Introduction to Diffusion MRI processing The diffusion process

  • Slides: 41
Download presentation
Introduction to Diffusion MRI processing

Introduction to Diffusion MRI processing

The diffusion process http: //pubs. niaaa. nih. gov/publications/arh 27 -2/146 -152. htm 2

The diffusion process http: //pubs. niaaa. nih. gov/publications/arh 27 -2/146 -152. htm 2

dt_recon • Required Arguments: • --i invol • --s subjectid • --o outputdir •

dt_recon • Required Arguments: • --i invol • --s subjectid • --o outputdir • Example: dt_recon --i 6 -1025. dcm --s M 111 --o dti 3

Main processing steps • # Eddy current and motion correction – (FSL eddy_correct) •

Main processing steps • # Eddy current and motion correction – (FSL eddy_correct) • # Tensor fitting – tensor. nii, eigvals. nii. eigvec? . nii – set of scalar maps: adc, fa, ra, vr, ivc • # Registration to anatomical space – (bbregister to lowb) • # Mapping mask, FA to Talairach space 4

Other Arguments (Optional) --b bvals bvecs --info-dump infodump. dat use info dump created by

Other Arguments (Optional) --b bvals bvecs --info-dump infodump. dat use info dump created by unpacksdcmdir or dcmunpack --ecref TP use TP as 0 -based reference time points for EC --no-ec turn off eddy/motion correction --no-reg do not register to subject or resample to talairach --no-tal do not resample FA to talairch space --sd subjectsdir specify subjects dir (default env SUBJECTS_DIR) --eres-save resdidual error (dwires and eres) --pca run PCA/SVD analysis on eres (saves in pca-eres dir) --prune_thr set threshold for masking (default is FLT_MIN) --debug print out lots of info --version print version of this script and exit --help voluminous bits of wisdom 5

Examples of scalar maps • FA: fractional anisotropy (fiber density, axonal diameter, myelination in

Examples of scalar maps • FA: fractional anisotropy (fiber density, axonal diameter, myelination in WM) • RA: relative anisotropy • VR: volume ratio • IVC: inter-voxel correlation (diffusion orientation agreement in neighbors) • ADC: apparent diffusion coefficient (magnitude of diffusion; low value organized tracts) • RD: radial diffusivity • AD: axial diffusivity • … 6

FA 7

FA 7

ADC 8

ADC 8

IVC 9

IVC 9

Tractography examples • Trackvis and Diffusion Toolkit (http: //www. trackvis. org/) 10

Tractography examples • Trackvis and Diffusion Toolkit (http: //www. trackvis. org/) 10

11

11

CST on (color) FA map 12

CST on (color) FA map 12

Under development: TRActs Constrained by Under. Lying Anatomy (TRACULA) Anastasia Yendiki HMS/MGH/MIT Athinoula A.

Under development: TRActs Constrained by Under. Lying Anatomy (TRACULA) Anastasia Yendiki HMS/MGH/MIT Athinoula A. Martinos Center for Biomedical Imaging 13

Tractography • Identify fiber bundles in cerebral white matter (WM) • Characterizing these WM

Tractography • Identify fiber bundles in cerebral white matter (WM) • Characterizing these WM pathways is important for: – Inferring connections b/w brain regions – Understanding effects of neurodegenerative diseases, stroke, aging, development … From Gray's Anatomy: IX. Neurology 14

Diffusion in brain tissue • Differentiate tissues based on the diffusion (random motion) of

Diffusion in brain tissue • Differentiate tissues based on the diffusion (random motion) of water molecules within them • Gray matter: Diffusion is unrestricted isotropic • White matter: Diffusion is restricted anisotropic 15

Diffusion MRI • Magnetic resonance imaging can provide “diffusion encoding” • Magnetic field strength

Diffusion MRI • Magnetic resonance imaging can provide “diffusion encoding” • Magnetic field strength is varied by gradients in different directions • Image intensity is attenuated depending on water diffusion in each direction • Compare with baseline images to infer on diffusion process Diffusion encoding in direction g 1 g 2 g 3 g 4 g 5 g 6 No diffusion encoding 16

Deterministic vs. probabilistic • Determine “best” pathway between two brain regions • Challenges: -

Deterministic vs. probabilistic • Determine “best” pathway between two brain regions • Challenges: - Noisy, distorted images - Pathway crossings - High-dimensional space • Deterministic methods: Model geometry of diffusion data, e. g. , tensor/eigenvectors [Conturo ‘ 99, Jones ‘ 99, Mori ‘ 99, Basser ‘ 00, Catani ‘ 02, Parker ‘ 02, O’Donnell ‘ 02, Lazar ‘ 03, Jackowski ‘ 04, Pichon ‘ 05, Fletcher ‘ 07, Melonakos ‘ 07, …] ? • Probabilistic methods: Also model statistics of diffusion data [Behrens ‘ 03, Hagmann ‘ 03, Pajevic ‘ 03, Jones ‘ 05, Lazar ‘ 05, Parker ‘ 05, Friman ‘ 06, Jbabdi ‘ 07, …] 17

Local vs. global • Local: Uses local information to determine next step, errors propagate

Local vs. global • Local: Uses local information to determine next step, errors propagate from areas of high uncertainty • Global: Integrates information along the entire path 18

Local tractography • Define a “seed” voxel or ROI to start the tract from

Local tractography • Define a “seed” voxel or ROI to start the tract from • Trace the tract by small steps, determine “best” direction at each step • Deterministic: Only one possible direction at each step • Probabilistic: Many possible directions at each step (because of noise), some more likely than others 19

Some issues • Not constrained to a connection of the seed to a target

Some issues • Not constrained to a connection of the seed to a target region • How do we isolate a specific connection? We can set a threshold, but how? • What if we want a nondominant connection? We can define waypoints, but there’s no guarantee that we’ll reach them. • Not symmetric between tract “start” and “end” point 20

Global tractography • Define a “seed” voxel or ROI • Define a “target” voxel

Global tractography • Define a “seed” voxel or ROI • Define a “target” voxel or ROI • Deterministic: Only one possible path • Probabilistic: Many possible paths, find their probability distribution • Constrained to a specific connection • Symmetric between seed and target regions 21

Probabilistic tractography Have set of images Want most probable path … • Determine the

Probabilistic tractography Have set of images Want most probable path … • Determine the most probable path based on: – What the images tell us about the path Assume a multi-compartment model of diffusion [Jbabdi et al. , Neuro. Image ‘ 07] – What we already know about the path Incorporate prior knowledge on path anatomy from training subjects 22

Multi-compartment model Behrens et al. , MRM ‘ 03 Jbabdi et al. , Neuro.

Multi-compartment model Behrens et al. , MRM ‘ 03 Jbabdi et al. , Neuro. Image ‘ 07 • Multiple diffusion compartments in each voxel: – Anisotropic compartments that model fibers (1, 2, …) 0 2 – One isotropic compartment that models everything left over (0) • We infer from the data: – Orientation angles of anisotropic compartments – Volumes of all compartments – Overall diffusivity in the voxel • Multiple fibers only if they are supported by data 1 23

Anatomical priors for WM paths • WM pathways are well-constrained by surrounding anatomy •

Anatomical priors for WM paths • WM pathways are well-constrained by surrounding anatomy • Sources of prior anatomical information: – Shape of the path in a set of training subjects – Anatomical regions around the path in the training subjects • Other types of anatomical constraints often used: – WM masks – Constraints on path angle – Constraints on path length 24

TRACULA • TRActs Constrained by Under. Lying Anatomy • Global probabilistic tractography • Prior

TRACULA • TRActs Constrained by Under. Lying Anatomy • Global probabilistic tractography • Prior info on tract anatomy from training subjects – No manual intervention in new subjects – Robustness w. r. t. initialization and ROI selection – Anatomically plausible solutions • Manual labeling of paths on a set of training subjects, performed by an expert • Anatomical segmentation maps of the training subjects, produced by Free. Surfer 25

Preliminary results Data courtesy of Dr. R. Gollub, MGH • Manual labeling of: –

Preliminary results Data courtesy of Dr. R. Gollub, MGH • Manual labeling of: – Corticospinal tract (CST) – Superior longitudinal fasciculus (SLF) 1, 2, 3 – Cingulum • DTI reliability data set from Mental Illness and Neuroscience Discovery (MIND) Institute – 10 healthy volunteers scanned twice – DWI: 2 x 2 x 2 mm resolution, 60 gradient directions – T 1: 1 x 1 x 1 mm resolution • Use manual labeling of 9 subjects to obtain path priors and path initialization for 10 th subject 26

Reliability study Manual labeling by Allison Stevens and Cibu Thomas Visualization tool by Ruopeng

Reliability study Manual labeling by Allison Stevens and Cibu Thomas Visualization tool by Ruopeng Wang CST SLF 27

Test-retest reliability No info from training subjects With info from training subjects Visit 1

Test-retest reliability No info from training subjects With info from training subjects Visit 1 Visit 2 28

Application: Huntington’s disease Data courtesy of Dr. D. Rosas, MGH Healthy Huntington’s stage 1

Application: Huntington’s disease Data courtesy of Dr. D. Rosas, MGH Healthy Huntington’s stage 1 Huntington’s stage 2 Huntington’s stage 3 29

MD changes in patients CST SLF 1 SLF 2 SLF 3 0. 1 Cingulum

MD changes in patients CST SLF 1 SLF 2 SLF 3 0. 1 Cingulum 0. 001 P-values for T-test on mean MD of Huntington’s patients (N=33) and controls (N=22) 30

Correlation with disease stage Left CST Right SLF 1 SLF 2 SLF 3 FA

Correlation with disease stage Left CST Right SLF 1 SLF 2 SLF 3 FA -. 3 MD. 3 . 4 -. 3 CB -. 3 SLF 1 SLF 2 SLF 3 -. 5 CB -. 3 -. 2 . 7. 6. 4 p<10 . 5 -7 -5 . 7. 6. 3 p<10 -8 -5 -. 2 RD. 3 . 4 . 6 . 5 . 4 . 6 . 6 . 3 AD. 3 . 4 . 7 . 6 . 4 . 8 . 5 . 3 FA: Fractional anisotropy MD: Mean diffusivity RD: Radial diffusivity AD: Axial diffusivity CST: Corticospinal tract SLF: Superior longitudinal fasciculus CB: Cingulum body 31

Application: Schizophrenia Data courtesy of Dr. R. Gollub, MGH CST SLF 1 SLF 2

Application: Schizophrenia Data courtesy of Dr. R. Gollub, MGH CST SLF 1 SLF 2 SLF 3 Cingulum 0. 1 0. 001 P-values for T-test on mean RD of schizophrenia patients (N=25) and controls (N=18) 32

FA and RD changes * * * ° * p<. 05 ° p<. 10

FA and RD changes * * * ° * p<. 05 ° p<. 10 ng ul um ° ig ht ci ul ng ci * R Le ft * um * 33

Current development • TRACULA: A method for diffusion tractography that combines a global probabilistic

Current development • TRACULA: A method for diffusion tractography that combines a global probabilistic approach with prior knowledge on path anatomy • More detailed models of tracts • Improved inter-subject registration • Coming soon to a Free. Surfer near you! 34

Acknowledgements Support provided in part by: • National Center for Research Resources – P

Acknowledgements Support provided in part by: • National Center for Research Resources – P 41 RR 14075 – R 01 RR 16594 – The NCRR BIRN Morphometric Project BIRN 002, U 24 RR 021382 • National Institute for Biomedical Imaging and Bioengineering – K 99 EB 008129 – R 01 EB 001550 – R 01 EB 006758 • National Institute for Neurological Disorders and Stroke – R 01 NS 052585 • Mental Illness and Neuroscience Discovery (MIND) Institute • National Alliance for Medical Image Computing – Funded by the NIH Roadmap for Medical Research, grant U 54 EB 005149 35

Acknowledgements MGH/Martinos Lilla Zöllei Allison Stevens David Salat Bruce Fischl & Jean Augustinack Oxford/FMRIB

Acknowledgements MGH/Martinos Lilla Zöllei Allison Stevens David Salat Bruce Fischl & Jean Augustinack Oxford/FMRIB Saad Jbabdi Tim Behrens 36

ONGOING: Registration of tractography • Goal: fiber bundle alignment • Study: compare CVS to

ONGOING: Registration of tractography • Goal: fiber bundle alignment • Study: compare CVS to methods directly aligning DWI-derived scalar volumes • Conclusion: high accuracy cross-subject registration based on structural MRI images can provide improved alignment • Zöllei, Stevens, Huber, Kakunoori, Fischl: “Improved Tractography Alignment Using Combined Volumetric and Surface Registration”, accepted to Neuro. Image 37

Mean Hausdorff distance measures for three fiber bundles CST ILF UNCINATE 38

Mean Hausdorff distance measures for three fiber bundles CST ILF UNCINATE 38

Average tracts after registration mapped to the template displayed with iso-surfaces FLIRT FA-FNIRT CVS

Average tracts after registration mapped to the template displayed with iso-surfaces FLIRT FA-FNIRT CVS 39

Stages: • 1. Convert dicom to nifti (creates dwi. nii) • 2. Eddy current

Stages: • 1. Convert dicom to nifti (creates dwi. nii) • 2. Eddy current and motion correction using FSLs eddy_correct, • creates dwi-ec. nii. Can take 1 -2 hours. • 3. DTI GLM Fit and tensor construction. Includes creation of: • tensor. nii -- maps of the tensor (9 frames) • eigvals. nii -- maps of the eigenvalues • eigvec? . nii -- maps of the eigenvectors • adc. nii -- apparent diffusion coefficient • fa. nii -- fractional anisotropy • ra. nii -- relative anisotropy • vr. nii -- volume ratio • ivc. nii -- intervoxel correlation • lowb. nii -- Low B • bvals. dat -- bvalues • bvecs. dat -- directions • Also creates glm-related images: • beta. nii - regression coefficients • eres. nii - residual error (log of dwi intensity) • rvar. nii - residual variance (log) • rstd. nii - residual stddev (log) • dwires. nii - residual error (dwi intensity) • dwirvar. nii - residual variance (dwi intensity) • 4. Registration of lowb to same-subject anatomical using • FSLs flirt (creates mask. nii and register. dat) • 5. Map FA to talairach space (creates fa-tal. nii) • Example usage: • dt_recon --i 6 -1025. dcm --s M 87102113 --o dti 40

After dt_recon • • • # Check registration tkregister 2 --mov dti/lowb. nii --reg

After dt_recon • • • # Check registration tkregister 2 --mov dti/lowb. nii --reg dti/register. dat --surf orig --tag • • • # View FA on the subject's anat: tkmedit M 87102113 orig. mgz -overlay dti/fa. nii -overlay-reg dti/register. dat • • # View FA on fsaverage tkmedit fsaverage orig. mgz -overlay dti/fa-tal. nii • • • # Group/Higher level GLM analysis: # Concatenate fa from individuals into one file # Make sure the order agrees with the fsgd below mri_concat */fa-tal. nii --o group-fa-tal. nii # Create a mask: mri_concat */mask-tal. nii --o group-masksum-tal. nii --mean mri_binarize --i group-masksum-tal. nii --min. 999 --o group-mask-tal. nii # GLM Fit mri_glmfit --y group-fa-tal. nii --mask group-mask-tal. nii --fsgd your. fsgd --C contrast --glm groupanadir 41