Resources and Goals
Much of this work has been adapted by the FSL guide for DTI reconstruction: http://camino.cs.ucl.ac.uk/index.php?n=Tutorials.DTI. We will show you a few steps that have been implemented in rcamino: camino_pointset2scheme, camino_modelfit, camino_fa, camino_md, and camino_dteig.
Reading in the Data
First, we download the data into a temporary directory the unzip it:
tdir = tempdir()
tfile = file.path(tdir, "example_dwi.zip")
download.file("http://cmic.cs.ucl.ac.uk/camino//uploads/Tutorials/example_dwi.zip",
destfile = tfile)
files = unzip(zipfile = tfile, exdir = tdir, overwrite = TRUE)
Making b-vectors and b-values
As dtifit requires the b-values and b-vectors to be separated, and this data has b-values of \(1000\) when the b-vectors is not zero. This is very important and you must know where your b-values and b-vectors are when doing your analyses and what units they are in.
library(rcamino)
b_data_file = grep("[.]txt$", files, value = TRUE)
scheme_file = camino_pointset2scheme(infile = b_data_file,
bvalue = 1e9)
/tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/pointset2scheme -inputfile '/tmp/RtmpLPgO2V/grad_dirs.txt' -bvalue 1000000000 -outputfile /tmp/RtmpLPgO2V/file3a26266e4b06.scheme
Checking our data
Here we ensure that the number of b-values/b-vectors is the same as the number of time points in the 4D image.
Loading required package: oro.nifti
oro.nifti 0.8.1
img_fname = grep("4Ddwi_b1000", files, value = TRUE)
img = readnii(img_fname)
ntim(img)
[1] 33
grads = readLines(b_data_file)
length(grads)
[1] 33
# cleanup
rm(list= "img"); gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 632653 33.8 1168576 62.5 940480 50.3
Vcells 1080071 8.3 139717724 1066.0 168960222 1289.1
Running Image Conversion
We will save the result in a temporary file (outfile), but also return the result as a nifti object ret, as retimg = TRUE. We will use the first volume as the reference as is the default in FSL. Note FSL is zero-indexed so the first volume is the zero-ith index:
float_fname = camino_image2voxel(infile = img_fname,
outputdatatype = "float")
/tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/image2voxel -inputfile '/tmp/RtmpLPgO2V/4Ddwi_b1000.nii.gz' -outputfile '/tmp/RtmpLPgO2V/file3a265adae75b.Bfloat' -outputdatatype float
Note, from here on forward we will use either the filename for the output of the eddy current correction or the eddy-current-corrected nifti object.
Fit the diffusion tensor
mask_fname = grep("mask", files, value = TRUE)
model_fname = camino_modelfit(
infile = float_fname,
scheme = scheme_file,
mask = mask_fname,
outputdatatype = "double"
)
/tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/modelfit -inputfile '/tmp/RtmpLPgO2V/file3a265adae75b.Bfloat' -outputfile '/tmp/RtmpLPgO2V/file3a2666f59a66.Bdouble' -inputdatatype float -schemefile /tmp/RtmpLPgO2V/file3a26266e4b06.scheme -bgmask /tmp/RtmpLPgO2V/brain_mask.nii.gz -maskdatatype float -model dt
Getting FA vlaues
fa_fname = camino_fa(infile = model_fname)
cat '/tmp/RtmpLPgO2V/file3a2666f59a66.Bdouble' | /tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/fa -inputmodel dt -outputdatatype double > '/tmp/RtmpLPgO2V/file3a263771e6fe.Bdouble'
Converting FA values back into an image
library(neurobase)
fa_img_name = camino_voxel2image(infile = fa_fname,
header = img_fname,
gzip = TRUE,
components = 1)
/tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/voxel2image -inputfile /tmp/RtmpLPgO2V/file3a263771e6fe.Bdouble -header /tmp/RtmpLPgO2V/4Ddwi_b1000.nii.gz -outputroot /tmp/RtmpLPgO2V/file3a2631be38b8_ -components 1 -gzip
fa_img = readnii(fa_img_name)
Converting with piping
We can chain Camino commands using the magrittr pipe operation (%>%):
library(magrittr)
fa_img2 = model_fname %>%
camino_fa() %>%
camino_voxel2image(header = img_fname, gzip = TRUE, components = 1) %>%
readnii
cat '/tmp/RtmpLPgO2V/file3a2666f59a66.Bdouble' | /tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/fa -inputmodel dt -outputdatatype double > '/tmp/RtmpLPgO2V/file3a2660e56e80.Bdouble'
/tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/voxel2image -inputfile /tmp/RtmpLPgO2V/file3a2660e56e80.Bdouble -header /tmp/RtmpLPgO2V/4Ddwi_b1000.nii.gz -outputroot /tmp/RtmpLPgO2V/file3a265ffe0ac8_ -components 1 -gzip
all.equal(fa_img, fa_img2)
[1] TRUE
Visualizing FA images
Using ortho2, we can visualize these FA maps:
Getting MD vlaues
Similar to getting FA maps, we can get mean diffusivity (MD) maps, read them into R, and visualize them using ortho2:
md_img = model_fname %>%
camino_md() %>%
camino_voxel2image(header = img_fname, gzip = TRUE, components = 1) %>%
readnii
cat '/tmp/RtmpLPgO2V/file3a2666f59a66.Bdouble' | /tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/md -inputmodel dt -outputdatatype double > '/tmp/RtmpLPgO2V/file3a266db52fa5.Bdouble'
/tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/voxel2image -inputfile /tmp/RtmpLPgO2V/file3a266db52fa5.Bdouble -header /tmp/RtmpLPgO2V/4Ddwi_b1000.nii.gz -outputroot /tmp/RtmpLPgO2V/file3a2642c8cda7_ -components 1 -gzip
Export DTs to NIfTI
Using camino_dt2nii, we can export the diffusion tensors into NIfTI files. We see the result is the filenames of the NIfTI files, and that they all exist (otherwise there’d be an errors.)
nifti_dt = camino_dt2nii(
infile = model_fname,
inputmodel = "dt",
header = img_fname,
gzip = TRUE
)
/tmp/RtmpcyCddB/Rinst383b3793ab66/rcamino/camino/bin/dt2nii -inputfile /tmp/RtmpLPgO2V/file3a2666f59a66.Bdouble -header /tmp/RtmpLPgO2V/4Ddwi_b1000.nii.gz -inputmodel dt -outputroot /tmp/RtmpLPgO2V/file3a26657298b_ -gzip
stopifnot(all(file.exists(nifti_dt)))
print(nifti_dt)
[1] "/tmp/RtmpLPgO2V/file3a26657298b_exitcode.nii.gz"
[2] "/tmp/RtmpLPgO2V/file3a26657298b_lns0.nii.gz"
[3] "/tmp/RtmpLPgO2V/file3a26657298b_dt.nii.gz"
We can read these DT images into R again using readnii, but we must set drop_dim = FALSE for diffusion tensor images because the pixel dimensions are zero and readnii assumes you want to drop “empty” dimensions
dt_imgs = lapply(nifti_dt, readnii, drop_dim = FALSE)