R Notebook

I was curious to know what the difference between MPRAGE & MPRAGE-DIST, and here’s a synopsis of my learning…

Inhomogeneity depends on the subject’s anatomy, the orientation of that anatomy with respect to the B0 field, as well as several other factors such as the scan-dependent magnetic shim settings. This generally means that inhomogeneity is not known in advance, and should be measured (“field mapping”) at each scan session in order to construct an accurate forward model of the distortion process (Chambers et al., 2015).

There are numerous ways to complete beta inhomogenity correction for EPI images (if fieldmaps are not available). In all circumstances though, it seems most efficient to use one’s field maps collected durring that session, if available.

• The post-hoc fsl method that I reviewed can be found here: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4662423/

  • They register the distorted fMRI to a T1-weighted image, as it has high in tissue contrast and provides mutual information about inhomogenity (if there is not alot of motion between MPRAGE and resting state aquistions).

  • Here are their results: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4662423/figure/F1/

  • MPRAGE distortion correction is automatically programmed into our default pipeline, because, really there is no better alternative. As stated above, the process seems to be modeled forward into MPRAGE reconstruction.

  • Output from dcm2niix:

    • /MPRAGE == MPRAGEDistCorrND
    • /MPRAGE-DIST == MPRAGEDistCorr

1. Bias Correction /MPRAGE

This should not be completed on MPRAGE post-hoc, as it results in over correction (?) \(\dagger\)

subdir="~/DATA/EPIC_SCRIPT/Dicom/7089/7089_1"
setwd(subdir)
fsl_biascorrect(file = "SAGT1MPRAGEDistCorND.nii.gz",
                outfile = "ND_SAGT1MPRAGEDistCor_BiasCorrect", #User defines outfile name
                opts = "-v")

1a. Voxel-Intensity Distributions: /MPRAGE

x-axis is voxel intensity
y-axis is distribution for 1 slice
each line represents a slice (2,6,10,14,18) respectivly

1b. Contrast Map: Results of inhomogenity correction of /MPRAGE

sub.bias <- niftiarr(nim, nim-fast_img)
# quantile the difference image using these as breaks
q=quantile(sub.bias[sub.bias !=0],probs = seq(0,1,by=0.1))
# get a diverging gradient palette
fcol=div_gradient_pal(low="blue",mid="yellow",high ="red")
ortho2(nim,sub.bias,col.y = alpha(fcol(seq(0,1, length=10)),
                                  0.5), ybreaks = q, ycolorbar=TRUE, text = paste0("Original
Image Minus N4", "\n Bias-Corrected Image")) 

2. Bias Correction /MPRAGE-DIST

subdir="~/DATA/EPIC_SCRIPT/Dicom/7089/7089_1"
setwd(subdir)
fsl_biascorrect(file = "SAGT1MPRAGEDistCor.nii.gz",
                outfile = "SAGT1MPRAGEDistCor_BiasCorrect", 
                opts = "-v")

2a. Voxel-Intensity Distributions: /MPRRAGE-DIST

The code for plotting:

2b. Contrast Map: Results of inhomogenity correction of /MPRAGE-DIST

Contrast Map 3: Results of inhomegentity correction built into DICOM data structure

Note the central areas of grey scale are overlapping sources of distortion-correction due to perturbed K-Space. This is essentially like plotting the residuals in a linear regression model (?) \(\dagger\)

• http://mriquestions.com/what-is-k-space.html

Commparison 4: MPRAGE(ND) minus Bias Corrected MPRAGE-DIST.

Example results of over-correction (?) or application of dist. correction field?

One notable observation is the voxel-intensity distr. plot is identical to 2a.