Measuring Diffusion Properties in Tissue: The Diffusion Tensor & Derived Indices

1ISMRM Weekend course – 5th of May 2012

Measuring Diffusion Properties in Tissue:The Diffusion Tensor & Derived Indices

Gwenaëlle DouaudFMRIB, University of Oxford

2

Historical description of the diffusion tensor

ISMRM Weekend course – 5th of May 2012

• Formalised by Peter Basser and colleagues (1994a; 1994b)

In an isotropic, unconstrained environment:

D = scalar In an anisotropic, constrained environment:

D = 33 definite symmetric positive D = “diffusion tensor”


Historical description of the diffusion tensor

In an anisotropic, constrained environment:

D = 33 definite symmetric positive D = “diffusion tensor”

2

3

1

with i {1,2,3}

4

• First clinical application: brain ischemia (Moseley et al., 1990; Warach et al., 1992)

Diffusion tensor indices: Mean Diffusivity

• Formalised by Peter Basser and Carlo Pierpaoli (1995; 1996)

Magnitude of diffusion


Bammer, 2003

Warach et al., 1995

Visible only 3 hours after onset of stroke

5

• Formalised by Peter Basser and Carlo Pierpaoli (1995; 1996)

RA = “Relative Anisotropy”

Anisotropy of diffusion


Bammer, 2003

Diffusion tensor indices: diffusion anisotropy



• Formal comparison of three anisotropy indices (Papadakis et al., 1999) FA: mapped diffusion anisotropy with the greatest detail and SNR VR: provided the strongest CNR, but with noise contamination, resolution RA: intermediate between FA and VR



• Further comparison between FA and RA (Hasan et al., 2004)


• Developmental study: anatomical and tractography-defined ROI (Lebel et al., 2008)

Fractional Anisotropy and Mean Diffusivity:ROI and voxel-based/TBSS analysis

9



• Developmental study: comparison between ROI and VBD (Snook et al., 2007) The discrepancies related to:

- ROI approach inherently limited- issues with the spatial normalisation for VBD analysis- masking localised changes by averaging all the voxels

within a ROI



• Increasing sensitivity and interpretability of results compared with voxel-based approaches based purely on non-linear registration: TBSS (Smith et al., 2006)

TBSS VBDMean

ControlMean Patient



• Large developmental and ageing study using TBSS (Westlye et al., 2010)


Fractional Anisotropy and Mean Diffusivity:first applications

• Cognitive performance (review in Moseley et al., 2002)

• Development (review in Beaulieu et al., 1999)

• Psychiatric diseases (review in Lim and Helpern, 2002): schizophrenia (Buchsbaum et al., 1998), alcoholism (Pfefferbaum et al., 2000) , depression etc.

• Ageing (review in Moseley, 2002)

• White matter diseases (review in Horsfield and Jones, 2002): multiple sclerosis (Horsfield et al., 1998), Alzheimer’s disease (Hanyu et al., 1998), ALS, CJD etc.


Fractional Anisotropy and Mean Diffusivity:Two complementary measures

• An intriguing finding in multiple sclerosis (Ciccarelli et al., 2001)

Patient

ControlWhite matter

basal ganglia

Increased FA in the basal ganglia in MS: selective Wallerian degeneration, with the intact connections showing an increase of their coherence?


Fractional Anisotropy and Mean Diffusivity:Two complementary measures

• Same findings seen in the basal ganglia in Huntington’s disease (Douaud et al., 2009)


• Using the information from the PDD (Schwartzmann et al., 2005)

2

3

1

Douaud et al., 2009


• Decrease of the dispersion of the PDD in Huntington’s disease (Douaud et al., 2009)


• No difference in the white matter of mild cognitive impairment (MCI) patients using MD or FA (Douaud et al., 2011)

Mode of anisotropy (MO): 3rd moment of the tensor, introduced by Basser (1997), formalised by Ennis and Kindlamnn (2006).

-1Planar, disc-like 1~ 2:

e.g., areas of crossing fibres

+1Linear, cigar-like

2~ 3: e.g., areas of one

predominating fibre population

Beyond Fractional Anisotropy and Mean Diffusivity:the Mode of anisotropy


• Significant difference in the white matter of MCI patients, with an increase of MO

MCI>CON

AD>CON AD>CON

Douaud et al., 2011



• Significant increase of MO in MCI and AD also related to selective degeneration in crossing fibres region (here, centrum semiovale)?

Douaud et al., 2011


20

• Validated in animal models by Song and colleagues (2002; 2005)

= 1

Parallel diffusivity or axial diffusivity: assessing axonal injury


Perpendicular diffusivity or radial diffusivity: assessing myelin injury

Beyond Fractional Anisotropy and Mean Diffusivity:Parallel and Perpendicular Diffusivity


Beyond Fractional Anisotropy and Mean Diffusivity:Parallel and Perpendicular Diffusivity

• Validated in humans using the model of callosotomy (Concha et al., 2006)

Axonal degradation

Myelin degradation


Diffusion tensor indices: in vivo models and histological

validation• Basis of diffusion anisotropy in the brain: comprehensive review by

Beaulieu (2002) Anisotropy due to membrane, not myelin Myelin modulates anisotropy Axonal cytoskeleton does not contribute to anisotropy



validation• Combined in vivo diffusion/histological study: animal model (van Camp et

al., 2012)



validation• Combined in vivo diffusion/histological study: epilepsy (Concha et al.,

2010)


The depressing slide…

• Sensitivity to noise increases with higher order moments:noise(MD) < noise(FA) < noise (MO)

• Crossing-fibres: effects can be averaged out

Groeschel et al., ISMRM 2011




• Crossing-fibres: problematic interpretation of and






• Crossing-fibres: not only influence on FA, but also MD (Vos et al., 2012)


Vos et al., ISMRM 2011





• Crossing-fibres: not only influence on FA, but also MD

• Effect of partial volume effect on FA, MD etc. (Jones, ISMRM 2011)Correction for CSF contamination? (Metzler-Baddeley et al., 2012)

• Choice of sequence: anisotropic voxels (Vos et al., ISMRM 2011)


• Choice of sequence: 12 orientations, 5 b-values: more sensitive to using MD30 orientations, 2 b-values: more sensitive to using FA


…Take home message

• FA and MD complementary to interpret results: higher FA is not necessarily better

• Do not forget about PVE, noise and crossing fibres to help interpret results

• and give additional information (e.g., Salat et al., 2010)

Vos et al., ISMRM 2011

• Choice of sequence: isotropic voxels, optimised for specific question

If ambiguous interpretation: MO, dispersion of PDD, Westin indices (Westin et al., 1997)


…Take home message

• FA and MD complementary to interpret results: higher FA is not necessarily better

• Do not forget about PVE, noise and crossing fibres to help interpret results

• Choice of sequence: isotropic voxels, optimised for specific question

If ambiguous interpretation: MO, dispersion of PDD, Westin indices (Westin et al., 1997)

• and give additional information (e.g., Salat et al., 2010)

• Crossing fibres helps detect subtle differences (Tuch et al., 2005; Douaud et al., 2011)

Tuch et al., 2005

31

FMRIB, University of Oxford

- Karla Miller - Stamatios Sotiropoulos- Timothy Behrens- Steve Smith- Peter Jezzard - Saad Jbabdi

THANK YOU

Special thanks to:

SHFJ, CEA, Orsay

- Cyril Poupon- Yann Cointepas- Denis Riviere- Jean-Francois Mangin



Other diffusion tensor-derived measures

• “Westin” indices describing the tensor shape: sphericity, linearity and planarity (Westin et al., 1997)

• Inter-voxel measures: lattice index (Pierpaoli)

• Fourth moment: kurtosis (Assaf)

Documents

Measuring Diffusion Properties in Tissue: The Diffusion Tensor & Derived Indices