Tamara Vanderwal “Gradients go to the movies: The topography and development of large-scale cortical organization during naturalistic viewing”

Transcript

• 00:05Good morning, everyone.
• 00:06So yes, we're going to look at what happens
• 00:09when the gradients go to the movies.
• 00:11So sometimes in my lab we like
• 00:14to think about. Movies as tasks,
• 00:16and we start to care a lot about trying
• 00:18to control what's happening in a movie.
• 00:21So this is footage from Inscapes where
• 00:23we very carefully constructed this
• 00:25state in which we were trying to avoid
• 00:28social processing, verbal processing,
• 00:30all these different things.
• 00:32That's because resting states
• 00:33actually a task, right,
• 00:35that some people can't do in this movie,
• 00:37which is called when Hyder
• 00:39met Simal Yuri Hesson.
• 00:41And I wanted to create this
• 00:43very social but very simple.
• 00:46Narrative.
• 00:46So it goes on and on and on
• 00:48for a little while.
• 00:49This third movie that you're about to see
• 00:51is the one that we're working on right now.
• 00:53It's called all day wrong.
• 00:56We've busted into live action with this one,
• 00:59and this is a symptom provocation task,
• 01:01so you can try to think of what
• 01:03might we be trying to get at.
• 01:05So this is a 10 minute OCD
• 01:07symptom provocation movie.
• 01:15And she grew and she grew.
• 01:17So oftentimes we talk about using
• 01:20movies to decrease noise, right?
• 01:22So that's things like head
• 01:24motion changes and arousal.
• 01:25We talk about trying to drive the signal.
• 01:27So if I'm interested in certain
• 01:28circuits or certain symptoms,
• 01:30I can try to get at that, right.
• 01:31We're very interested in the
• 01:33variability that is happening in the
• 01:34bold signal during movie watching.
• 01:36We think there's something
• 01:37quite special going on there.
• 01:39And then we talk about caring
• 01:41less about what the movies are
• 01:43actually doing and using movie
• 01:45watching itself as a brain state.
• 01:46So that's what I'm mostly
• 01:48going to talk about today.
• 01:50So this question came primarily
• 01:52from Daniels principal gradient,
• 01:54which we've seen a lot of.
• 01:56And the question was what would happen
• 01:59to the principal gradient if all of those
• 02:02regions were actually active and engaged,
• 02:04right?
• 02:04So if both poles of the gradient,
• 02:06both hetero modal cortex
• 02:08and primary sensory cortex,
• 02:10if those were all active and engaged,
• 02:12what would happen to the principal gradient?
• 02:15So my great doctoral student Ahmed Samara,
• 02:19who I think is on zoom, took on this project.
• 02:22He used the HCP data set,
• 02:24which you can see the structure of here.
• 02:26This is very much taking movie
• 02:28watching as a brain state, right.
• 02:30We have four different 15 minute movie
• 02:32runs and during each run there's like
• 02:34little snippets of different movies,
• 02:36so three minute clips,
• 02:385 minute clips,
• 02:39whole bunch of different stuff.
• 02:40So we're just kind of kitchen
• 02:42sinking the brain.
• 02:44And we were able.
• 02:45And I had fun with this because you
• 02:47can't usually do this when you're
• 02:48working with kids, but we could take.
• 02:52Both sets of data, both conditions,
• 02:55make them perfectly equal
• 02:56with regard to head motion.
• 02:57Normally I don't get to do that,
• 02:58so that was fun.
• 03:00And also perfectly controlled
• 03:01for the amount of data.
• 03:02So the number of volumes in each of these,
• 03:04right?
• 03:05So these are these pristine functional
• 03:07connectivity matrices for each condition,
• 03:09about 45 minutes,
• 03:1050 minutes of data in each of those.
• 03:14So then we just basically run
• 03:16our vanilla at this point,
• 03:18diffusion embedding dimensionality
• 03:20reduction to get our gradients.
• 03:23And what do we find?
• 03:24We find that across conditions,
• 03:26the components are basically explaining
• 03:29very similar amounts of variance.
• 03:31We recapitulate our classic
• 03:34resting state gradients,
• 03:36and what do we start to see
• 03:38when we get to movies?
• 03:39So some interesting differences,
• 03:41right?
• 03:42First of all,
• 03:42the sensory motor regions jump out the same,
• 03:45so that looks familiar,
• 03:47but there's some differences there.
• 03:49One difference that we're gonna see
• 03:51throughout these movie gradients
• 03:53is that the hetero modal pole
• 03:54is not just default network,
• 03:56it is equally occupied by both frontal,
• 03:58parietal and default regions.
• 04:00So we think that's pretty interesting.
• 04:03You also see on that that the
• 04:05visual regions have shifted to
• 04:07the middle of the gradient.
• 04:09Gradient 2 is what we kind of call
• 04:11this visual to non visual gradient.
• 04:13That's a little bit of an oversimplification,
• 04:16but you can kind of see what
• 04:18we're talking about big picture.
• 04:20And then our favorite gradient
• 04:22is this movie gradient 3.
• 04:24And there's a few reasons why we're
• 04:26jazzed about this one right now.
• 04:28But it's unique.
• 04:29So this one does not show up really
• 04:31in a recognizable form at all
• 04:33in the resting state gradients.
• 04:35And it combines some, you know,
• 04:38primary auditory regions, but all of your.
• 04:41Auditory language processing regions,
• 04:43some of which are a little bit
• 04:45more higher order and it's giving
• 04:46them all the same gradient score.
• 04:48So it is grouping those things all
• 04:50together and they're anchoring that gradient,
• 04:53right.
• 04:53So this is a unique and we call this
• 04:56auditory language movie gradient.
• 04:59We can take the lowest 10% of those
• 05:03scores and we can do some neuro synth
• 05:05mapping and you can see that these are
• 05:08very functionally segregated, right?
• 05:12So this is interesting.
• 05:13All of a sudden we don't have,
• 05:15we have hierarchical gradients still,
• 05:18right that that principle is the same,
• 05:20but we now have this different
• 05:22level of granularity.
• 05:24So as someone who wants to study development
• 05:26and different psychiatric populations.
• 05:29The question becomes,
• 05:30does this granularity get us anything right?
• 05:32So some of the studies in
• 05:34gradient work that have looked at,
• 05:36say,
• 05:36chaotic populations,
• 05:37generally what they're finding is
• 05:39a squashed principal gradient.
• 05:41So in depression you have a
• 05:43squashed principle gradient.
• 05:43In autism you have a
• 05:45squash principle gradient.
• 05:46If we have these three gradients to look at,
• 05:49do we start to see a little
• 05:51bit of differentiation or can
• 05:53we tell a different story?
• 05:54So thinking about development,
• 05:55we thought just based on what we
• 05:58know about cortical development,
• 06:00both functionally and structurally,
• 06:01that in kids probably those first two
• 06:04gradients would look pretty much the same.
• 06:06But we thought maybe our special
• 06:08movie gradient #3,
• 06:09the the auditory language gradient,
• 06:11might show some differences.
• 06:15So we will cut to the chase and show
• 06:17you this is now jumping data set.
• 06:19So this is in the healthy brain
• 06:21network biobank.
• 06:21These kids watched 10 minutes
• 06:23of Despicable Me and about 3
• 06:261/2 minutes of the present.
• 06:27And what you can see here and I'm
• 06:29not showing you but will we will
• 06:31eventually that the first two
• 06:32gradients look pretty much the
• 06:34same across kids and adolescents.
• 06:36But when you get to this
• 06:38auditory language gradient,
• 06:38you start to see these very
• 06:41interesting differences.
• 06:42And so Ahmad's point about this.
• 06:44Is that when you're in the kids?
• 06:46This is very much.
• 06:48And auditory gradient and as you
• 06:50progress through development though
• 06:52these are cross-sectional data,
• 06:54you get to see the the different
• 06:56higher order regions and these
• 06:58different language processing
• 06:59regions are now part of the gradient.
• 07:02They weren't when you were a little kid.
• 07:06So we're going to do one more thing.
• 07:08This is a garbage can that I pass
• 07:10on my way to work every morning.
• 07:12My skull is a cage,
• 07:13and I yearn to wander. So good.
• 07:19So we are going to go out of the
• 07:22skull and into gradient space.
• 07:24So we can look,
• 07:25this is back in the adult data now
• 07:27and we can look at the transitions
• 07:29within this with ingredient
• 07:31space between these two states.
• 07:32And So what we thought we were seeing was
• 07:35that some regions were moving far right
• 07:38from rest to movie and some were not.
• 07:41So we just computed the distance
• 07:43that each region was moving
• 07:44within this space and then we can
• 07:47map that back onto the cortex.
• 07:49And what you see is this like very
• 07:52clear clustering around the STS.
• 07:53But those are the regions that within
• 07:56gradient space are traversing a really
• 07:58far distance when you do these stage shifts.
• 08:01So kind of interesting.
• 08:02So this work right now is on bio archive.
• 08:05We just submitted the revisions,
• 08:07you can check that out if you want.
• 08:09We also looked at the reliability of it
• 08:11and did some brain behavior stuff too.
• 08:13So I will stop by saying thank you
• 08:15to my lab and to everybody else.
• 08:18And then this.
• 08:19Did I do this?
• 08:20Yeah, this one's for Doctor Brakey,
• 08:22because I don't feel like we got
• 08:23enough press for our beautiful cover.
• 08:25So I'm just going to leave that up now,
• 08:27and I'm happy to answer any questions.
• 08:38I was wondering. Don't get very focused.
• 08:46Like familiarity with the movie
• 08:47that they're watching the scanner.
• 08:50Yeah. So I think novelty is something that
• 08:53we haven't systematically looked at yet.
• 08:55I would point out that in tasks,
• 08:57there are massive novelty effects
• 08:59that we don't ever talk about.
• 09:01Resting state. Are there novelty effects?
• 09:03We don't seem to care, right?
• 09:04But I do think we should look at it.
• 09:06My my gut would be that in kids,
• 09:10the novelty effect would
• 09:11actually be very small, right?
• 09:13So kids have an amazing
• 09:15appetite for repetition.
• 09:16They want to see it again, see it again,
• 09:18see it again, and they just don't care.
• 09:19They're in. They're as interested.
• 09:21The 70th time as they were the first time.
• 09:23So I think we'd actually have
• 09:25a smaller novelty effect.
• 09:26But yeah, we should look at it.
• 09:29Do you think that?
• 09:31Find the data set where you're.
• 09:36The same.
• 09:40Yeah, so it's a really neat question.
• 09:42We're just starting to collaborate
• 09:45with Garov Patel's lab in Columbia.
• 09:47They have an amazing data set where
• 09:50individuals with schizophrenia
• 09:52watched listen to stories.
• 09:54So they have an auditory
• 09:56only naturalistic condition,
• 09:58and then they watch a movie,
• 09:59but without the auditory stuff.
• 10:02So that's super weird, right?
• 10:04So you're watching everything
• 10:04and you see the mouse moving,
• 10:06but you're not hearing the soundtrack.
• 10:08So they have these three.
• 10:09You know, different.
• 10:10So they're starting to
• 10:10look at and so we have,
• 10:12we want to look at the gradients
• 10:13and those and see what happens.
• 10:16Curious so.
• 10:19I noticed how.
• 10:22Things out all the time.
• 10:24When you got a movie to watch
• 10:26the actual content in the sound.
• 10:27I don't know how much of a difference
• 10:29that you see here about the fact that
• 10:30you're getting kind of bombarded
• 10:31by effectively noise during the
• 10:33rest of the study during the task
• 10:35that that construction information.
• 10:36How do you even go about finding
• 10:38people that it's really.
• 10:41Yeah, it's a really big deal, right.
• 10:43So music is a massive part of our brains
• 10:45and our functioning and our hearing things,
• 10:47and in a movie it's a really big deal.
• 10:50So a lot of people who have used inscapes.
• 10:52Do it without sound, which to me like
• 10:54that's a whole different paradigm.
• 10:56Yeah, you could.
• 10:57You could definitely look at that and I
• 11:00think there'd be significant effects. Ohh.
• 11:04And this is the best.
• 11:09One of the best, probably the best.
• 11:13Amazing job. It only took us what? Two years.
• 11:19If you go back to the slide,
• 11:21I mean it looks to me like that.
• 11:23Yeah, one more. Yeah, this looks
• 11:24like it's been 3 dimensional space.
• 11:26It's just. Yeah, really. Have you?
• 11:35Yeah, so you can look at them.
• 11:38Right, so we started looking at
• 11:40them just in like flat space.
• 11:41So this is gradient 1 by gradient 2.
• 11:44I get all geeked out about this
• 11:45because I think this starts to
• 11:47look like a more perfect gradient.
• 11:48So Daniel and I like to argue about
• 11:51which is the most perfect gradient.
• 11:53Definitely look at that, right.
• 11:55We did like all these measurements
• 11:57like distance to nearest
• 11:58neighbor really equal in movies.
• 12:00So step wise distance between all
• 12:01of those points is very, very,
• 12:03very similar in movie distance to centroid,
• 12:06we did all those sorts of stuff.
• 12:08Yeah, I think it's really interesting.
• 12:11It's like there's this apex triangle, right?
• 12:14So you have these three
• 12:15hierarchical gradients that go up.
• 12:17So we sort of think of it like
• 12:19the three movie gradients are
• 12:21represented in the principal gradient.
• 12:23But it's just. Collapsed almost.
• 12:27And so if you rotate it,
• 12:28then you can see all three of them.
• 12:29But it is it.
• 12:30Is it the same information or not?
• 12:33I don't know.
• 12:34We're still trying to figure it out.
• 12:362nd.