 And thank you very much to the audience for staying to the end and I know it's very challenging especially he didn't care So thanks a lot, and I apologize somewhat for my voice. I'm fighting with cold the last few days I hope you bear with me So I'd like to say a few words about our efforts in the last decade and Show you what you can do with a team of one or two programmers over than 70 when you have limited budgets So I started Does it work it does So so what got me into this field was some experiments in electrophysiology our colleagues at the Institute were interested in context-dependent information processing in in Rats a much sensory system. So they had an awake Immobilized rat they flicked the whiskers and they looked in the information processing in the Falamus and in the cortex in The situation where the rat was habituated or when it was awake It turned out that actually the activity in Talamus was very complex So when they did an acute experiment where they took Local potentials on the grid of four times five times seven points so 140 points and here is an example Set of results You see some example 16 potentials taken at these points and you see that there are some traps and there are some peaks Which are quite synchronized This is a consequence of the long range of the electric field. So to get this we completed the extra cellular we completed the current source density and What you see here, there are four different representations of the same slab of forebrain and in the middle you see just five consecutive coronal sections With some sinks and sources Happening about 3.2 milliseconds after deflection of whiskers, which we think is a response to a single whisker flick and this is This is not very responsive and this is 0.2 milliseconds later and you see Incoming input from multi whisker flick. So when you look at this This discurrence and these sinks and sources seem to be roughly in the right place in the right structures They are not perfectly synchronized. So you may say, okay No, maybe it's the consequence of the variability of the brains. Maybe it's just problems with the synchronization But what do you do if you have several animals and you would like to get some precise information about what's happening where? Now that was that was ten ten years ago as I say and Well, first what we need was a common reference frame, which in that case is a 3D rendering atlas. It was not available at the time We would also need a workflow for multimodal data integration So you'd have to be able to integrate two-dimensional histology data in 3D And we didn't know how to do this at the time and then could register that in common reference So well, you can't get it. You just have to do it yourself now what we found out actually was That while this time Jan Biali and his friends in Norway produced actually this excellent 3D rudder in atlas and they were actually kind enough to give us To make this available for us, but there were some problems. I mean because the data are proprietary We weren't quite clear what we can do with this So we decided to work on a tool which would be an overlay which would allow you If you purchase legally your own atlas or if you have data compatible with this To do this processing by yourself So we have this project which you call 3D bar 3D brain atlas Reconstructor which essentially allows you to go through a set of slides like this or compatible data to three dimensional structures This is an offline tool which you can which is again open source It's it's in python and you can easily install this and we also develop the web service which has Somewhat limited functionality, but essentially is a repository of all of the brain atlas in data And you can access it really so what we got in this exercise Well, we developed a workflow to generate 3D brain atlas is from 2D delineated data To generate 3D representations of structures and we have developed a format Which you call modestly common atlas format to store brain atlas is of this type And again, this is an offline tool and the web service There is actually a third option if you can afford it Which is to create a 3D atlas from scratch? And we are well aware of several recent efforts And we just said presentation from Allen brain Institute and we know the most from mouse and the most from rat I just would like to present a briefcase study from our own worst of opossum as they call it So what do you need what do you need if you want to Create a brain atlas from scratch and you have a choice between using Section-based techniques where the advantage is that they are very specific They give you access to find your anatomical information at the cellular level, but they have some problems Namely when you work on this you have high level of distortions and they don't they lack spatial consistency on the other hand You might use 3D imaging techniques which have some advantages for instance, you can use them in vivo and You get automatically three-dimensional data sets The disadvantages that usually the resolution is not so good and They are not very specific. You don't have very good contrast to make the delineations So ideally you would like to combine this these two approaches to You would like to integrate this kind of data To use their advantages and get rid of their disadvantages So this is what we did in a project of opossum brain atlas Why not mouse? I mean, why should we do mouse? I mean, how can you compete with our colleagues now? Why opossum? I mean opossum is is an animal which is gaining popularity and in particular in developmental neurobiology and the reason is that there is a short gestation of this animal on the order of two weeks and the newborns are born in early stage and we just happen to have colleagues in our institute who are Interested in developmental neurobiology of these animals and there is no brain atlas available So to construct this atlas we took amor scans was more term and we had micro micro tomography to get the skull Then the brain was extracted and we took the block face Images while cutting slices and Then half the slices were stained for nisla half were stained for myelin Then this data need to be integrated So we integrated the two-dimensional data To a common To a common space getting rid of the distortion as much as we could and then this was Corregistered to the to the image from the MR scan I'm not going to go into into details of the process. I don't have time for this Finally, all of these data were placed in this theoretical reference frame of the to get from my micro city So here you can see the result. I'm sure this is not very Responsive so again here we have the MR scan. This is the block face the nisla and the myelin spaces and Here is just a random cut So in the quality of the reconstruction, so this these three images have been reconstructed from from corner slices Finally all this all this data after integration in the common space Sorry They have been delineated so we have all together around 110 structures and All of this is available in in three different forms So you can just go to the 3d bar dot org and you can free bar s dot org and you can just Explore this data or you can go to the Scalabrain Atlas and you can play with them or you can just find Find them in bulk and download and play with your favorite software So again, what we got in this in this exercise Well, we got some some data which are made publicly available. This is the opposite brain Atlas And to be developed a workflow to integrate 2d images into 3d which we call possum We just again publicly available So again, I think that what we are showing is is not novel in terms of of the essence of what we show So what is what I think is different is our strategy that what we try to do is is is Make it available for the For the community both data and both software and they are available today not five years from now So what you can do when you have these tools? available Let me just give an example. So this is here we come to the monkeys and this is an example of of Experiment coming from Monash University from Alcellar Rossa's group With whom he started collaboration a few years ago Actually the collaboration started from a discussion at the INCF Domobuf at the fence meeting in Barcelona So anyway, if you want to do traditional neuroanatomy, what does it look like? So say say you want to get a connect home in this case you want to get the connect home of the Marmoset monkey So inject some tracers You look at the stain cells You get some cells stained with different different dyes different tracers and You count them you place them on I mean you mark them on your slice And what you end up with you have the distribution of these marks all over the slices slides slices So now if you want to have some connectivity information you want to Count them and you want to find how many of them are in different structures? So that's the typical end result. You've got a table. We just shows you how What kind of connectivity you have between different structures? So all is well But what happens? I mean we so Catherine asked this question to one of the previous speakers What if actually we decide tomorrow to change our percolation? We might have a reason to use different percolations and so what I mean You cannot move back from this table or from an image back to a new percolation So the table that I presented you is in fact a loss your way of reporting and storing neuroanatomical data So if we if say we have a new idea a new percolation new metals To subdivide part of the brain or if we want to use data from previous investigations to refine your hypothesis We need something something else So we need a more flexible way of storing and reporting the data a Way which would allow us reinterpretation in light of new information or hypothesis which would be preserved enough in 3d space To allow us making purely spatially based analysis and would allow for quantitative analysis not just saying connection present or absent So what we did in this case Let's see if it works We took the recently published Marmoset brain Atlas by George parts Paxino's martial arts and others and We use the 3d bar software To reconstruct the percolation in 3d That's again available on the 3d bar website And we took the nissan images and reconstructed the template which is nicely co-registered with this data So this gives us this the space the reference space in which we would like to put the tracing information Now so this is the task that you are faced with We have a number or much a lot has a number of of specimens a number of monkeys Which have been a which have been injected with these traces and then their brains were sliced With in the slices of 40 micron thickness when every thief slice was was used for for fluorescence imaging of the of the cells and every thief was you was Missile stained Then the locations of the cells were registered to the neighboring missile sections and now the task you have is just You would like to take this kind of data and placed it in this three-dimensional reference space So you need to have a transformation which goes from the left to the right So we use for this the the possum framework, which we developed while while working on the on the opossum brain atlas So here's your typical data set the input image stack You see it's maybe not perfectly aligned to the it maybe doesn't look perfectly like a brain So what you want to do first you want to well we start with some fine Reconstruction to get it closer to the To the reference template when you do the formable reconstruction to get rid of the non-correlated Deformations and finally so this is done with the possum framework and and then we use ans to co-register with the template So this is non-linear co-registration. So let's see it more dynamically. Hopefully. Okay, it works So that's your initial stack and here you have the fine transformation And this is the formal transformation This is something happened here. Now. This is being corrected with non-linear registration so you end up with a decent transformation of the original input input stack into the into the reference template and Now once you have once you have this transformation, they can actually translate all the So specific information into your template And this is something which is essentially independent of the parcelation So when you can use any parcelation you like and you can do with this data anything And you can do any analysis you like. So this is much more. This is a much richer data set And now of course you can represent it also in different in different ways So you can look at the 3d representation So again this this dot here is is the point of injection of tracer the black dots are the Stain cells and different colors show different areas. So this is taken from the Atlas which we processed So you can also use flat mapper representation if that's what you prefer You can also just Automatically now count the cells so you've got a count based map and This is just to show that I mean this is an ongoing project. So this was this I mean most of the stuff I'm showing here is the work of Peter Micah actually who is sitting here Who started working on this in Warsaw and now continues this project here in in Monash University So this is the the data which have been processed. So it's so far 27 injections out of out of 300 Okay, so that's essentially all I wanted to show you so I wanted to Show you just how we what you can do with a small but dedicated team of programmers and what I wanted to stress is that but What we feel is is needed in your informatics in neuroscience is essentially And if you want to really achieve scalability We have to make sure that we make both data and tools available when you look at the land landscape of today's new informatics There are many projects There are many outstanding projects which are very useful But I think they still don't reach the full potential because I mean sometimes You have just the data available and then there are tools which might be of use to the community Which are which are closed or the opposite or very often very commonly is the tool Which is so convoluted with the data that is just not very practical to use them separately so We essentially want to go the the Unix way and we try to provide small chunks of data Which could be hopefully of use to the community So so far it's been the 3d bar which is the general tool to integrate 2d drawings into 3d structures 3d bars is the repository of open-brain atlas is providing Some part of the functionality of this Possum the general tool to integrate 2d images into 3d and we have some some data Which is the opossum brain atlas and the marmoset critical I have also showed you some some Results of our projects on the possum brain atlas and the marmoset connectivity Which utilize some of these tools and I only hope that some of you will find We will find these tools useful and I'll be very happy to see some exciting projects in the in the future And thank you for your attention Thank you very much Two small questions the first Why didn't you do or are you doing intensity correction? I mean when you you're doing it intensity correction as well because The nissle stain looked a little bit heterogeneous with respect to Intensity correction when you look from most rostrel to most caudal sections You have shown there's alignment the stacking the alignment using non-linear, but but there was I Was not sure whether you have done intensity correction In the marmoset case Maybe I'm maybe I'm cheating whole section not Not locally So the so intensity was Made uniform but but again through the The way it has been unified was that there was single I-10 city transformation for the whole section not not local I-10 city corrections and Because of that reason that Myelin stuck was as you could see probably was quite sort of Like zebra like Yeah For the purpose, so maybe didn't play such an important role because you are still detecting your labeling presumably May have a second one. Okay, so I like very much also the It's a connectivity information that you map to the Marmoset, of course, this is very interesting Did you label in your atlas the position of the cells and Or did you also consider the tracks itself? So again, I mean so these data are not mine. They are much a lot of us I mean, I don't know how much I can say about this of the project. Okay, so some I mean especially with much a lot is not here so I Will guess that I didn't label all tracks with these techniques around this is my guess too But but I mean this would be really cool and exciting So if you want to comment, I mean so it will know more what we can say Those are very good questions. I mean just you know, so these were retrograde Tracers, so they labeled only this the neuron soma so the cells were are actually discrete point in space There's no tract or any any fiber stain with this with this matters Okay Very good Thanks again very much Daniel