Module 12.2: Kinds of Designs Part 2

Loading...

Do curso por Johns Hopkins University

Principles of fMRI 1

389 classificações

Explore Related Videos

At Coursera, you will find the best lectures in the world. Here are some of our personalized recommendations for you

Johns Hopkins University

Principles of fMRI 1

389 classificações

This course covers the design, acquisition, and analysis of Functional Magnetic Resonance Imaging (fMRI) data. A book related to the class can be found here: https://leanpub.com/principlesoffmri

Na lição

Week 2

This week we will discuss the fMRI signal, experimental design and pre-processing.

Module 11: Experimental Design13:58

Module 12.1: Kinds of Designs Part 111:42

Module 12.2: Kinds of Designs Part 2 8:12

Conheça os instrutores

Martin Lindquist, PhD, MSc
Professor, Biostatistics
Bloomberg School of Public Health | Johns Hopkins University
Tor Wager
PhD
Department of Psychology and Neuroscience, The Institute of Cognitive Science | University of Colorado at Boulder

Now let's look at a multiple subtraction design.

So the idea here is to triangulate on the association

between event type A, some event type, and

a particular process by subtracting away multiple kinds of events.

And this helps us avoid some of the issues with pure insertion if

the different control conditions I'm subtracting off each have different kinds

of characteristics, and each of them is potentially

confusable with the process of interest in different ways.

So, a case study here is studies of the Fusiform Face Area, by Nancy Kanwisher et

al, and they did some really nice studies, and her's a slide from one of them,

in which faces are compared with many different kinds of things.

So they can pair faces with objects.

But of course, those differ in their different spatial frequencies.

So they can pair intact faces with scrambled faces,

where the basic spatial frequencies are preserved.

Faces with houses, faces with hands.

So maybe it's about body parts.

They can control for body parts.

Faces with animal faces,

faces with pictures of faces, drawings of faces, and other conditions.

So really across these multiple subtractions,

they are trying to triangulate on the idea that the FFA is activated when you see

a face, but not when you do other things.

Let's look now at a process overlap or dissociation design.

And one way of thinking about this is to think in terms of double dissociation,

which goes back to neuropsychology literature from way back.

And the idea is that task A activates more than task B in one area,

and I can find another area of the brain in which B activates more than A.

And this implies separate processes in A and B.

So this isn't knock-down-drag-out evidence for separable processes because you can

get nonlinear effects that can sort of cause that doubled association.

But it's pretty good evidence on process separability.

A stronger argument is called separate modifiability.

So let's say we have task A activates one area but

B does not, and B activates another area but not A.

So now we have the doubled association, but

with the added constraint that there's really no response

to the thing that doesn't activate the area, it activates it less.

And this implies separable processes in A and B.

So let's look at an example of this from a recent study, from our lab for

convenience.

And the issue here was, do physical pain and

romantic rejection share common brain representation?

Especially in the dorsal anterior cingulate,

which is the area we're going to focus now.

So what we're going to do is compare patterns of activity,

this is multivariate, that are predictive of heat pain,

versus patterns that are predictive of viewing pictures of the ex-partner,

relative to their respective control conditions.

So what we see here, is that we can identify a pattern within the anterior

cingulate of activity where the response in that pattern, overall,

responds to physical pain more than to warmth, its control condition.

But there's no additional response to the ex-partner,

which is the rejection condition, versus the friend.

So, that's one half of a doubled association or

separate modifiability process.

Now, let's see, we can identify another pattern which is tracking rejection.

In this pattern, responds significantly to the rejection event, ex-partner, but

there's no response to heat, to painful heat.

So what does that mean?

That means that these two conditions and

these two patterns are separately modifiable.

So the representations of pain and rejection, at least by these measures,

in the dorsal anterior cingulate, are not shared.

Let's now look at factorial designs,

which is another basic way of enhancing our inferential power.

So in a factorial design, I manipulate two factors at once.

And I can test for dissociations in the areas that are activated.

And I can also test for interactions between the two factors.

So here's an example.

It's a task switching experiment where people view these complex objects with

shapes and orientations on the screen, and the upshot is

that there are four different types of attention shifts in this experiment.

And they're grouped into two factors.

So in factor 1, people are switching between an attribute of the object,

let's say whether an object is sort of vertically or horizontally aligned.

And when it's held in memory.

They have to remember and read off memory.

In factor 2, it's what we'll call an external switch,

where the objects are available on screen and

they have to shift the focus of their attention from one to the other.

In this study, people could do either type of switch or both simultaneously.

So the four trial types there are cover the two by two factorial space.

And the maps at the bottom just show you some of the activations in internal task

switching in the frontal cortex to external task switching in the frontal

cortex and the posterior areas, and then there are also some interactions.

Interactions are where something special happens when you have to switch

both things at once, something on the screen and something internal.

And that happened in the insula.

So this is just one example among many of how we can employ

factorial design approaches in our experiments in brain imaging.

A final example is a parametric modulation design.

And this is one in which I can manipulate variables in a parametric fashion

within-person, usually with three or more levels.

We can also just use measured variables like performance, or emotion ratings, or

other things like that.

And the nice thing about this is that it can also help with the specificity of our

inferences.

It can provide stronger evidence than basic contrast that there's

a relationship between brain activity in an area or voxel, and

the process that you're studying, that you've manipulated parametrically.

So here are two of my favorite examples from early research.

On the left, what you see is blood flow on the y-axis increasing as

complexity in a Tower of London task,

which is a task that involves reasoning and executive function, increases.

And we're looking at the dorsolateral prefrontal cortex.

And what you can see here Is that there's a graded parametric increase

in the blood flow as the task complexity increases, so

that increases our confidence that the DLPSC activity is related to performance,

or task performance in some way.

On the right, we're looking at the ventromedial prefrontal cortex, and

here there are five levels that involve different levels,

all different levels of the goal or purchase value of an item.

So people were asked to, say, how much do you want to buy this hat?

How much do you want to buy this pencil?

Or this mug?

And when people said, no I don't value that very much, the activity is low.

And the more they value the item, the more there is a parametric increase in

the level of brain activity, this time in the ventromedial prefrontal cortex.

So there are many examples, but

these are some of my favorites of parametric modulation designs.

And we'll look at examples of how to do all of these designs statistically

when we talk about the general linear model.

So let's wrap up.

We've talked about several kinds of designs, subtraction designs,

basic contrast, individual differences designs, correlations between brain and

behavior across individuals.

We've talked about process overlap and

dissociation designs which involve multiple comparisons, and

in some cases the logic of doubled association or separate modifiability.

We've talked about factorial designs and we've talked about parametric modulation.

So that wraps up this module.

And thanks for your attention.

See you next time.

Explore nosso catálogo

Registre-se gratuitamente e obtenha recomendações, atualizações e ofertas personalizadas.

O Coursera proporciona acesso universal à melhor educação do mundo fazendo parcerias com as melhores universidades e organizações para oferecer cursos on-line.

© 2019 Coursera Inc. Todos os direitos reservados.

Baixar na App Store

Baixar no Google Play