positive values or anterior to posterior for the opposite. You can see only theta with minus which means that from the anterior theta drives posterior theta and all four subjects were excluded as outliers from the whole cohort. The time lag even though I estimate the whole approach in B variate and one can say that there are many indirect connections via third node are also above almost 100 milliseconds. We can see different time lag between phase and amplitude domain. As a comparison I use phase lag index which is very much presented by STAM very robust volume conduction in order to detect posterior anterior trend for phase dynamics and transfer entropy for amplitude dynamics but I didn't succeed to detect significant values because this is across the whole cohort. The important thing is how to my assumption but this is not my assumption is if two brain areas are connected then they should be connected with one type of interaction within amplitude, within phase, within the same frequency or between two frequencies. So if we estimate the using the delay symbolic transfer entropy or any other type of interactions the connectivity between every frequency band that we selected and also between different frequency bands you can see the diagonal verb and this example is the delay symbolic transfer entropy within frequency bands in the upper triangular is the cross frequency coupling. We can find this matrix called commodulogram that can represent both inter-endifice coupling then we can apply surrogate bofuronic corrections and we can there are three cases. The first case is that only one survived the bofuronic corrections the second one is that's more than two then I selected one with the maximum delay symbolic transfer entropy or none survived so this means that two brain areas are not connected and of course we're not expected to have a full weighted graph. If you apply the same thing in our cases in our database for amplitude and phase you can see clearly see even the tracing stage are both within and also cross frequency coupling especially with delta, delta theta, delta alpha 1, delta alpha 2 you can see the first column in both commodulograms both matrices. Cross frequency coupling is very important for the communication of two different two distant brain areas that oscillate on a dominant frequency like from the theta or parahedra alpha 2 and it's very important also as an index of maturation and you can see that less cross frequency coupling for example in normal aging also in alzheimer. And if we define a very simplified but very informative developmental index if we sum the probability distribution of inter frequency coupling which is the upper triangular and divided by the sum of inter frequency coupling which is the main diagonal independent of phase and amplitude and if we degrade them in four age groups you can see a small increment between first and second let's say decade. And decrement in 50 to 60 compared to 28 up to 14 both aperture and phase. So it's very informative to incorporate and degrade in one single fractional brain network all the different type of interactions simultaneously and do so we need apart from shadow gate and statistics and fast discovery rate for correct for multiple horizon. And to diminish the effect of fast positive we need to find a model or just the less bowling transfer entropy and because I'm working lots of the base in schizophrenia matter of my brain injury alzheimer and this give us a lot of power and it's very important also to work in a more dynamic fashion. This is a preliminary result I'm trying to to add more subjects up to 200 and also to how to combine different systems make system like CTF with electra how to how the whole approach will be different but it's just the first significant interpretation. Here are the three references based on the less bowling transfer entropy applied in a metal arithmetic task the method results and this a new one how to discriminate correct from trial in a metal arithmetic task. Thank you for your attention. Thank you very much. Questions. What is the advantage of symbolizing your time series compared to just considering. It's it's more robust to noise and then we're proposing a step. So it's. Using as you can see that's why I adapt transfer entropy couldn't result in so consistent results. So it's. I cannot explain let's say new physiology wise that but it's definitely the results demonstrate that this is more significant for noise we know we played with simulated data but original data set that's more robust to noise and noises could be any type of noise not system noise because of the source localization we cannot. Let's say accurate find the sources but it's more more accurate and also based on the discrimination because I'm working with five defined biomarkers. It's much more robust compared to using transfer entropy or any other information theoretic approach but it should be done. On the common space not you shouldn't apply civilization on a single time series or applied some. Yes. Thank you very much.