All packages we will discuss are loaded on the RStudio Server:

• oro.nifti - reading/writing NIfTI images
  • made the nifti object/data class: like an array - but with header information
    • the main data class we will use
• neurobase - extends oro.nifti and provides helpful imaging functions

Let’s load them:

library(oro.nifti)
library(neurobase)

We will use the readnii function (from neurobase) to read in a nifti object (this is an R object).

Here we read in the “training01_01_t1.nii.gz” file, and assign it to an object called t1:

t1 = readnii("training01_01_t1.nii.gz")

Now, an object t1 is in memory/the workspace.

class(t1)

[1] "nifti"
attr(,"package")
[1] "oro.nifti"

By default, if you simply pass the object, it is printed, we can also do print(t1):

t1

NIfTI-1 format
  Type            : nifti
  Data Type       : 4 (INT16)
  Bits per Pixel  : 16
  Slice Code      : 0 (Unknown)
  Intent Code     : 0 (None)
  Qform Code      : 2 (Aligned_Anat)
  Sform Code      : 1 (Scanner_Anat)
  Dimension       : 408 x 512 x 152
  Pixel Dimension : 0.43 x 0.43 x 0.82
  Voxel Units     : mm
  Time Units      : sec

These work with an image and a number (img + 2) or two images of the same dimensions img1 + img2.

• Comparison: >, >=, <, <=, == (equals), != (not equal)
• Logical: ! - not, & - and, | - or (a “pipe”)
• Arithmetic: +, -, *, /, ^ - exponents
• Standard math functions: log, abs, sqrt

t1 + t1 + 2 # still a nifti

NIfTI-1 format
  Type            : nifti
  Data Type       : 4 (INT16)
  Bits per Pixel  : 16
  Slice Code      : 0 (Unknown)
  Intent Code     : 0 (None)
  Qform Code      : 2 (Aligned_Anat)
  Sform Code      : 1 (Scanner_Anat)
  Dimension       : 408 x 512 x 152
  Pixel Dimension : 0.43 x 0.43 x 0.82
  Voxel Units     : mm
  Time Units      : sec

Again, we can use a logical operation. Let’s create an image indicating values over \(400\):

class(t1 > 400) # still a nifti

[1] "nifti"
attr(,"package")
[1] "oro.nifti"

head(t1 > 400) # values are now logical vs. numeric

[1] FALSE FALSE FALSE FALSE FALSE FALSE

We will refer to images such as t1 > 400 as a “mask”, simply binary images with logical values in them (or 0s and 1s)

The subsetting here is similar to that of arrays. Since t1 is 3-dimensional the subsetting goes to the 3rd dimension:

t1[5, 4, 3]

[1] 0

t1[5, 4, ] # returns a vector of numbers (1-d)
t1[, 4, ] # returns a 2-d matrix
t1[1, , ] # returns a 2-d matrix

• You can subset with a logical array of the same dimensions!
• We can view values of the t1 greater than 400 (head only prints the first 6 values):

head(t1[ t1 > 400 ]) # produces a vector of numbers

[1] 402 412 435 448 453 430

The which function works to get indices, but you can pass the arr.ind = TRUE argument to get “array” indices:

head(which(t1 > 400, arr.ind = TRUE))

     dim1 dim2 dim3
[1,]  180  258    1
[2,]  175  259    1
[3,]  176  259    1
[4,]  177  259    1
[5,]  178  259    1
[6,]  179  259    1

But can get the “vector” indices as well:

head(which(t1 > 400, arr.ind = FALSE))

[1] 105036 105439 105440 105441 105442 105443

Subsetting can work on the left hand side of assignment too:

t1_copy = t1
t1_copy[ t1_copy > 400 ] = 400 # replaced these values!
max(t1_copy) # should be 400

[1] 400

max(t1) 

[1] 1691

Note, although t1_copy was copied from t1, they are not linked - if you change values in t1_copy, values in t1 are unchanged.

We now can write out this modified t1_copy image:

writenii(nim = t1_copy, 
         filename = "training01_t1_under400.nii.gz")
file.exists("training01_t1_under400.nii.gz")

[1] TRUE

We have seen that file.exists returns TRUE if a file exists

• useful in conjunction with all: all(file.exists(VECTOR_OF_FILES))

To convert a nifti to a vector, you can simply use the c() function:

vals = c(t1)
class(vals)

[1] "numeric"

Essentially “strings out” the array. If you do array(c(t1), dim = dim(t1)), this will put things back “in order” of the t1.

Vectorizing is useful for making data.frames (covered later) when you want to do modeling at a voxel level.

df = data.frame(t1 = c(t1), mask = c(t1 > 400)); head(df)

  t1  mask
1  0 FALSE
2  0 FALSE
3  0 FALSE
4  0 FALSE
5  0 FALSE
6  0 FALSE

Use paste if you want to put strings together with spaces, paste0 no spaces by default.

file.path(directory, filename) will paste directory and filename w/file separators (e.g. /)

c(paste("img", ".nii.gz"), paste0("img", ".nii.gz"))

[1] "img .nii.gz" "img.nii.gz" 

x = file.path("output_directory", paste0("img", ".nii.gz")); print(x)

[1] "output_directory/img.nii.gz"

nii.stub will strip off the nifti extension. If bn = TRUE, it removes the directory as well:

c(nii.stub(x), nii.stub(x, bn = TRUE))

[1] "output_directory/img" "img"                 

• oro.nifti - reading/writing NIfTI images
• neurobase - extends oro.nifti and provides helpful imaging functions
• fslr - wraps FSL commands to use in R
  • registration, image manipulation, skull stripping
• ANTsR - wrapper for Advanced normalization tools (ANTs) code
  • registration, inhomogeneity correction, lots of tools
• extrantsr - allows ANTsR to work with objects from oro.nifti

Data Package we will use

• ms.lesion - contains training/testing data of patients with multiple sclerosis (MS)
  • from an open MS MRI data set (Lesjak et al. 2017)

• We have (briefly) covered some R data classes and types to get you started
• We will be using nifti objects
  • They are special 3-dimensional arrays
  • Contain numbers or logicals
• readnii and writenii are used for reading/writing nifti objects to NIfTI files
• We have briefly covered subsetting and image manipulation
  • more on that later

http://johnmuschelli.com/imaging_in_r