1 03 - tensor calculus 03 - tensor calculus - tensor analysis.
Diffusion Weighted Imaging Tensor Analysis Vincent A. Magnotta Associate Professor March 21, 2011.
-
Upload
norman-nash -
Category
Documents
-
view
220 -
download
0
Transcript of Diffusion Weighted Imaging Tensor Analysis Vincent A. Magnotta Associate Professor March 21, 2011.
Diffusion Weighted ImagingTensor Analysis
Vincent A. MagnottaAssociate Professor
March 21, 2011
Diffusion Tensor Analysis Flow Chart
DTIData
Collection(DICOM)
ImagesFormat
Conversion
GenerationOf Diffusion
Tensor
RigidCo-RegisterWith AC-PCAligned T1
Non-RigidCo-RegisterWith AC-PCAligned T1
CreateDiffusion
ScalarImages
ResampleImages
Into ACPCSpace
Concatenate Data
DTIPrep
1. Verify Acquisition2. Artifact Detection3. Motion Correction4. Update Gradient
Directions5. Remove Bad Data
ExtractB0
Image
Diffusion Tensor Image Analysis• Image format conversion
– Change from DICOM to Nifti or NRRD image formats– Rotate applied diffusion gradients
• Motion Correction– Account for patient motion and eddy current artifacts
• Generation of Diffusion Tensor– Includes possible edge preserving low pass spatial filtering– Use rotated diffusion directions
• Create Diffusion Tensor scalar maps– Mean diffusivity– Fractional Anisotropy– Relative anisotropy– Radial Diffusivity– Axial Diffusivity
• Co-register with anatomical image– Rigid– Non-Rigid (B-Spline)
Image Format Conversion• Convert from DICOM to NRRD format
– Nearly Raw Raster Data– Defines origin, spacing, orientation, Diffusion Gradients,
and Measurement Frame– Coordinate frame for the applied diffusion gradients
• All information obtained from DICOM header– Siemens, Philips, and GE scanners
DicomToNrrdConverter \ --inputDicomDirectory /home/vince/images/dti_images \ --outputDirectory /home/vince/images \ --outputVolume /home/vince/images/SUBJECT_DWI.nhdr
DTI Concatenation
• Concatenate multiple DTI runs together– Improve SNR of tensor estimation– Runs can contain any number of gradient
directions and orientations
gtractConcatDwi --outputVolume dti.nhdr \ --inputVolume dti_parta.nhdr,dti_partb.nhdr
Artifacts
Improving DTI measures: DTIPrep• From UNC, Zhexing Liu• Purpose of DTIPrep: provide individual and group quality
control of DWI/DTI data sets in GUI and command line mode– Detect and remove artifacts that often appear in DWI data– Prevent artifacts from creating DTI estimation errors in tensor
principle orientation (premature fiber tracking termination) and scalars
– Prevent low consistency in quality control associated with current visual checking of DWI data sets
DTIPrep: Quality Control Pipeline
• Image information checking• Diffusion information checking• Slice-wise intensity artifact checking• Interlace-wise venetian blind artifact checking• Baseline averaging• Eddy-current and head motion artifact correction• Gradient-wise checking (motion artifact checking)
DTIPrep: Quality Control Pipeline
• Image information checking– Image space– Image directions– Image size– Image spacing– Image origin– Cropping
• Diffusion information checking– b value– Diffusion gradient
vectors– Tolerance tests– Replacement of diffusion
gradient vectors with those in acquisition protocol
DTIPrep: Quality Control Pipeline
• Venetian blind artifact detection• Baseline averaging
– Motion between baseline scans is removed by rigidly registering all baseline scans and averaging them together
– The averaged baseline image is used as a reference for subsequent eddy-current and head motion artifact correction for all gradients
• Eddy-current and head motion artifacts correction• Resulting image is SUBJECT_DWI_Qced.nhdr
DTIPrep –DWINrrdFile /home/vince/images/SUBJECT_DWI.nhdr \ --xmlProtocol /home/vince/images/default.xml \ --default --check --outputFolder /home/vince/images
DTIPrep Outputs• NRRD file containing
– Single baseline average image (motion corrected)– Corrected Diffusion gradients
• Passed quality control (slice-wise & interlace)• Head motion corrected (Rigid register to baseline with gradient direction adjustments
relative to anatomical frame of reference)• Eddy current corrected (Affine register to baseline)
– SUBJECT_DWI_Qced.nhdr• Report on excluded diffusion gradients
– SUBJECT_DWI_QcReport.txt
• Optional outputs: NRRD files of excluded diffusion gradients from each quality control step
• DTIPrep outputs GTRACT
DTIPrep GUI
DTIPrep: Quality Control Pipeline2.4 Slice-wise intensity related artifacts checking
We propose to use Normalized Correlation (NC) between successive slices across all the diffusion gradients for screening the intensity related artifacts.
Analysis region
Slice number
Slic
e-to
-slic
e co
rrel
ation
val
ue
DTIPrep: Quality Control Pipeline2.5 Interlace-wise Venetian blind artifact checking
Venetian blind like artifacts can be detected via correlations and motion parameters between the interleaved parts for each gradient volume.
Tran
slati
on (m
m),
Angl
e of
rota
tion
(deg
rees
)
Gradient number
DTIPrep: Quality Control Pipeline2.8 Gradient-wise checking
Motion artifact residuals after eddy-current and head motion corrections can be detected via motion parameters between baseline and each of the gradients.
DTIPrep Impact on FA values
• Exclusion of optimal number of gradients minimized the standard deviation in FA values– Standard deviation was lowered in a single scan processed by
DTIPrep
Without DTIPrep With DTIPrep
Stan
dard
dev
iatio
n in
FA
Stan
dard
dev
iatio
n in
FA
9% 21% 26% 27%
Create Diffusion Tensor
• Create Tensor representation of diffusion process– Defined by 6 unique parameters– Allows for edge preserving low pass filtering (median)
whose radius is defined in voxels– Removal of background signal
gtractTensor \ --inputVolume Subject_DTIPREP.nhdr \ --outputVolume SUBJECT_Tensor.nhdr \ --medianFilterSize 1,1,1 --backgroundSuppressingThreshold 50 --b0Index 0
Rotationally Invariant Scalar Generation
• Eigen analysis of tensor data• Creates a variety of scalars:
– FA – Fractional Anisotropy– MD – Mean Diffusivity– RA – Relative Anisotropy– LI – Lattice Index– AD – Axial Diffusivity– RD – Radial Diffusivity
gtractAnisotropyMap \ --inputTensorVolume Subject_Tensor.nhdr \ --outputVolume SUBJECT_FA.nii.gz \ --anisotropyType FA
Image Extraction and Clipping
• Extract B0 image• Clip B0 image to remove skull using AFNI
extractNrrdVectorIndex --index 0\ --inputVolume Subject_DTIPREP.nhdr \ --outputVolume Subject_B0.nii.gz
3dAutomask -prefix Subject_DWI_B0_mask.nii.gz \ Subject_B0.nii.gz3dcalc -a Subject_DWI_B0_mask.nii.gz \ -datum short -expr "a*1" \ -prefix Subject_B0_maskShort.nii.gz3dcalc -a Subject_B0_maskShort.nii.gz \ -b Subject_DWI_B0.nii.gz -expr "a*b" \ -prefix Subject_DWI_B0_Brain.nii.gz
DWI to Anatomical Registration
• Utilize BRAINSFit image registration– Supports Mutual Information registration metric
• Non-linear image registration– B-splines can be used to correct for susceptibility
artifacts– Eliminates the need for field maps
BRAINSFit –movingVolume Subject_DWI_B0_Brain.nii.gz \ --fixedVolume Subject_clippedT1.nii.gz\ --transformType Rigid,BSpline \ --numberOfSamples 500000 \ --splineGridSize 12,12,12 \ --outputTransform SUBJECT_ACPC.mat \ --initializeTransformMode useMomentsAlign
Invert Transform
• Provide a mapping from AC-PC apace back to the DTI space
• Approximate inverse is computed using Thin Plate Spline (TPS) transforms
• Used to map ROIs into DTI space for fiber tracking
gtractInvertBSplineTransform \ --inputTransform SUBJECT_ACPC.mat \ --outputTransform SUBJECT_ACPC_Inverse.mat \ --inputReferenceVolume Subject_clippedT1.nii.gz
Resample DTI Scalars
• Place rotationally invariant scalars into the space of anatomical images
• Resample B0 image to check quality of registration
BRAINSResample \ --referenceVolume SUBJECT_T1.nii.gz \ --inputVolume SUBJECT_FA.nii.gz \ --warpTransform SUBJECT_ACPC.mat \ --outputVolume SUBJECT_FA_ACPC.nii.gz --interpolationMode Linear
Diffusion Tensor Scalar Measurements
• Lobar Talairach Analysis– Frontal, Temporal, Parietal, Occipital, and Cerebellar white
matter measurements– White matter region defined using both FA and tissue
classified images– BRAINS measurement script exists
Diffusion Tensor Scalar Measurements A-P
• Analysis of Anisotropy from Anterior-Posterior based on Talairach Atlas– Divide regions from A-D
and F-I in half– Retain sizes of E1, E2 and
E3 – BRAINS script exists
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
A A.5 B B.5 C C.5 D D.5 E1-2 E2-3 E3-3 F F.5 G G.5 H H.5 I I.5
Anterior to Posterior
Fra
ctio
nal A
niso
trop
y
Patients
Controls
P values
0.05 line
SPM Analysis
• Co-register to Atlas image– Apply transform to DTI
scalar image– Smooth Scalar images– Threshold to White
matter regions– Possible issues with
anatomic variability
Fiber Tracking - IntroductionBase on the directional information provided by DTI, fiber tracking can be used to explore the underlying white matter fiber structure non-invasively