meeting date: 2019-02-14
days until DIPY: 25

last run: 2019-02-14
last edited: 2019-02-14
link: http://rpubs.com/navona/thesis_week02

Summary

This week I have been focused on learning tractography via Slicer. Primarily, I have taken time to read relevant papers and documentation, and work through the O’Donnell groups’ Slicer tutorials.

I tracked how I spent my time here.

Papers

  1. Cited heavily in DWI/DTI/tractography
    Beaulieu, C. (2002). The basis of anisotropic water diffusion in the nervous system - a technical review. NMR in Biomedicine, 15(7-8), 435–455. doi:10.1002/nbm.782 

  2. Unscented Kalman Filter (UKF) tractography
    Chen, Z., Tie, Y., Olubiyi, O., Rigolo, L., Mehrtash, A., Norton, I., … O’Donnell, L. J. (2015). Reconstruction of the arcuate fasciculus for surgical planning in the setting of peritumoral edema using two-tensor unscented Kalman filter tractography. NeuroImage: Clinical, 7, 815–822. doi:10.1016/j.nicl.2015.03.009

  3. SlicerDMRI software
    Norton, I., Essayed, W. I., Zhang, F., Pujol, S., Yarmarkovich, A., Golby, A. J., … O’Donnell, L. J. (2017). SlicerDMRI: Open Source Diffusion MRI Software for Brain Cancer Research. Cancer Research, 77(21), e101–e103. doi:10.1158/0008-5472.can-17-0332

  4. New Slicer atlas
    Zhang, F., Wu, Y., Norton, I., Rigolo, L., Rathi, Y., Makris, N., & O’Donnell, L. J. (2018). An anatomically curated fiber clustering white matter atlas for consistent white matter tract parcellation across the lifespan. NeuroImage, 179, 429–447. doi:10.1016/j.neuroimage.2018.06.027

  5. Tractography challenges
    Essayed, W. I., Zhang, F., Unadkat, P., Cosgrove, G. R., Golby, A. J., & O’Donnell, L. J. (2017). White matter tractography for neurosurgical planning: A topography-based review of the current state of the art. NeuroImage: Clinical, 15, 659–672. doi:10.1016/j.nicl.2017.06.011

Documentation

  1. Unscented Kalman Filter (UKF) tractography
    UFK is a two-tensor method for estimating streamlines. In contrast to methods that fit a model to the signal independently at each voxel, UKF employs prior information at each tracking step to help stabilize model fitting, rendering it more sensitive to crossing fibers.

  2. Medical Reality Modeling Language (MRML) scene files
    These are a variant of the XML format, and allow for 3D visualization.

Tutorials

The O’Donnell group has posted Slicer tutorials originally presented at MICCAI 2018 (Spain), here. Three tutorials are relevant to us: (1) an introduction to Slicer, SlicerDMRI, the ‘whitematteranalysis’ pipeline, and extensions; (2) automated white matter parcellation using the new 2018 atlas; and (3) Python programming to create custom Slicer modules/ pipelines.


Tutorial 1/3 [completed]

  • Using the tutorials’ test data, I gained familiarity with the Slicer GUI, including loading data or ‘scenes’, changing settings such as image contrast, using various in-built modules, etc.
  • I created a brain mask, with ‘Diffusion > Process > Diffusion Brain Masking’
  • I estimated the diffusion tensor, with ‘Diffusion > Process > Diffusion Tensor Estimation’. At right is an image of the diffusion tensor I rendered from the test data (as usual, red: right-left; green: anterior-posterior; blue: inferior-superior).
  • I also took a look at interactive tractography using fiducial seeding, with ‘Diffusion > Tractography > Tractography Seeding’, though I don’t think this is relevant to us.

Next step: At present, I only know how to complete these steps in the GUI, not the terminal. I will work on scripting next week (I believe the lab has something of an existing code base). Also, the test data came pre-processed (and was in keeping with Glasser 2013, i.e., HCP cifty). I need to figure out how to do this pre-processing / if it’s ideal for our analysis / if steps should be different than for TBSS (???).


Tutorial 2/3 [completed]

First, I reviewed the new ORG-800FC-100HCP-minimal-atlas (O’Donnell Research Group, or “ORG”) atlas, described here. As I understand it, this atlas is unique for several reasons: (1) it is tract-based (cf. voxel based); (2) it uses groupwise/spectral clustering to parcellate the whole-brain into 800 fiber parcels and 256 white matter structures, including 58 deep white matter tracts and 198 superficial fiber parcels; (3) these classifications are based on initial computation using FreeSurfer regions (10,000 fibers randomly sampled from each subject’s full tractography for a total of 1 million fibers), followed by expert judgment; (4) the atlas parcellation is based on N=100 individuals (22-36 yo); (5) the atlas was tested on N=584 individuals (1d - 72 yo), spanning HCs and non-HCs with a number of neurological conditions, across different scanners. At left is a visualization of the atlas (all 800 fiber clusters have a unique colour, and similar clusters have similar colours).

Second, using the tutorial’s test data, I performed subject-specific clustering with ‘Diffusion > Tractography > Tractography Display’ enabled by the whitematteranalysis (WMA) package. The image to the right shows subject-specific clustering in the AF, UF, and ILF, which we hypothesized will show FA differences in PNS vs. non-PNS, for one healthy participant chosen at random.

Next step: In contrast to brain masking and fitting the tensor (which I only know how to do in the GUI), I only know how to perform these steps in the terminal. I don’t think this is a problem, but I would like to take a more thorough look at the whitematteranalysis package.


Tutorial 3/3 [incomplete]

Next step: - Complete this tutorial; I started but got stuck partway through.

QC tools

I also briefly reviewed the SlicerDiffusionQC quality-control software cited in Zhang (2018). The software offers QC of DWI data, but within the Slicer environment. It identifies bad gradients by comparing distance of each gradient to a median line, where the median line is obtained from KL divergences between consecutive slices (probabilistic). I used this software on a test participant.

Next step: Do we want to use SlicerDiffusionQC? It’s better than what we’re doing now…

Next steps

  1. Finish 3/3 Slicer training module (this week);
  2. Determine pre-processing pipeline pre-Slicer (dwiprep? consistent with Glasser 2013?);
  3. Run Slicer, using new ORG atlas, on Toronto SPINS data;
  4. As discussed last week, run MRC / ZHH participants through TBSS (without fieldmaps) - sorry, I fell behind on this

Other tasks

  1. 2-hour presentation in Khisti lab (Chapter 7 of the Fortino book, on 3 models of network communication, i.e., shortest path, diffusion, navigation) and respective efficiency indicators).
    [Notes here]
  2. Researched ridge regression for inclusion in stacked model for ABCD contest.
    [Notes here]
  3. Worked ~2 hours for Yuliya.