Basic Visualization 2

Loading...

Do curso por Johns Hopkins University

Introduction to Neurohacking In R

137 classificações

Johns Hopkins University

Introduction to Neurohacking In R

137 classificações

Neurohacking describes how to use the R programming language (https://cran.r-project.org/) and its associated package to perform manipulation, processing, and analysis of neuroimaging data. We focus on publicly-available structural magnetic resonance imaging (MRI). We discuss concepts such as inhomogeneity correction, image registration, and image visualization. By the end of this course, you will be able to: Read/write images of the brain in the NIfTI (Neuroimaging Informatics Technology Initiative) format Visualize and explore these images Perform inhomogeneity correction, brain extraction, and image registration (within a subject and to a template).

Na lição

Neuroimaging: Formats and Visualization

In this section, we will discuss different formats that brain images come in, as well as some of the commonly done magnetic resonance imaging (MRI) scans.

Basic Visualization 19:00

Basic Visualization 29:35

Conheça os instrutores

Dr. Elizabeth Sweeney
Rice Academy Postdoctoral Fellow
Biostatistics
Ciprian M. Crainiceanu
Professor
Biostatstics
John Muschelli III
Assistant Scientist
Biostatistics

So, how do we visualize some of the actual data rate?

It's nice to plot a brain, but what do we actually see in the image?

What is the distributions of the intensities of that entire image.

So, at the top we just do some very basic things that we want to say.

I want to do two plots, and I want to have two columns and one row,

so that's what the par mf row means.

The o, we pretty much set it so that the margins of the plot are zero.

These are actually standard they're not very specific in our imaging.

But we wanted to show you how you can actually create the plot below exactly.

So, what we first do is we say histogram on the NIfTI object.

So we do a histogram of all the intensities in the image.

We color the histogram and we change some of the y and x labels.

But by default, what we're really trying to do is say,

what is the distribution of the intensities look like?

And what you'll note is that they're very very large spike at zero.

And that's because a lot of the image is just the background.

So if we jump back we see a lot of area outside of the brain is actually zero.

So, when an image is acquired, they don't just zoom specifically on your brain.

They can't just zoom exactly in that area.

They actually take what's called a field of view, so they take your brain and

some leeway so that they make sure they don't cut anything off.

And a lot of that area is just background, which is set to zero, and so

that's why you see this big spike.

Now, in the second plot, we take intensities greater than 20.

So we take the NIfTI object, we subset intensities that are greater than 20,

20 is an arbitrary choice.

We just chose something rather small to just lop off the bottom

area of the histogram, so that we can visualize the majority of the image.

So we do the histogram of the intensities over 20.

And again these are in arbitrary units,

that's why it was rather arbitrary to take 20.

So we see a bit different distribution, and so we see a spike around maybe 450.

We see some long tail on the right.

But the majority of the intensities go from about 20, or 0 to 1000, and

then we see an even longer tail out to the right.

We see that the histogram is plotted up to 1500, and actually further than that.

So there are some voxels in the brain that have very,

very high intensity compared to the rest of the brain.

And some of these could be potentially artifacts.

Some of these could be just hyperintensities in the brain.

But this is actually good to look at because this will actually change

the way things are plotted.

So when we go back and we look at the brain here.

So the brightness of the voxels are mapped to the grey scale pallet.

And that's mapped in a certain way so that if things are really, really bright,

if you have a super bright voxal in your image.

So if you have a super bright area, then this entire image,

the majority of it will actually look a lot dimmer.

So sometimes it's very good to maybe lop of some of the top voxels and

set them to some value.

So for example, if we set everything above 1,000 to 1,000, that can actually

make the picture look brighter, but get the same level of information.

Maybe not in the high intensity areas, but maybe some of those aren't so of interest.

We call the back mapping, so what we do is we derive some sort of statistic or

we derive some sort of information from the image.

And then we may want to plot something overlaid on the image,

that might no actually be that clear.

So what we do here is we create a vector, a logical operator.

We say, is this object, is the nifty object for

all the voxels greater than 300 and less than 400.

So we just want to say, give me a logical yes, true or false,

whether a specific voxel is between 301 and 499 inclusive.

And I say that because they're pretty much mapped to integers.

So it's not like there's 300.5, at least the way the data is originally acquired.

So we take 301 to 399 and we say these voxels, give me those.

Give me true for that, everything else is false.

So that is what this is_btw_300_400.

So we said, give me something that says true if you're between these values,

false if you're not.

Then we copy the NIfTI nii_T1 object, it's a NIfTI object in R.

So we just copy that over and we call that nii_T1_mask.

And then we do one operation, we say nii_T1_mask and we subset, and

we say, not the vector we just made or the logical operators.

And then we assign those NA.

So what is this entire operation doing?

So we're saying, we want an image where if values are outside,

so our 300 and lower, or 400 or higher.

We want to actually set those to be NAs.

We want to throw them out and call them not applicable.

Okay, so what do we actually see in this image, so we see

the background image plotted in gray and we actually see the overlay in red.

What we're actually seeing in the image if you aren't necessarily acquainted with

structures in the brain yet.

Which we don't assume you are, this

big red area in the middle corresponds to the white matter of the brain.

So, where the axons are in the brain and we see some areas of the skull.

So this is saying that, areas of the brain between 301 and

399 are some areas of skull and a lot of the white matter.

So we can use that overlay function again but if we don't it tell

it which slice of the plot, and we don't want to do a single plot.

We don't actually plot, again every single slice of the brain,

going from the bottom all the way to the top.

So, I didn't necessarily specify that before in the image function,

but if you take the top left corner, that's the bottom of the brain.

And it starts going up as you go right.

So, left to right, you're actually going up the brain,

and then it goes to the row below it.

And it's continuing going up the brain in that kind of operation,

like you would be reading a book.

So left to right, and then you drop a row down and

that's actually going up the brain.

And we can see that again, large areas in the brain that are in red

are the white matter regions and areas of the skull.

So this just gives you again a little bit more comprehensive view

over all the brain rather than just one slice.

But like I said before it's very useful but

at times it can be actually a little bit overwhelming.

So we can do the same operation like we were saying before by doing it by slice,

doing it with the orthographic view.

So again, we pass in orthographic, we pass in the entire NIfTI object and

the the mask.

And then we put a little bit more information in here.

So we kept it at the same xyz, so the cross hair should be at the same

point that it was before, and we actually added a bit of text here.

So this is one of the reasons I'm a big fan of this orthographic view.

Because it tries to give you a relatively three dimensional per se view of

the brain.

Because it gives you all three planes at the same time, and

it allows you to write some text in there rather easily.

You can do that with the image function.

You would just say text, and you would label things so

that you know what you're actually viewing.

But the image function is not very good,

especially when you have all the slices of the brain presented.

If you have a specific slice, it's very good.

You can label it, it's very nice.

But if you do all them, it's not so easy.

But with orthographic view, is in the right hand corner.

You can put a lot of information in there.

You can put legends, and so forth.

So, it allows you to plot all three slices and

give you additional information in that bottom right corner.

So, to summarize, what functions have we gone over?

Again, the workhorse function that we will be touching on many, many,

many times is the readNIfTI function that will be able

to take in .ini or .niigzfiles from your computer, from the hard disk.

And these are NIfTI files on your computer and read them into R, so

that they're R objects.

R NIfTI objects and the image function,

the orthographic function and the overlay function all from the oro.NIfTI package.

Operate on these NIfTI objects to actually create what I would say is

relatively standard visualizations.

You can make them more complex as you go along but

we try to just illustrate some very basic ones.

And as the first pass we showed you the graphics image function,

how you could actually do that from an array frameworks rate.

So if you thought of a NIfTI object as a three dimensional array.

And you're normally used to applying matrices for

example using image form the graphics package.

Then that might be a first pass but

we believe that there are there are better ways to do it.And more intuitive ways

with these three functions from the [INAUDIBLE] NIfTI package.

Thank you.

[SOUND]

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.

© 2018 Coursera Inc. Todos os direitos reservados.

Baixar na App Store

Baixar no Google Play