After screening our participant to make that she didn't have any implanted metal

in her body and therefore would be unsafe for the scanner environment.

John, the MR technologist, is laying the subject down on the scanner bed and

he's going to move the bed up and move her back into the bore.

Over her head is a gradient coil, and on the coil is mounted a set of mirrors

which will allow her to see the images that I present from the scanner room.

Once she is on the bed and moved back into the magnet, she'll stay still.

That will give us a chance to localize where her brain is in her head, and

then we can take a structural image of it.

And then we will be able to present our stimuli and conduct functional imaging.

So the first thing that John's doing right now is sometimes

called a scout scan, or even a localizer scan, and

that's just giving him an idea of where the brain is in that bore.

So obviously a subject comes in.

They're in there.

We have to localize first what we're looking at.

And then what he's going to do at this point in time is he's going to create

a field of view, which we're then going to be able to use so

we can get a finer-grain look at the brain.

>> Okay, Erin, it's going to be about four minutes.

>> So this first scan is what we would call a high-resolution structural image.

And what the high-resolution structural image actually is, is in the scanner

environment we have a local high magnetic field, in this case, three Tesla.

And then we have a second more transient magnetic field that oscillates.

And what this does is the local high magnetic field at three Tesla allows

all the hydrogen atoms in the brain, and your body's made mostly of water so

there's a lot of hydrogen available, to line up in one direction.

By having a secondary magnetic field that oscillates,

that comes on and off, you can excite these hydrogen atoms in different ways.

So you can make them actually change the direction that they're oriented in.

And when they change, what happens is, is we can recover how fast they

settle back down into equilibrium, so back to that state that they're at at the local

high-intensity magnetic field in three Tesla.

And depending on the tissue type,

that will dictate the speed at which they recover to equilibrium.

And that's what actually leads to these really wonderful images you see when

you have an MRI taken of your brain or your lung or any other body part.

It allows you to noninvasilvely view tissue.

And so, in a medical setting if you're having this done,

it's useful if you want to identify tissues that don't belong.

For instance, tumors or

other sorts of growths on something like your brain would be able to be

picked up by different sort of varying contrasts that MR technologists can do.

Now what we're using it for is we want a really nice picture of the brain.

And the reason why we want a nice picture of the brain is when we're doing fMRI,

which is functional magnetic resonance imaging,

we're really interested in function.

So we're interested in function but

we also are interested in localizing function.

So we want to have a nice picture of the brain so we can identify particular brain

areas and regions of interest that we might speculate or

hypothesize are involved in certain cognitive or sensory processes.

So the first thing that we're going to do is we're going to find out

where everything is so that we can then hopefully,

when we do the functional scan, line up function to structure.

And so each scan in an fMRI experiment will always

start with this general high-resolution task.

And therefore, we can then go and do a lot with this sort of image.

The sequence that is actually being conducted right now to take this

structural image, the special type of sequence that was developed was

developed here at the University of Virginia.

And what it actually did was it allowed for better resolution at much

faster times, which is a real big plus in both the medical and research communities.

Now we actually have a high-resolution view of our participant's brain.

And what we can do now is we can find the areas that we're interested in

covering to look for functional activity, so

activity in the brain when we do our functional task in this case.

And with modern imaging, we actually now can cover the whole brain.

About a decade or more ago we had to pick certain parts of the brain we could look

at because that was all the scanner was capable of,

that's all the software was capable of.

But now we can actually in a general scan we can cover most of the brain.

I'm not sure if it's viewable in the camera, but

you might be able to see yellow lines that are coming across all the brains.

These are called slices.

And what they are is when we take a, what we call a volume which

encompasses a number of slices and we can then cover the whole brain and we can look

at each one of these slices individually to view the different parts of the brain.

So what John has done here is he's just prescribed where we're looking,

so how we're going to go about creating slices,

how we're going to create measurements while the subject is doing a task.

And that's what we're going to start here in a minute.

So what I'm going to do now is, through a little intercom she can hear me in here,

I'm going to talk to the participant and

tell her we're about to start doing functional runs.

What that means to the participant is I've already given her instructions before she

went into the magnet about particular tasks that we're collecting today.

I'm just going to remind her quickly of the instructions, and

then I'm going to tell her that the scan is going to start in a second.

She will see what I see on my computer out here on my experimenter computer,

she will see that through her glasses, or

through the reflective mirrors that I talked about before.

And I will be controlling what's on the projector from right here.

Okay, Erin, I'm going to start your functional task, okay?

>> Okay.

>> So again, in this task, what you're going to do is you're going to see

a series of cues that I've talked to you about before.

In one case,

the cues are going to indicate that you could receive a positive reward.

And in another case,

the cues are going to indicate that you might receive a negative reward.

Your task is to take the button box and

to press the button as soon as the prompt comes up.

The prompt is the solid white square.

If you press the prompt fast enough, then you might receive a reward or

you might avoid getting the punishment, okay?

>> Got it. >> Okay.

So through my experimental presentation software now, I'm going to

Start the stimulus.

And she actually has some instruction screens here that she can press through

with her thumb on the blue button.

And she could read the instructions again just to, again,

be sure that she understands the task and kind of

gives her a refresher of what we talked about before she went in the magnet.

When she's finished reading the instructions I'm going to tell John, and

John's going to start the functional scan.

And when the functional scan begins,

it's going to actually tell the stimulus computer to start.

So it's going to tell the program, it's going to tell my experiment to start.

Okay, we can start, Joe.

So, at the beginning of every scan, we actually have to have some time for

the magnetic field, itself, to stabilize.

So, the software that actually is in control of the magnet

is in control of the sequence, is smart about this.

And it doesn't let itself start it until it's ready.

Generally around five or six seconds.

And once it's ready, and once it's ready to start creating this oscillating

magnetic environment, in this case.

What we do is it sends a signal to the stimulus computer here,

which tells the computer to start the experiment.

And that's what we've done here.

Now the experiment is started.

In this experiment that our subject actually is looking at today, we're

looking at different areas of the brain that are involved in reward processing.

And what we're actually doing is we're trying to model how one

anticipates either receiving a reward or receiving a punishment in this way.

This one is a special version of it in which we're actually interested in

social perceptual and social rewards.

So we're actually using smiley faces as a reward and

frowning faces as punishment in this case.

Essentially, what our participant is doing is

she's given some information about the types of trials that are going to happen.

So these are trials that are going to indicate that you might receive a reward

or you might receive a punishment, and then she's going to do the task.

And the task is a speeded response test,

where she simply sees a cue on the screen and she responds as fast as possible.

If she responds fast enough, then she will either possibly be rewarded or

possible be punished in this case.

So this type of task is very common in a lot of different fields

of cognitive neuroscience, psychology fields and things like that.

And decision making fields and

really it's used often times to localize areas of the brain

that we think are involved in rewarding participation or reward processing.

One thing that you'll notice also when we go through the scan here.

The functional magnetic resonance imaging procedures that we're doing right now

are not going to immediately appear on the technologists computer.

So the images that you often see associated in publications or

in newspaper articles or things like that that shows blobs on a brain or

little colored spots across the brain are not things that

are actually going to be reconstructed and put up on the screen right now.

That generally is the result of a statistical analysis.

They can talk more about that if you guys want me to at a different point.

But what that actually does is, it's post-processing.

So we take the images, we go back, we then look in the images and

we look at statistical probabilities and we infer function from that.

But it's actually a labor intensive process that actually doesn't get done,

and it's not sort of a plug and play, you press the button and go and

we get the data and we get out of here.

So you never actually get functional data coming right from this type of scan.

That being said, there are some ways of getting some

real time functional approaches that people have tried in clinical settings.

And there's some success with it, but it's much more experimental,

and much more limited in what you could actually get out of it.

I should also say a word while this is going on,

the functional part, about what exactly we're collecting in this case.

So I keep saying we're collecting functional data.

So what does that mean.

Well before we took a three dimensional image of the brain.

So we took three dimensional image based on the different composition of

different tissues in the brain.

And In fMRI, the f stands for functional.

And what we mean by function is is what we're actually interested in as

neuroscientists oftentimes is neuronal function.

So, we're interested in if I'm going to deliver a reward for instance,

what is happening where in the brain when someone is anticipating this reward.

Well, we can't easily use MRI to understand how neurons are firing,

so we can't easily track neuronal firing with MRI.

But one thing that we can use, and one thing that we can monitor and

track in an EMR environment, is Bloodflow.

And the reason for this is that oxygenated and

deoxygenated hemoglobin, so blood this either is carrying oxygen or

blood that has been deprived of oxygen, has a different magnetic property.

Specifically, deoxygenated hemoglobin is much more magnetic than

oxygenated hemoglobin.

So, functional magnetic resonance imaging uses something called BOLD,

which is blood oxygenation level dependent contrast.

So what we're actually looking at with this acronym here,

this bold acronym, is blood flow.

It's the delivery of fresh blood to

an area of the brain that is requiring this blood.

And what kind of areas of the brain would require blood at particular times.

Those that are active.

Those that have just undergone a high degree of activity.

So neuronal firing causes energy.

We need to replenish those ion pumps that allow for the neuronal firing.

What we have here is we have blood that carries resource to those areas, and

therefore this blood delivery is a secondary measure of the neurons firing.

So, we're interested in neuronal activity.

We don't have an easy way of understanding or

measuring neuronal activity in the MRI environment.

But we do have this particular property of hemoglobin

that we can capitalize on, and we can track that over time in a scan.

So that's a fourth dimension.

So instead of having a three dimensional scan of simply the structure of the brain.

We can have this fourth dimension, which is change over time.

Change in the oxygenation level of the blood over time in various parts of

the brain, of that whole area that we covered when he told where to scan from.

One caveat of this is that neurons fire very quickly,

in the order of milliseconds, and neuronal firing is something that.

Here, well, the scans finished so we'll let me just talk to the subject here.

And then, I can get back to that point.

Okay, Erin, we'll come and get you in just a second, okay?

>> Okay.

>> So neurons fire very quickly.

The hemodynamic response, which is what the blood flow is.

We call that the hemodynamic response, is actually quite slow.

So, the hematinic response can peak anywhere,

from four to six seconds after neurons fire.

So, FMRI is said to have a really high level of spacial resolution,

in that we can localize functions in particular areas, and

we can see really clearly to different parts of the brain and

in the different properties of hemodynamic response in different parts of the brain.

But it's said to have a poor temporal resolution in that we can't understand how

processes change on a real fine scale over time based off of neurons firing.

So we'll have John go and get the subject out of there now.

When you see and print the results of a functional magnetic resonance imaging

study, what you often see is what some people refer to as brain blobs.

I'm not particularly fond of the term, but what you actually are looking

at is you're looking at these really nice structural images that we've

seen before that we were able to take of our participant's brain.

And then you see these little colorations on there.

And these colorations are there to indicate activity in the brain.

So what I think most people would see, maybe from a less informed or a naive

perspective in this case, would be okay this must indicate activity in the brain.

This must indicate that this area of the brain is active for this task.

So this is where we actually can localize the functions that we're interested in.

So that is true.

But when you take that further, okay, well what does the color actually represent?

What does that actually mean?

Really what it is it's a statistical map.

So we took the images that we had and

we took the tracking of the hemodynamic response that we did.

And we did it at each voxel, which is what you think of as a pixel,

across all of the brain space that we analyzed.

And what we did is we said, okay if this area is going to be active

then what we should see is when a particular stimulus was on the screen,

we should see that that stimulus caused a change

in the hemodynamic response in that particular voxel.

And so what we can do is we can say okay,

show us statistically which voxels show the greatest relationship or

coupling between an event happening that we caused, that the experimenter had in

their experiment, and the hemodynamic response changing in that voxel.

Now, we can then say this is going to be a lot of noise.

There's going to to be a lot of issues.

There's going to be a lot of false positives.

There's going to be a lot of things that look like an area was active, but

it was just a fluke, it just was a change in that area, just given that time.

So what we want to do is we want to use sort of modern statistical approaches

to give us an idea through all that noise, where is the signal?

Where are these areas that reliably show a response to a particular type of stimulus

across this whole brain space?

And then we can conduct a statistical test which will then

only identify those voxels and then we can plot those.

Well, here we have plotted them three dimensionally

on different areas of the brain.

So you can see these little red spots on this brain, areas here and

little red spots here on this dorsal anterior cingulate cortex area.

These would be the results of our image processing that we did

after we collected the data.

So basically, what you're looking at when you see these brain blobs is actually

the result of statistical mapping across three dimensions.

So it's actually just a spacial statistic that's showing a low probability that it

was by chance that we have a coupling here between human dynamic response changes and

the stimulation that we provided which in this case was a visual task.

Now when you see studies where they've collected 15 participants and

then they've made some inference that the 15 people's'

brains are all working the same in this case.

What they've actually done, is they've taken the statistical maps for

each one of their subjects, so each one of their participants.

And they've taken the statistical maps that have shown this coupling for

each participant between human endemic response and stimulation.

And they've had every single subject undergo the same exact procedure.

And then they've averaged them together.

So by taking group averaging methods statistically, you could take

the individual maps, average them together and find the areas of commonality.

The areas of commonality will then give you one large group map.

And the inference, then, for a particular cognitive or perceptual task,

is that these are the areas of the brain that are involved

in the task that you were engaged in.

So, in any particular sensory cognitive type of task.

And that's how fMRI really comes to be in that case.

So again you're actually going from a secondary measure of

neuronal function, which is humanistic response.

Then you're applying the statistical methodology to that.

To track what areas are showing a pattern of response that we would expect

given stimulation we provided.

The statistical methodology overwhelmingly that's used to analyze

fMRI data is just a general newer model approach.

It's a multiple regression approach.

Which makes an assumption going in

about what the hemodynamic response function should look like.

So we make an assumption that we know what

the hemodynamic response will look like in certain areas.

Which can be problematic in most cases, but it's actually quite robust in others.

We model before we go in, so we say here's what we are looking at.

We have our hypotheses, based on when we had things on the screen and

based on that being tagged to when the data were collected.

That our events, our stimulation would cause a change

in brain areas that were involved in processing what we were showing.

And knowing that, we can go through and say, okay, places that are involved in

this task should show a change at all of these relevant times.

Or should show somewhat of a change during these relevant times.

We're making an assumption that we can then say this is what the change

should look like.

So we're going to model that,

and then we're going to take our model and we're going to put it through.

And we're going to say which areas statistically adhere to our model.

Probably, in a large majority papers that you've come across will have

used those techniques for understanding or analyzing the brain activity.

There's newer model free approaches.

Which, are there to allow us to make less assumptions going in and actually then

to take the data and to find consistencies across space and across time.

And then tell us what areas are of interest versus

that we can then go back and compare to what we did.

So these are often called model free approaches for that reason or

data driven approaches.

And there's a lot of different statistical procedures that allow for

these types of approaches to get around this idea of making assumptions of

what's actually going on before really letting the data speak for themselves.

"Touring an fMRI Lab," Jamie Morris

藏传佛教禅修与现代世界：小乘佛教

与讲师见面