dwi_preprocessing & data preparation

This is the set of scripts that will be used to preprocess the NODDI data in the KIMEL lab for the PACTMD study.

To Do List

• Add dwi preprocessing scripts to repository
• Update all scripts to run from current directory & ensure that I can run from raw data
• Update the surface script - make sure all participants who are eligible are included

Briefly, the steps we are including are:

First Script merging_shells_01.sh

1. use dwidenoise (PCA based denoising) from the mrtrix package
2. run the updated version of EDDY (can handle higher b-values) using the fieldmap from step 1
3. run dwidebiascorrect from the mrtrix package this ended up not being included after all

Second script

1. reconstruct GE fieldmap: mk_ge_fieldmap_02.sh note, in the future, I want to use synthetic b0 or ANTs registration to correct distortion rather than fieldmaps which are unreliable

Third Script

1. run FSL eddy's algorithm - see run_eddy_SLURM_03.sh

Fourth Script

1. run the NODDI algorithm. We will be using both the mdt toolbox https://mdt-toolbox.readthedocs.io/en/latest_release/ to fit the NODDI parameters.
2. see the run_mdt_cuda_04.sh script to run the multishell estimation using NVIDIA cuda acceleration. This is so much better than the base version from MATLAB - each person's data can be estimated in ~10 minutes instead of 30-40 minutes. image below from: https://www.researchgate.net/profile/Ariana_Anderson/publication/283259865/figure/fig4/AS:561702509793284@1510931722192/Example-images-from-model-parameters-from-NODDI-the-intracellular-volume-fraction.png

Fifth Step (multiple scripts)

1. The first approach is to use GBSS (https://github.com/arash-n/GBSS). Note I had to alter this substantially to get it to work. Here are my versions of Arash Nazeri's scripts (see GBSS folder)

Sixth Step

1. convert the data to surface space using CIFTIFY (Erin Dickie's work). This can be run using the transform_NODDI_To_MNI_Surface_06.sh script. The benefits of using surface based approches are a) we can now smooth the data without fear of smudging different tissue classes into one another (thus increasing power and accuracy), and b) unlike the GBSS algorithm, we can get a more continuous estimate of the entire surface without gaps using the midthickness values. In the figure below for example, orientation dispersion index (ODI) is represented by panel C. Image is taken from Fukutomi, H., Glasser, M. F., Zhang, H., Autio, J. A., Coalson, T. S., Okada, T., ... & Hayashi, T. (2018). Neurite imaging reveals microstructural variations in human cerebral cortical gray matter. Neuroimage, 182, 488-499.