Automated image analysis with QIT

Welcome! This is a guide for automated quantitative analysis of diffusion MRI data using QIT. If you have collected many diffusion MRI scans in an imaging study, this workflow can help you extract quantitative measures of structural anatomical characteristics across your group and ultimately use these measures to explore their statistical relationship with demographic and behavioral variables.

Overview

The workflow is an end-to-end tool that supports a variety of ways to analyze diffusion MRI data, in which you start from raw imaging data and produce geometric models and quantitative summaries of tissue microstructure, morphometry, and connectivity. Specifically, you can perform region-based and tractography-based analysis and combine these with tissue parameter mapping using diffusion tensor imaging and advanced multi-shell approaches. In addition to diffusion MRI, the workflow has components that streamline the use of state-of-the-art packages for morphometric analysis using T1-weighted MRI data, and further, allow them to be combined with diffusion MRI data. If you are new to diffusion MRI and would like to learn more, Diffusion Tensor Imaging: a Practical Handbook and NMR in Biomedicine: Special Issue on Diffusion MRI of the brain are both good starting points.

In the following sections, this guide will cover how to install the necessary software on your computer, how to prepare your imaging data, how to run various analyses, and how to combine the results across subjects.

Installing software

This section will guide you through the installation of QIT and its dependencies. The workflow is designed for the command line, and if you need need to learn about it, you can check out the tutorials at Software Carpentry. Note: while the workflow primarily uses QIT, it also depends on several other 3rd party tools that are designed to run only on Linux and Mac. So while QIT itself can run on Windows, to run the complete workflow on Windows you would need a virtualization solution, such as VMware, VirtualBox, or the Windows Subsystem for Linux.

Installing QIT

You should first follow the instructions on the Installation page, and make sure that the bin directory is included in your shell PATH variable. You can read [https://linuxize.com/post/how-to-add-directory-to-path-in-linux/ this page] to learn more about adding a directory to your path. After installing, you can check that things worked by running the following command and seeing if it produces something analogous (your version should be this one or later):

$ qit --version
QIT build revision: 2005\:2019, build time: 03/25/2020 05\:09 PM`

Installing dependencies

Next, you should install the other software packages required by the workflow, which are two common and publicly available 3rd party neuroimaging tools: DTI-TK and FSL. You can follow the installation instructions on the linked pages, and afterwards, you can test that they are correctly installed by trying these commands:

$ which bedpostx
/opt/share/fsl-5.0.10/bin/bedpostx

$ which TVMean
/opt/share/dtitk/bin/TVMean

Note: there is a GPU version of FSL bedpostx that can dramatically speed things up. If you install it using the instructions on the linked page, you can enable it in the QIT workflow by adding GPU=1 to the qitdiff command.

Installing more dependencies (optional)

As an optional step, if want to use T1 MRI data in the analysis, you should also install [https://surfer.nmr.mgh.harvard.edu/ FreeSurfer] and [https://stnava.github.io/ANTs/ ANTs]. These are not required for most parts of the pipeline, but they can offer more precise segmentation of certain brain structures, which are described in more detail later in the guide. You can follow the installation instructions on the linked pages, and you can test that they are correctly installed by trying these commands:

$ which recon-all
/opt/share/freesurfer/bin/recon-all

$ which antsRegistrationSyn.sh
/opt/share/ANTs-install/bin/antsRegistrationSyn.sh

Preparing your data

This section provides instructions for converting your data into a format that is compatible with the workflow. You may already have data in a suitable format, so not all of these steps will be necessary. The workflow expects the input to be a diffusion-weighted MRI in the NIfTI and associated text files for the b-vectors and b-values. If you need data to try out, you can download this example clinical quality single shell dataset:

Download Download dmri.tutorial.data.zip

If you’d like to skip the processing, you can also check out these example results:

Download Download qitdiff.demo.tar.gz

File format conversion

Data from the scanner typically is stored in the DICOM file format, which stores the all data from each session in a directory that combines imaging, sequence parameters, and patient information across many files contained within. You will need to convert DICOM data to the NIfTI file format, a simpler format that is more amenable to imaging research. There are a few tools available for converting between NIfTI and DICOM, but a good starting point is the dcm2niix tool. The linked page shares instructions for installing the software and applying it to your DICOM data to produce NIfTI volumes. Afterwards, you should be left with nii.gz files from your session, one of which will be a multi-channel diffusion MRI. You should also find bvecs and bvals files, which list the b-vectors and b-values sequence parameters from the scan, respectively.

Artifact Correction

There are a variety of artifact that can be corrected in diffusion MRI, and it is recommended that you apply any corrections that are compatible with your data collection scheme. QIT offers some basic corrections for denoising, signal drift correction, and motion correction, and these will be noted in the next section. However, if possible, it is recommended that you use advanced preprocessing tools when possible. In particular, if you have collected data with an additional set of reversed-phase encoding scans, you can use FSL EDDY to greatly improve the quality of your data by simultaneously correcting for motion, eddy-current, susceptibility-induced geometric distortion. You should explore using an OpenMP accelerated version, as it can be quite time consuming. To apply it to your data, you can use the FSL or the HCP pipelines. Once you’ve finished processing it, you should find the corrected diffusion MRI and b-vectors, which will be named eddy_out.nii.gz and eddy_out.eddy_rotated_bvecs, respectively. The b-values are not changed by the correction, so you should use the raw values.

If you are interested in other artifact correction features of QIT, you can check out these modules: VolumeFilterNLM, VolumeDwiCorrectDrift, and VolumeDwiCorrect, which provide denoising, drift correction, and motion correction, respectively.

General considerations

In this section and the following three, we describe the different ways that you can run the QIT diffusion workflow. We’ll start by discussing the general interface to the script, and then describe specific types of analysis in detail. The primary interface to the QIT diffusion workflow is a program named qitdiff. If you run this command without any arguments, you’ll find a usage page, like so:

$ qitdiff

  Name: qitdiff

  Description:

    Process diffusion-weighted imaging data using the qit diffusion workflow.
    The first time you run this script, you must provide the DWI, bvecs, and
    bvals for your subject.  After that, you only need to specify the targets. 

  Usage: 

    qitdiff [opts] --subject subject [targets]

  Input Parameters (must be include in the first run, but not afterwards):

     --dwi dwi.nii.gz:   the diffusion weighted MRI file, stored in NIFTI
     --bvecs bvecs:      the diffusion b-vectors file, stored as text
     --bvals bvals:      the diffusion b-values file, stored as text
     --subject subject:  the subject directory where the results will be saved

  Optional Parameters (may be included in the first run, but not afterwards):

     --denoise <num>:    denoise the dwi with the given noise level
     --mask <fn>:        use the given precomputed brain mask
     --bet <num>:        use the given FSL BET fraction parameter 
     --erode <num>:      erode the brain mask by the given number of voxels
     --motion:           correct for motion and create a summary report
     --nomatch:          skip gradient matching step
     --tone <fn>:        specify a T1-weighted MRI
     --freesurfer <dir>: use the given precomputed freesurfer results
     --tracts <dir>:     include user-defined tracts in the workflow 

  Author: Ryan Cabeen

There are three flags for specifying the input to the workflow: --dwi, --bvecs, and --bvals. The b-vectors and b-values can be organized in column or row form, and QIT will automatically determine the coordinate transform necessary to align the b-vectors to the imaging data (so there is no need to flip or transpose b-vectors). You must also specify a subject directory with --subject, but if you omit this flag, the current working directory will be used. There are several other optional flags that are listed after the required flags. You may skip these for now if you like.

Besides the input data, you must also specify one or more targets. A target is a string that identifies what you want the workflow to produce, e.g. regional summaries of FA or fiber bundle geometry. In the next section of the tutorial we’ll describe the different targets that are available. But to give a concrete example, you could produce region DTI parameter statistics with the JHU atlas using a command like so:

$ qitdiff --subject qitsubject --dwi scan/dwi.nii.gz --bvecs scan/bvecs.txt --bvals scan/bvals.txt diff.region/jhu.labels.dti.map

One of the features of the QIT workflow is that you can add additional analyses and re-use the previous work you did. So for example, if you wanted to add TBSS to your analysis, you could simply run a command like this:

$ qitdiff --subject qitsubject atlas.region/jhu.labels.tbss.dti.map

For example, this will skip the data import and atlas registration steps, as they were already completed. Note that you can also omit the flags for the input data, since they have already been imported.

Region-of-interest analysis

Region-of-interest (ROI) analysis is a simple and widely used way to summarize diffusion MRI data. In this approach, the data is spatially normalized with an atlas that contains a parcellation of brain areas, stored in a 3D mask. Then, diffusion model parameters are statistically summarized in each ROI of the atlas; for example, by computing the average fractional anisotropy in each brain area. There are quite a few variations on this approach, and QIT supports many of them. For example, you can combine the ROI approach with Tract-based Spatial Statistics (TBSS), you can apply erosion of the masks, and you can compute the statistics in either native or atlas space. Each of these possibilities is available with a different target that you can provide to qitdiff. Here is a table summarizing the different kinds of ROI targets that are available:

  • diff.region/jhu.labels.dti.map:
    • DTI parameters statistics for each region in native space
  • diff.region/jhu.labels.erode.dti.map:
    • DTI parameters statistics for each region in native space with additional erosion step
  • atlas.region/jhu.labels.dti.map:
    • DTI parameters statistics for each region in atlas space
  • atlas.region/jhu.labels.erode.dti.map:
    • DTI parameters statistics for each region in atlas space with an additional erosion step
  • atlas.region/jhu.labels.tbss.dti.map:
    • DTI parameters statistics for each region in atlas space with additional TBSS processing

The above examples use the Johns Hopkins University white matter atlas, but there are many other alternatives in qitdiff. Below is a table summarizing these other options. There are targets available for these, in which you simply replace jhu.labels in the targets above.

  • jhu.labels: Johns Hopkins University deep white matter atlas)
  • jhu.tracts: Johns Hopkins University white matter tract atlas
  • fsa.ccwm: FreeSurfer-based corpus callosum parcellation
  • fsa.scgm: FreeSurfer-based subcortical gray matter
  • fsa.dkwm: FreeSurfer-based superficial white matter
  • hox.sub: Harvard-Oxford subcortical parcellation
  • ini.bstem: USC INI brainstem parcellation
  • cit.amy: Caltech amygdala subfields
  • cit.sub: Caltech subcortical parcellation

Furthermore, there are many types of diffusion parameters that can be used in an ROI analysis. The examples listed above used diffusion tensor imaging (DTI) parameters, but if you have multi-shell data, there are other possibilities supported by qitdiff. The table below lists these other possibilities, and you can create targets for them by substituting dti with the appropriate model identifier.

  • dti : Diffusion Tensor Imaging (DTI)
  • fwdti: Free-water Elimination DTI (with a fixed diffusivity ball)
  • dki: Diffusion Kurtosis Imaging
  • noddi: Neurite Orientation Dispersion and Density Imaging
  • mcsmt: Multi-compartment Spherical Mean Technique
  • bti: Bi-tensor Imaging (DTI with an unconstrained ball)

For a concrete example, if you wanted to compute NODDI parameters in subcortical brain areas, you could run something like so:

$ qitdiff --subject qitsubject --dwi scan/dwi.nii.gz --bvecs scan/bvecs.txt --bvals scan/bvals.txt diff.region/fsa.scgm.noddi.map

With all of these options, you may be wondering what to pick, and in this case, a good starting point is the target atlas.region/jhu.region.tbss.dti.map. This is a fairly standardized approach that summarizes DTI parameters in white matter areas from the JHU atlas with preprocessing using TBSS. For example, this is similar to the protocol used in the ENIGMA network (note: the template and registration algorithm are not identical).

Tractography-based analysis

Tractography analysis is an alternative approach for diffusion MRI analysis that is focused on modeling connectivity as opposed to regions. In this approach, geometric models of white matter connectivity are reconstructed from fiber orientation data estimated from diffusion MRI. With sufficiently high quality data, this approach can extract the major pathways of the brain, known as fiber bundles. The QIT diffusion workflow includes such a bundle-specific analysis, enabling the quantitative characterization of both whole bundles and subdivisions along their length. These can be run using the following qitdiff targets:

  • diff.tract/bundles.map:
    • whole bundle parameters, such as volume, length, etc.
  • diff.tract/bundles.whole.vertex.dti.map:
    • whole bundle DTI parameters, such as FA, MD, etc.
  • diff.tract/bundles.along.vertex.dti.map:
    • along-bundle DTI parameters, e.g. FA measured at each of a sequence of bundle subdivisions

Similar to ROI analysis, you can also extract multi-shell diffusion MRI parameters using any of the model identifiers from the previous section. For a concrete example, if you wanted to compute NODDI parameters along each bundle, you could run something like so:

$ qitdiff --subject qitsubject --dwi scan/dwi.nii.gz --bvecs scan/bvecs.txt --bvals scan/bvals.txt diff.tract/bundles.along.vertex.noddi.map

Multi-modal analysis

Besides diffusion MRI, the QIT workflow has integrated morphometric analysis using T1-weighted MRI. These steps require ANTs and FreeSurfer to be installed (as described above). After that, you can perform cortical surface based analysis using FreeSurfer using the following command:

$ qitdiff --subject qitsubject --tone t1.nii.gz tone.fs.map

If you’ve already run FreeSurfer, you can also import the previous results, e.g.

$ qitdiff --subject qitsubject --freesurfer fs_subject_dir tone.fs.map

You can also perform a multi-modal image analysis that combines the T1-weighted and diffusion MRI data. For example, you could compute DTI parameter statistics in subject-specific subcortical and superficial white matter ROIs using this command:

$ qitdiff --subject qitsubject --dwi scan/dwi.nii.gz --bvecs scan/bvecs.txt --bvals scan/bvals.txt --tone t1.nii.gz tone.region/fs.scgm.dti.map tone.region/fs.dkwm.dti.map

There are various other T1-related targets available, which are summarized here:

  • tone.fs.brain:
    • Freesurfer brain image and tissue types converted into nifti
  • tone.fs.region:
    • Freesurfer regions-of-interest converted into nifti
  • tone.fs.surfaces:
    • Freesurfer meshes converted into VTK
  • diff.fs.brain:
    • Freesurfer brain image and tissue types deformed into native diffusion space
  • diff.fs.region:
    • Freesurfer regions-of-interest deformed into native diffusion space
  • diff.fs.surfaces:
    • Freesurfer meshes deformed into native diffusion space
  • tone.region/fs.scgm.dti.map:
    • DTI parameters statistics for each Freesurfer subcortical ROI
  • tone.region/fs.dkwm.dti.map:
    • DTI parameters statistics for each Freesurfer Desikan white matter ROI
  • tone.region/fs.dkgm.dti.map:
    • DTI parameters statistics for each Freesurfer Desikan gray matter ROI

Multi-modal Tractography Analysis

The QIT workflow also supports performing tract-based analysis in T1 space. This can be helpful for visualizing bundles with the relatively high resolution reference of the T1 scan and cortical surface models. To do this, you simply need to replace diff.tract with tone.tract in the workflow target. For example, the following command:

$ qitdiff --subject qitsubject --dwi scan/dwi.nii.gz --bvecs scan/bvecs.txt --bvals scan/bvals.txt --freesurfer fs_subject_dir tone.tract/bundles.map

Note: the above commands perform linear intra-subject registration. This makes sense if you have minimal artifact or performed artifact correction. If you are working with clinical data where there may be morphometric differences between the T1 and DWI scans, you can enable deformable intra-subject registration by adding this flag: INTRA_DEFORM=1

Aggregating results

Once you’ve finished processing your data, you may want to combine the results into data tables, that is, aggregating metrics from all of the research subjects into a spreadsheet. The QIT workflow was designed to make this easy by saving results in a standardized format. You may notice that each target ends in map; this indicates that the target is a directory that contains summary statistics. Each map directory contains an array of CSV files store simple name-value pairs. For example, the bundle-specific analysis produces this CSV file among many others:

$ head diff.tract/bundles.map/volume.csv 
name,value
lh_acoustic,13570.0
lh_thal_tempinf,9663.0
rh_arc_ant,51518.0
rh_thal_pma,21428.0
rh_med_lem,11603.0
rh_aft,6183.0
...

Suppose you have many subjects, and you have organized the data such that each subject has a QIT workflow subject directory, e.g. test_103818, inside a directory named subjects. Further, you have a list of subject identifiers that correspond to the subject directory names, and these names are stored one-per-line in a text file named sids.txt. Then, you can create a data table and inspect the output using these commands:

$ qit --verbose MapCat --pattern subjects/%{subject}/diff.tract/bundles.map/volume.csv  --vars subject=sids.txt --skip --output bundles.volume.csv
$ cat bundles.volume.csv | grep lh_acoustic | head
test_103818,lh_acoustic,13386.0
test_105923,lh_acoustic,17255.0
test_111312,lh_acoustic,12189.0
test_114823,lh_acoustic,15710.0
test_115320,lh_acoustic,14599.0
...

The resulting table is in a long format, but you can also convert it to a wide one:

$ qit TableWiden --input bundles.volume.csv --output bundles.volume.wide.csv
$ head bundles.volume.wide.csv
subject,lh_acoustic,lh_arc,...
test_103818,13386.0,8950.0,...
test_105923,17255.0,13213.0,...
test_111312,12189.0,11033.0,...
...

In this table, each column is a different brain area, and each row is a subject. We used bundle volume in the example, but the same approach works for all of the other metrics stored in map directories. If you would like to create tables for many different variables, there is a more complex but powerful script called qitmaketables that you can check out.

Fiber Bundle References

The fiber bundle pipeline uses the following abbreviations:

Identifier Description
ant_comm Anterior commissure, connecting temporal poles
cc_all Corpus callosum, all connections combined
cc_dlpfc Corpus callosum, dorsolateral prefrontal cortex connections
cc_dmpfc Corpus callosum, dorsomedial prefrontal cortex connections
cc_occip Corpus callosum, occipital lobe connections
cc_parcen Corpus callosum, central parietal cortex connections
cc_parinf Corpus callosum, inferior parietal cortex connections
cc_parmar Corpus callosum, marginal parietal cortex connections
cc_parmed Corpus callosum, medial parietal cortex connections
cc_parsup Corpus callosum, superior parietal cortex connections
cc_pma Corpus callosum, premotor area connections
cc_postcentral Corpus callosum, postcentral gyrus connections
cc_precentral Corpus callosum, precentral gyrus connections
cc_sma Corpus callosum, supplementary motor area connections
cc_tempinf Corpus callosum, inferior temporal connections
cc_tempmid Corpus callosum, middle temporal connections
cc_tempsup Corpus callosum, superior temporal cortex connections
cc_vlpfc Corpus callosum, ventrolateral prefrontal cortex connections
cc_vmpfc Corpus callosum, ventromedial prefrontal cortex connections
forceps_major Forceps major, connections through the splenium of the corpus callosum
forceps_minor Forceps minor , connections through the genu of the corpus callosum
mid_cerebellar Middle cerebellar peduncle
lh_acoustic Acoustic radiations, from thalamus to superior temporal lobe
lh_aft Acumbofrontal tract, connecting nucleus accumbens and prefrontal cortex
lh_amyfug Amygofugal tract, connecting amygdala and hypothalamus
lh_arc_ant Arcuate fasciculus, anterior segment
lh_arc_long Arcuate fasciculus, long segment
lh_arc_post Arcuate fasciculus, posterior segment
lh_atr Anterior thalamic radiation
lh_cerebellar_inf Inferior cerebellar peduncle
lh_cerebellar_sup Superior cerebellar peduncle
lh_cing_body Cingulum bundle, body
lh_cing_dor Cingulum bundle, dorsal connections of cingulate cortex
lh_cing_genu Cingulum bundle, segment around genu of the corpus callosum
lh_cing_long Cingulum bundle, long segment connecting frontal and temporal lobes
lh_cing_ven Cingulum bundle, ventral portion in temporal lobe
lh_corona_ant Corona radiata, anterior portion
lh_corona_post Corona radiata, posterior portion
lh_cst Corticospinal tract
lh_fat Frontoaslant tract
lh_fornix Fornix
lh_ifof Inferior fronto-occipital fasciculus
lh_ilf Inferior longitudinal fasciculus
lh_lat_lem Lateral leminiscus
lh_med_lem Medial leminiscus
lh_mlf Middle longitudinal fasciculus
lh_opticrad Optic radiation
lh_slfi Superior longitudinal fasciculus I, connections to superior frontal cortex
lh_slfii Superior longitudinal fasciculus II, connections to middle frontal cortex
lh_slfiii Superior longitudinal fasciculus III, connections of inferior frontal cortex
lh_str Superior thalamic radiation
lh_thal_pcc Thalamocortical pathway, posterior cingulate connections
lh_thal_pole Thalamocortical pathway, temporal pole connections
lh_thal_stem Thalamocortical pathway, brainstem connections
lh_thal_dlpfc Thalamocortical pathway, dorsolateral prefrontal cortex connections
lh_thal_dmpfc Thalamocortical pathway, dorsomedial prefrontal cortex connections
lh_thal_occ Thalamocortical pathway, occipital lobe connections
lh_thal_parcen Thalamocortical pathway, central parietal cortex connections
lh_thal_parinf Thalamocortical pathway, inferior parietal cortex connections
lh_thal_parmar Thalamocortical pathway, marginal parietal cortex connections
lh_thal_parmed Thalamocortical pathway, medial parietal cortex connections
lh_thal_parsup Thalamocortical pathway, superior parietal cortex connections
lh_thal_pma Thalamocortical pathway, premotor area connections
lh_thal_postcentral Thalamocortical pathway, postcentral gyrus connections
lh_thal_precentral Thalamocortical pathway, precentral gyrus connections
lh_thal_sma Thalamocortical pathway, supplementary motor area connections
lh_thal_tempinf Thalamocortical pathway, inferior temporal connections
lh_thal_tempmid Thalamocortical pathway, middle temporal connections
lh_thal_tempsup Thalamocortical pathway, superior temporal cortex connections
lh_thal_vlpfc Thalamocortical pathway, ventrolateral prefrontal cortex connections
lh_thal_vmpfc Thalamocortical pathway, ventromedial prefrontal cortex connections
lh_umotor U-fibers of motor cortex
lh_usomato U-fibers of somatosensory cortex
lh_unc Uncinate fasciculus, connecting prefrontal cortex to temporal pole
lh_unc_lat Uncinate fasciculus, connection lateral prefrontal cortex to temporal pole
lh_unc_med Uncinate fasciculus, connection medial prefrontal cortex to temporal pole
lh_vof Vertical occipital fasciculus
lh_vof_lat Vertical occipital fasciculus, lateral portion
lh_vof_med Vertical occipital fasciculus, medial portion

Note: lh and rh indicate the left and right hemisphere bundles, respectively. Only the left bundles are listed above for brevity

Acknowledgements

If you find QIT is valuable in your work, we ask that you clearly acknowledge it any publications or grant proposals. This is greatly appreciated, as it improves the reproducibility of your findings, and it helps our team maintain resources for continued support and development. You can cite QIT by including a link to the main website at http://cabeen.io/qitwiki and including the following citation in the manuscript:

Cabeen RP, Laidlaw DH, Toga AW. 2018. Quantitative Imaging Toolkit: Software
for Interactive 3D Visualization, Data Exploration, and Computational Analysis
of Neuroimaging Datasets. Proceedings of the Annual Meeting of the Society for
Magnetic Resonance in Medicine (ISMRM).  Paris, France 2018 (p. 2854)

If you use the region-of-interest analysis, you can include the following reference, which describes and evaluates these components:

Cabeen, R.P., Bastin, M.E. and Laidlaw, D.H., 2017. A Comparative evaluation of
voxel-based spatial mapping in diffusion tensor imaging. Neuroimage, 146,
pp.100-112.

If you used the tractography-based analysis, you can include the following references, which describe and evaluate the methods for tractography and template construction:

Cabeen, R.P., Bastin, M.E. and Laidlaw, D.H., 2016. Kernel regression
estimation of fiber orientation mixtures in diffusion MRI. Neuroimage, 127,
pp.158-172.

Cabeen, R.P., Toga, A.W., 2020. Reinforcement tractography: a hybrid approach
for robust segmentation of complex fiber bundles. International Symposium on
Biomedical Imaging (ISBI) 2020

Besides QIT references, you should also please cite the other dependent tools and resources where appropriate:

Jenkinson, M., Beckmann, C.F., Behrens, T.E., Woolrich, M.W. and Smith, S.M.,
2012. Fsl. Neuroimage, 62(2), pp.782-790.

Behrens, T.E., Berg, H.J., Jbabdi, S., Rushworth, M.F. and Woolrich, M.W.,
2007. Probabilistic diffusion tractography with multiple fibre orientations:
What can we gain?. Neuroimage, 34(1), pp.144-155.

Zhang, H., Yushkevich, P.A., Alexander, D.C. and Gee, J.C., 2006. Deformable
registration of diffusion tensor MR images with explicit orientation
optimization. Medical image analysis, 10(5), pp.764-785.

Zhang, S., Peng, H., Dawe, R.J. and Arfanakis, K., 2011. Enhanced ICBM
diffusion tensor template of the human brain. Neuroimage, 54(2), pp.974-984.

Zhang, H., Yushkevich, P.A., Alexander, D.C. and Gee, J.C., 2006. Deformable
registration of diffusion tensor MR images with explicit orientation
optimization. Medical image analysis, 10(5), pp.764-785.

Basser, P.J., Mattiello, J. and LeBihan, D., 1994. MR diffusion tensor
spectroscopy and imaging. Biophysical journal, 66(1), pp.259-267.

Zhang, H., Schneider, T., Wheeler-Kingshott, C.A. and Alexander, D.C., 2012.
NODDI: practical in vivo neurite orientation dispersion and density imaging of
the human brain. Neuroimage, 61(4), pp.1000-1016.

Hoy, A.R., Koay, C.G., Kecskemeti, S.R. and Alexander, A.L., 2014. Optimization
of a free water elimination two-compartment model for diffusion tensor imaging.
Neuroimage, 103, pp.323-333.

Kaden, E., Kelm, N.D., Carson, R.P., Does, M.D. and Alexander, D.C., 2016.
Multi-compartment microscopic diffusion imaging. NeuroImage, 139, pp.346-359.

Fieremans, E., Jensen, J. H., & Helpern, J. A. (2011). White matter
characterization with diffusional kurtosis imaging. Neuroimage, 58(1), 177-188.

Sepehrband, F., Cabeen, R.P., Choupan, J., Barisano, G., Law, M., Toga, A.W.
and Alzheimer's Disease Neuroimaging Initiative, 2019. Perivascular space fluid
contributes to diffusion tensor imaging changes in white matter. NeuroImage,
197, pp.243-254.

Mori, S., Oishi, K., Jiang, H., Jiang, L., Li, X., Akhter, K., Hua, K., Faria,
A.V., Mahmood, A., Woods, R. and Toga, A.W., 2008. Stereotaxic white matter
atlas based on diffusion tensor imaging in an ICBM template. Neuroimage, 40(2),
pp.570-582.

Smith, S.M., Jenkinson, M., Johansen-Berg, H., Rueckert, D., Nichols, T.E.,
Mackay, C.E., Watkins, K.E., Ciccarelli, O., Cader, M.Z., Matthews, P.M. and
Behrens, T.E., 2006. Tract-based spatial statistics: voxelwise analysis of
multi-subject diffusion data. Neuroimage, 31(4), pp.1487-1505.

Pauli, W.M., Nili, A.N. and Tyszka, J.M., 2018. A high-resolution probabilistic
in vivo atlas of human subcortical brain nuclei. Scientific data, 5, p.180063.

Tyszka, J.M. and Pauli, W.M., 2016. In vivo delineation of subdivisions of the
human amygdaloid complex in a high‐resolution group template. Human brain
mapping, 37(11), pp.3979-3998.

Tang, Y., Sun, W., Toga, A.W., Ringman, J.M. and Shi, Y., 2018. A probabilistic
atlas of human brainstem pathways based on connectome imaging data. Neuroimage,
169, pp.227-239.

Desikan, R.S., Ségonne, F., Fischl, B., Quinn, B.T., Dickerson, B.C., Blacker,
D., Buckner, R.L., Dale, A.M., Maguire, R.P., Hyman, B.T. and Albert, M.S.,
2006. An automated labeling system for subdividing the human cerebral cortex on
MRI scans into gyral based regions of interest. Neuroimage, 31(3), pp.968-980.

Fischl, B., 2012. FreeSurfer. Neuroimage, 62(2), pp.774-781.

Avants, B.B., Tustison, N. and Song, G., 2009. Advanced normalization tools
(ANTS). Insight j, 2(365), pp.1-35.

Closing

This concludes the tutorial for the QIT diffusion workflow. This was written to provide only a brief overview, so you may be left wondering about how some things work. If so, please feel free to reach out with any questions or thoughts via cabeen@gmail.com.