UNC Quantitative DTI Analysis Guido Gerig Isabelle Corouge
UNC: Quantitative DTI Analysis Guido Gerig, Isabelle Corouge Students: Casey Goodlett, Clement Vachet, Matthieu Jomier National Alliance for Medical Image Computing http: //na-mic. org
UNC: Quantitative DTI Analysis • Clinical needs: – Access to fiber tract properties: WM “Integrity” – Fibertract-oriented measurements: Diffusion properties within cross-sections and along bundles – Statistics of diffusion tensors: Beyond FA/ADC • Approaches: – Replace voxel-based by fiber-tract-based analysis – Fiber. Viewer: Set of tools for quantitative fiber tract analysis: Geometry and Diffusion Properties • Clustering, Outlier Detection, Parametrization, Establishing intersubject correspondence – Statistical analysis of DTI National Alliance for Medical Image Computing http: //na-mic. org
Quantitative DTI Analysis Conventional Analysis: ROI or voxel-based group tests after alignment UNC NA-MIC Approach: • Quantitative Analysis of Fiber Tracts • DTI Tensor Statistics across/along fiber bundles • Statistics of tensors Patient Control Tracking/ clustering National Alliance for Medical Image Computing http: //na-mic. org selection FA FA along tract
Processing Tools Fib. Trac: Input DT-MRI, Filtering, Tensor Calc. , FA, ADC, Tractography National Alliance for Medical Image Computing http: //na-mic. org Fiber. Viewer: Clustering, Bundling, Parametrization, Statistics, Visualization
Example: Fiber-tract Measurements Major fiber tracts uncinate fasciculus cingulum uncinate fasciculus FA along cingulate FA along uncinate Corouge, Isabelle, Gouttard, Sylvain and Gerig, Guido, "Towards a Shape Model of White Matter Fiber Bundles using Diffusion Tensor MRI" , Proc. IEEE Computer Society, Int. Symp. on Biomedical Imaging, to appear April 2004 Gerig, Guido, Gouttard, Sylvain and Corouge, Isabelle, "Analysis of Brain White Matter via Fiber Tract Modeling", full paper IEEE Engineering in Medicine Alliance Medical Image Computing and National Biology Society EMBSfor 2004, Sept. 2004 http: //na-mic. org
Processing Steps • Tractography – Data structure for sets of attributed streamlines • Clustering • Parametrization • Diffusion properties across/along bundles • Graph/Text Output • Statistical Analysis National Alliance for Medical Image Computing http: //na-mic. org Ø Slicer (? ) Ø ITK Polyline data structure (J. Jomier) Ø Normalized Cuts (ITK) Ø B-splines (ITK) Ø NEW: DTI stats in nonlinear space (UTAH) Ø Display/Files Ø Biostatistics / ev. DTI hypothesis testing (UTAH)
Concept: Statistics along fiber tracts Origin (anatomical landmark) FA
Accomplished 09/04 – 02/05 Fiber. Viewer Prototype System (ITK) • Clustering (various metrics, normalized graph cut) • Parametrization • FA/ADC/Eigen-value Statistics • Uses Spatial. Objects and Spatial. Object-Viewer • ITK Datastructure for attributed streamlines • Tests in two UNC clinical studies (neonates, autism) • Validation of reproducibility: ISMRM’ 05 National Alliance for Medical Image Computing http: //na-mic. org
ITK Polyline Datastructure National Alliance for Medical Image Computing http: //na-mic. org
3 D Curve Clustering with Normalized Graph Cuts • NGC: Shi and Malik, IEEE 2000 • Set-up of Matrix: Metric: Mean of distances at corresponding points of parametrized curves • Matlab prototype ready, ITK version in development (Casey Goodlett, UNC) Graph Cut National Alliance for Medical Image Computing http: //na-mic. org
3 D Curve Clustering Longitudinal fasciculus 501 streamlines Uncinate fasciculus Clustering can separate neighboring bundles Not possible with region-based processing National Alliance for Medical Image Computing http: //na-mic. org
3 D Curve Clustering seeding Whole longitudinal fasciculus: 2312 streamlines National Alliance for Medical Image Computing http: //na-mic. org 6 clusters
Validation: 6 repeated DTI Registration of ROI Extraction Selection of a ROI T B 01 B 02 Scan 2… Scan 1 T B 01 B 06 Extraction … Scan 6 …Scan 6 Extraction Direct Average of the 6 scans National Alliance for Medical Image Computing DTI Average http: //na-mic. org DTI Average
Tract-based Diffusion Properties Statistics across 6 repeated scans: Curves of Mean. FA and Mean. ADC, with Standard Deviation FA ADC National Alliance for Medical Image Computing http: //na-mic. org FA
Tract-based Diffusion Properties Curves of Mean. FA/ Mean. ADC in comparison to the Average DTI FA ADC National Alliance for Medical Image Computing http: //na-mic. org FA
Work in Progress: Statistics of Tensors (UTAH & UNC) • Statistics of DTI requires new math and tools • Linear Statistics does not preserve positive-definit. • Tom Fletcher UNC Ph. D 2004 (w. Joshi/Pizer), now UTAH – Riemannian symmetric (nonlinear) space – New similarity measure – Method for interpolation of tensors National Alliance for Medical Image Computing http: //na-mic. org
we all like to pick the highlights, who picks the “dirty reality” problems? ? • Papers: “Bad slices were eliminated from processing” • But: +12 dir/ +4 averages / +25 slices: 1200 images? ? National Alliance for Medical Image Computing http: //na-mic. org
we all like to pick the highlights, who picks the “dirty reality” problems? ? • UNC Solution: ITK DTIchecker (Matthieu Jomier) • Automatic screening for intensity artifacts, motion artifacs, missing/corrupted slices • Writes report / Script file National Alliance for Medical Image Computing http: //na-mic. org
we all like to pick the highlights, who picks the “dirty reality” problems? ? • • • Lucas MRI and MRS Center, Stanford University, CA : Spin echo EPI dti_epi Pulsed Gradient/Stejskal-Tanner diffusion weighting UNC uses Stanford Bammer/Mosley “tensorcalc” software for DTI processing Eddy Current Distortion Correction (here 23 directions) Tensorcalc (“T 1”) DWI/DTI recon toolbox with powerful built-in image registration tools. http: //rsl. stanford. edu/research/software. html / http: //www-radiology. stanford. edu/majh/ http: //snarp. stanford. edu/dwi/maj/ The diffusion weighted images are unwarped using the method described in de Crespigny, A. J. and Moseley, M. E. : "Eddy Current Induced Image Warping in Diffusion Weighted EPI", Proc , ISMRM 6 th Meeting, Sydney 661 (1998) and Haselgrove, J. C. and Moore, J. R. , "Correction for distortion of echoplanar images used to calculate the apparent diffusion coefficient", MRM 1996, 36: 960 -964 ( Medline citation). National Alliance for Medical Image Computing http: //na-mic. org
Next 6 months • Methodology Development: – – DTI tensor statistics: close collab. with UTAH Deliver ITK tools for clustering/parametrization to Core 2 Feasibility tests with tractography from Slicer Deliver Fiber. Viewer prototype platform to Core 2 to discuss integration into Slicer • Clinical Study: DTI data from Core 3 – Check feasibility of tract-based analysis w. r. t. DTI resolution (isotropic voxels(? )), SNR – Apply procedure to measure properties of: • Cingulate (replicate ROI findings of Shenton/Kubiki) • Uncinate fasciculus (replicate ROI findings) • Dartmouth 3 mm DTI data National Alliance for Medical Image Computing http: //na-mic. org
NA-MIC DTI Processing Needs • Generic DTI reconstruction – – Arbitrary #directions Artifact checking/removal Eddy-current distortion correction Tensor calculation • Tensor Filtering (nonlinear, geodesic space) • Tensor interpolation, linear- and nonlinear registration • Tensor+ reconstruction/representation (DSI) • Standards for datastructures (DTI, tensors, streamlines, diffusion-gradient-file) National Alliance for Medical Image Computing http: //na-mic. org
Local shape properties of wm tracts • Geometric characterization of fiber bundles • Local shape descriptors: curvature and torsion Adults Neonate Max. curvature positions: Possible candidates for curve matching National Alliance for Medical Image Computing http: //na-mic. org
- Slides: 22