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).
Introduction
Loading...
来自 Johns Hopkins University 的课程
Introduction to Neurohacking In R
140 个评分
与讲师见面
Dr. Elizabeth SweeneyRice Academy Postdoctoral Fellow
Biostatistics
Ciprian M. CrainiceanuProfessor
Biostatstics
John Muschelli IIIAssistant Scientist
Biostatistics
[SOUND] Welcome to Neurohacking in R,
a course presented to you by John Muschelli,
Elizabeth Sweeney, and Jipian Kinchung.
I'm Jipian Kingchung.
We are the in the Biostatistics Department at Johns Hopkins School of Public Health.
We'll start with the first lecture called Neurohacking.
And what we will do, we will go over the definition of neurohacking,
what do we mean by that?
Then we'll go over [COUGH] the goals,
the structure, and the suggested prerequisites.
So first of all, what do we mean by neurohacking?
This is an image that should provide an idea about what we mean by neurohacking.
Here what you see is an image, an axial image, through the brain.
It's an axial image.
If you don't quite follow every word that I'm saying right now, that's okay.
We'll say that throughout the course, exactly what every word means.
But this is an axial slice of a T1 image, and it represents
a subject who has an MS lesion that is identified in the middle of the brain.
And you also see a red box with
numbers associated on the right.
It may be interesting to look a little bit closer to the image to the left and
see how the left corner of the box is actually darker.
And as you progress from the left corner to the right corner of the image,
it becomes lighter, and then it becomes darker again.
If you look in the right side of the box, you see that the numbers themselves,
which correspond to the actual image that you see on the left do the same thing.
You start in the left corner from 120, 130.
Then as you progress through the image towards the right corner,
numbers becomes larger first.
In the range of 200 you see some 202 some 200, 192, and it may become smaller again.
They get in the range of 160, not as small as they were in the left corner, but
quite much smaller than actually they are in the middle of the image.
We will see much later that this is actually an image
where you see a contrast enhancement identifying the brain
blood barrier penetration by gadolinium chelates.
As I said, there are a lot words that don't make a lot of sense right now, but
we will make our way through that as we go along.
So we also have a formal definition of what neurohacking is.
Neurohacking in a nutshell is a word that we came up with, and
it essentially means that we would like to have a way of doing imaging analysis,
neuroimaging analysis in particular,
where the data is actually first of all reproducible.
And the entire analytic pipeline is reproducible.
So once you start with the data set you
go from data acquisition all the way to data analysis.
And what are the course goals for this course?
Well first of all we would like to interactively explore data with R.
We would like to do image processing in R, fsl, and ANTs.
We would like to read, write, plot, and manipulate data with R.
So, this is kind of the basic idea,
to just get a feel for what neuroimaging analysis is.
And we will do that progressively.
We will start with simple tools and
then we will progress through more complex tools as we go along.
We also like to develop some skills, and
one of these skills is to just take the difference between two images.
On this slide, you see two T1 images from a baseline scan,
or a visit one scan on the left side and a visit two scan in the middle.
And what you would like to do is to take the difference between the two and
identify where, if any, changes happen in the image.
This is something that is used extensively clinical practice when there
is an interest in identifying whether there is a change
in a particular image from one visit to another.
We will see that this is very easy to ask as a question, but
it actually requires a little bit of care in terms of how one goes
from one image to the next and how we take this difference.
Another thing that we would like to do is actually to understand
biologically where a particular lesion in the brain is.
In this case, on this slide, you see on the left side
a FLAIR image of a person who has brain cancer.
You see that the brain cancer is hyperintent.
It's a little bit lighter than the surrounding tissue.
And in the middle, you see the same exact
area, that brain cancer segmented by hand shown as a mask.
Essentially, it's a map which is zero one.
Where you see white it means that brain cancer is there,
and where you see black it means that it is not there.
What we would like to do is to take these particular ROIs and use a template,
a map, which is essentially a way of showing on a template brain,
on this map exactly what is the biological information.
What areas will the brain cancer cover and
not cover in a particular template?
So in other words, we like to see where exactly is
this template, this brain cancer in the brain.
Even though we see it, it actually is quite a bit of work to go
from seeing it to actually quantifying where it actually is and
providing that as a useful tool for clinical practice.
So what should you expect to see?
Well this is the general course structure.
This is what we would like to cover.
The first theme is what I'm doing right now, is an overview.
It's just giving the basic ideas, the basic goals, the basic strategies
of what we want to do and also provide a way to set things up.
Provide the basic ideas about where we are and
what is the type of data that we prepared, and how to analyze,
what are the types of tools that are required for analyzing the data.
Theme 2 is building progressively on Theme 1.
So we'll talk more about data structure.
We'll talk more about how to plot data, what are the file types,
visualization, and we'll also talk about multi-sequence structural MRI.
And also the differences between structural MRI and
some of the other modalities for brain imaging.
Theme 3 becomes, it's a little bit more complex,
and starts to discuss preprocessing.
In particular, we'll discuss inhomogeneity correction,
intensity normalization, registration, and also the tools in R.
So everything will actually be supported in R, and done in R.
And so that one can go directly from DICOM images to NIfTI images and
all the way through the preprocessing pipeline and
the processing pipeline towards finished products.
Theme 4 will deal with registration,
which is essentially how to transform different types of images so
that locations in images are the same.
We'll discuss about ROI quantification which is related to
what I showed a little bit earlier where we have a brain cancer region.
And we would like to say where in particular is this brain cancer located in
the brain, and we'll talk about segmentation.
Segmentation refers to essentially every algorithm that
identifies parts of the brain versus other parts of the brain.
For example, it could be something as simple as identifying white matter
segmentation versus gray matter versus CSF.
Or it could also be that we are interested in finding particular lesions or
particular parts of the brain that are hyper intense.
But it will also relate to other segmentations of the brain.
In particular we will focus more on white matter, gray matter, and CSF segmentation.
The last but not the least thing is dynamic visualization in R.
R has not been historically very friendly to dynamic visualization.
But we were able to use some of the tools out of there for
dynamic visualization where you can actually go through slices and
get a better idea about what happens.
Of course, all these things are based on other people's work.
And one of the main themes of what we are doing is to take these other tools and
integrate them in a way that you don't ever have to leave R to actually use them.
And we'll talk about papayar and itksnapr,
which are two R programs that will help us go through these dynamic visualization.
Some of the suggested prerequisites are actually the entire
course is self-contained.
So you can start from start to finish,
go through the course, download the data, and not have any problems.
However we find it very helpful to have a knowledge of Linux and/or Unix.
Basic knowledge of programming is very helpful.
It becomes harder as you go along if you don't know programming.
A basic knowledge of array data structures, understanding what a 2D,
what a 3D data structure is.
Understanding how to navigate that a little bit, having intuition about it,
that would be very, very useful.
We also found useful, again, not required, to take
the Coursera Data Science Specialization taught by our colleagues here at Hopkins.
