Diffusion Models and Metrics

Before practicum on Friday, please complete following:

• Read [5. Fiber resolving methods]
  • 5.1. Model-based methods
  • 5.2. Model-free methods
  • 5.3. Spherical deconvolution
  • 5.4. Comparison
• Watch a video about diffusion tensor imaging

During practicum on Friday:

Overview

Review: Diffusion MRI Preprocessing (10 min)

Diffusion MRI models (10 min)

• Diffusion Models
  • Model-based methods (DTI, FSL’s BSM, DKI, NODDI)
  • Model-free methods (DSI Studio’s GQI)
  • Spherical deconvolution methods (MRtrix’s CSD, MSMT-CSD)

Model/Method	Full Name	Tool	Publications	B-table requirements	Handle Free Water	Quantify Restricted Diffusion	Resolve Multiple Fibers	Metrics
Model-Based
DTI (Basser, 1994)	Diffusion tensor imaging		353,000	B0, >= 1 b-value(s)	No (except for DTI-FWE)	No	No (except for Multi-Tensor)	fa, ad, rd, md
DKI (Jensen, 2005)	Diffusion kurtosis imaging	DKE	6,400	B0, >= 2 b-values	No	Yes	No (except for Multi-Tensor)	ak, rk, mk
NODDI (Zhang, 2012)	Neurite Orientation Dispersion and Density Imaging	NODDI Matlab Toolbox	4,680	B0, >= 2 b-values	Yes	Yes	No (except for Multi-Fiber NODDI )	iso, odi, ndi (icvf)
BSM (Behrens,2003)	Ball & Stick Model	BEDPOSTX,FSL	1,830	B0, >= 1 b-value(s)	Yes	Yes	Yes	qa, iso, rdi
Model-Free (Q-Space Imaging)
DSI (Wedeen, 2005)	Diffusion Spectrum Imaging	TrackVis	5,480	B0, >10 b-values on grid	No	No	Yes	-
QBI (Tuch,2004)	Q-ball Imaging	TrackVis	3,460	B0, 1 b-value	No	No	Yes	gfa
GQI (Yeh, 2010),QSDR (Yeh,2011)	Generalized Q-sampling Imaging, Q-Space Diffeomorphic Reconstruction	DSI Studio	860, 452	B0, >= 1 b-value(s)	Yes	Yes	Yes	qa, iso, rdi
Spherical Deconvolution
CSD (Tournier,2007)	Constrained Spherical Deconvolution	MRtrix3, DIPY	4,010	1 b-value	No	No	Yes	afd
MSMT-CSD (Jeurissen,2014)	Multi-Shell, Multi-Tissue CSD	MRtrix3, DIPY	229	>= 2 b-values	Yes	Yes	Yes	afd

Personal recommendation:

for deriving voxel-wise metrics

• always include DTI metrics: (1) nearly all acquisition can fit a tensor model. (2) ITS metrics and related biophysics well known. (3) if a simpler method works, no need for more complicated one, unless the the purpose is getting better sensitivity/specificity.
• DKI, NODDI, GQI, and CSD provide more advanced metrics. BSM, DSI, QBI are usually used for fiber tracking.
• avoid using streamline counts as the metrics.

for mapping white matter tracts

• Both MSMT-CSD and CSD provides fiber orientation distribution (FOD). MSMT-CSD is more reliable than CSD.
• Both GQI, QBI, and DSI provides diffusion orientation distribution function (dODF). GQI can use all DWI acquisitions, whereas QBI and DSI have their specific acquisition requirements.
• GQI provides more conservative results, whereas CSD provides better sensitivity to crossing/branching fibers.

Hands-on: DTI, GQI, and QSDR Reconstruction (10 min)

• Download control subject 1 session 1 data from the SCA2 Diffusion Tensor Imaging study
• Reconstruction using DTI, GQI, QSDR
• Check and compare FIB files
• Batch SRC reconstruction

Metrics (10 min)

P. Mukherjee, J.I. Berman, S.W. Chung, C.P. Hess and R.G. Henry American Journal of Neuroradiology April 2008, 29 (4) 632-641; DOI: https://doi.org/10.3174/ajnr.A1051

• Microscopic metrics: anisotropy, diffusivities, and other voxel-based metrics
  • Anisotropy: axonal density & myelination
  • Isotropy: free water (not restricted) , cellularity (restricted)
  • Interpretation
• Macroscopic metrics: shape metrics

• Graph-based metrics: network measures

Hands-on: dMRI analysis (15 min)

• Region-based analysis
  • Open a FIB file at Step T3
  • Use [Region][Statistics] to get diffusion metrics from atlases (Documentation).
• Tract-based analysis
  • Open a FIB file at Step T3 and map a white matter pathway (e.g. use autotrack)
  • Get diffusion and shape metrics at CST using [Tracts][Statistics]
  • Plot along CST metrics at z-direction or fiber direction. (Documentation)
• Insert other modalities

• Group-wise region/tract analysis using connectometry database
  • Reconstruct all SRC files using QSDR
  • Construct a database: samples
  • Region-based analysis
  • Tract-based analysis

---

Practicum assignment: ROI-based and Track-based analysis comparing SCA2 with controls

1. Download SCA2 eddy-processed SRC Files. Ignore those in the follow-up folder.
2. Reconstruct data at [Step T2 Reconstruction] using QSDR
3. Create connectometry databases using [Step C1: Create a connectometry Database] and select dti_fa as the metric. The created database can be checked using [Step C2].
4. Open the database at [Step T3: Fiber Tracking]
5. Use region-based analysis to get fa at pons and compare between patients and controls.
6. Use tract-based analysis to get fa at CST and compare between patients and controls.
7. Cut CST into multiple segments and compare them between patients and controls.