BUndle ANalytics (BUAN) framework

This tutorial walks through the steps for reproducing Bundle Analytics [Chandio20] results on Parkinson’s Progression Markers Initiative (PPMI) [Marek11] data derivatives. Bundle Analytics is a framework for comparing bundle profiles and shapes of different groups. In this example, we will be comparing healthy controls and patients with parkinson’s disease. We will be using PPMI data derivatives generated using DIPY [Garyfallidis14].

First we need to download streamline atlas [Yeh18] with 30 white matter bundles in MNI space from

For this tutorial we will be using a test sample of DIPY Processed Parkinson’s Progression Markers Initiative (PPMI) Data Derivatives. It can be downloaded from the link below

Note

If you prefer to run experiments on the complete dataset to reproduce the paper [Chandio20] results please see the “Reproducing results on larger dataset” section at end of the page for more information.

There are two parts of Bundle Analytics group comparison framework, bundle profile analysis and bundle shape similarity analysis.

Group Comparison of Bundle Profiles

For generating bundle profile data (saved as .h5 files): You must have downloaded bundles folder of 30 atlas bundles and subjects folder with PPMI data derivatives.

Following workflows require specific input directory structure but don’t worry as data you downloaded is already in the required format. We will be using bundles folder you downloaded from streamline atlas link and subjects_small folder downloaded from test data link.

Where, bundles folder has following model bundles:

bundles/
├── AF_L.trk
├── AF_R.trk
├── CCMid.trk
├── CC_ForcepsMajor.trk
├── CC_ForcepsMinor.trk
├── CST_L.trk
├── CST_R.trk
├── EMC_L.trk
├── EMC_R.trk
├── FPT_L.trk
├── FPT_R.trk
├── IFOF_L.trk
├── IFOF_R.trk
├── ILF_L.trk
├── ILF_R.trk
├── MLF_L.trk
├── MLF_R.trk
├── ML_L.trk
├── ML_R.trk
├── MdLF_L.trk
├── MdLF_R.trk
├── OPT_L.trk
├── OPT_R.trk
├── OR_L.trk
├── OR_R.trk
├── STT_L.trk
├── STT_R.trk
├── UF_L.trk
├── UF_R.trk
└── V.trk

The subjects_small directory has following structure:

subjects_small
├── control
│   ├── 3805
│   │   ├── anatomical_measures
│   │   ├── org_bundles
│   │   └── rec_bundles
│   ├── 3806
│   │   ├── anatomical_measures
│   │   ├── org_bundles
│   │   └── rec_bundles
│   ├── 3809
│   │   ├── anatomical_measures
│   │   ├── org_bundles
│   │   └── rec_bundles
│   ├── 3850
│   │   ├── anatomical_measures
│   │   ├── org_bundles
│   │   └── rec_bundles
│   └── 3851
│       ├── anatomical_measures
│       ├── org_bundles
│       └── rec_bundles
└── patient
    ├── 3383
    │   ├── anatomical_measures
    │   ├── org_bundles
    │   └── rec_bundles
    ├── 3385
    │   ├── anatomical_measures
    │   ├── org_bundles
    │   └── rec_bundles
    ├── 3387
    │   ├── anatomical_measures
    │   ├── org_bundles
    │   └── rec_bundles
    ├── 3392
    │   ├── anatomical_measures
    │   ├── org_bundles
    │   └── rec_bundles
    └── 3552
        ├── anatomical_measures
        ├── org_bundles
        └── rec_bundles

And each subject folder has the following structure:

├── anatomical_measures
│   ├── ad.nii.gz
│   ├── csa_peaks.pam5
│   ├── fa.nii.gz
│   ├── md.nii.gz
│   └── rd.nii.gz
├── org_bundles
│   ├── streamlines_moved_AF_L__labels__recognized_orig.trk
│   ├── streamlines_moved_AF_R__labels__recognized_orig.trk
│   ├── streamlines_moved_CCMid__labels__recognized_orig.trk
│   . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
│   . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
│   . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
│   ├── streamlines_moved_UF_L__labels__recognized_orig.trk
│   ├── streamlines_moved_UF_R__labels__recognized_orig.trk
│   └── streamlines_moved_V__labels__recognized_orig.trk
└── rec_bundles
    ├── moved_AF_L__recognized.trk
    ├── moved_AF_R__recognized.trk
    ├── moved_CCMid__recognized.trk
    . . . . . . . . . . . . . . . .
    . . . . . . . . . . . . . . . .
    . . . . . . . . . . . . . . . .
    ├── moved_UF_L__recognized.trk
    ├── moved_UF_R__recognized.trk
    └── moved_V__recognized.trk

If you want to run this tutorial on your data, make sure that the directory structure is The same as shown above. Where, anatomical_measures folder has nifti files for DTI measures such as FA, MD, and CSA/CSD pam5 files. The org_bundles folder has extracted bundles in native space. The rec_bundles folder has extracted bundles in common space.

Note

Make sure all the output folders are empty and do not get overridden.

Create an out_dir folder (eg: bundle_profiles):

mkdir bundle_profiles

Run the following workflow:

dipy_buan_profiles bundles/ subjects_small/ --out_dir "bundle_profiles"

For running Linear Mixed Models (LMM) on generated .h5 files from the previous step:

Create an out_dir folder (eg: lmm_plots):

mkdir lmm_plots

And run the following workflow:

dipy_buan_lmm "bundle_profiles/*" --out_dir "lmm_plots"

This workflow will generate 30 bundles group comparison plots per anatomical measures. Plots will look like the following example:

alternate text

Result plot for left arcuate fasciculus (AF_L) on FA measure

We can also visualize and highlight the specific location of group differences on the bundle by providing output p-values file from dipy_buan_lmm workflow. The user can specify at what level of significance they want to see group differences by providing threshold value of p-value to buan_thr (default 0.05). The color of the highlighted area can be specified by providing RGB color values to buan_highlight (Default Red)

Run the following commandline for visualizing group differences on the model bundle:

dipy_horizon bundles/AF_L.trk lmm_plots/AF_L_fa_pvalues.npy --buan --buan_thr 0.05

Where, AF_L.trk `` is located in your model bundle folder ``bundles and AF_L_fa_pvalues.npy is saved in output folder lmm_plots of dipy_buan_lmm workflow

The output of this commandline is an interactive visualization window. Example snapshot:

alternate text

Result plot for left arcuate fasciculus (AF_L) with a highlighted area on the bundle in red color. The highlighted area represents the segments on bundles with significant group differences that have pvalues < 0.05.

Let’s use a different highlight color this time on CST_L bundle:

dipy_horizon bundles/CST_L.trk lmm_plots/CST_L_fa_pvalues.npy --buan --buan_thr 0.05 --buan_highlight 1 1 0
alternate text

Result plot for left corticospinal tract left (CST_L) with a highlighted area on the bundle in yellow color. The highlighted area represents the segments on bundles with significant group differences that have pvalues < 0.05.

Shape similarity of specific bundles across the populations

Create an out_dir folder (eg: sm_plots):

mkdir sm_plots

Run the following workflow:

dipy_buan_shapes subjects_small/ --out_dir "sm_plots"

This workflow will generate 30 bundles shape similarity plots. Shape similarity score ranges between 0-1, where 1 being highest similarity and 0 being lowest. Plots will look like the following example:

alternate text

Result plot for right uncinate fasciculus (UF_R) for 10 subjects. First 5 subjects belong to the healthy control group and last 5 subjects belong to patient group. In the diagonal, we have shape similarity score of 1 as it is calculated between a bundle and itself.

Reproducing results on larger dataset:

Complete dataset of DIPY Processed Parkinson’s Progression Markers Initiative (PPMI) Data Derivatives can be downloaded from the link below:

Please note this is a large data file and might take some time to run. If you only want to test the workflows use the test sample data.

All steps will be the same as mentioned above except this time the data donwloaded will have different folder name subjects instead of subjects_small.

For more information about each command line, you can go to https://github.com/dipy/dipy/blob/master/dipy/workflows/stats.py

If you are using any of these commands do cite the relevant papers.

[Chandio20] (1,2)

Chandio, B.Q., Risacher, S.L., Pestilli, F., Bullock, D., Yeh, FC., Koudoro, S., Rokem, A., Harezlak, J., and Garyfallidis, E. Bundle analytics, a computational framework for investigating the shapes and profiles of brain pathways across populations. Sci Rep 10, 17149 (2020)

[Marek11]

Marek, Kenneth and Jennings, Danna and Lasch, Shirley and Siderowf, Andrew and Tanner, Caroline and Simuni, Tanya and Coffey, Chris and Kieburtz, Karl and Flagg, Emily and Chowdhury, Sohini and others. The parkinson progression marker initiative (PPMI). Progress in neurobiology, 2011.

[Garyfallidis14]

Garyfallidis, E., M. Brett, B. Amirbekian, A. Rokem, S. Van Der Walt, M. Descoteaux, and I. Nimmo-Smith. “DIPY, a library for the analysis of diffusion MRI data”. Frontiers in Neuroinformatics, 1-18, 2014.

[Yeh18]

Yeh F.C., Panesar S., Fernandes D., Meola A., Yoshino M., Fernandez-Miranda J.C., Vettel J.M., Verstynen T. Population-averaged atlas of the macroscale human structural connectome and its network topology. Neuroimage, 2018.