3.2 Methods
3.2.4 Image analysis framework
Raw DWI images were processed and DTI tractography analysis was performed using the UNC−Utah NA−MIC Framework for DTI Fiber Tract Analysis Verde2014, which is schematized in Figure 3.1. In brief, DWI images first underwent rigorous qual- ity control, eddy current, and motion correction with DTIPrep2 [102; 128]. We next
evaluated DTI signal to noise and tractography feasibility using 3D Slicer [127]. Here it was discovered that the right−left glyph direction was flipped, and image measurement
1 2 3 4
Final DTI Atlas
Quality Control Fiber Tractography
Fiber Cleaning &
Parameterization Diffusion Parameter Profiles Along Tract Statistical Analysis
Plot Statistics on Tract
−40 −30 −20 −10 0 10 20 30 40 50 −0.03 −0.02 −0.01 0 0.01 0.02 0.03 arclength FA
uncL FA GroupAgeSex Estimated Coefficient
Group Age Sex
Figure 3.1: UNC−Utah NA−MIC framework for DTI fiber tract analysis. The frame- work begins with rigorous quality control of individual participant data inspecting for DWI artifacts (1), DTI direction encoding (2), glyph orientation (3), and landmark (fiducial) tractography. Together these steps correct common DWI artifacts, assess data quality, and tractography feasibility. Quality controlled DTI from each partici- pant is then used to create a study specific DTI atlas via linear and nonlinear regis- tration. On this atlas, fiber bundle tractography is performed. Fiber tracts are then cleaned and parameterized so each location along the tract has a relative address. Then diffusion parameters at each address along the tract are extracted from native space for individual participants, creating diffusion parameter profiles for each tract. These profiles are then entered into statistical analysis. Resulting p−values and/or betas can then be visualized on the fiber bundle.
frames were edited in the original DWI header file to correct this directionality for all participants. Then DWI and DTI quality control was repeated with the correct mea- surement frame. Brain masks were created from the b=0 (b0) images and the isotropic
diffusion weighted (IDWI) images with AutoSeg3 [129], and edited in ITK−SNAP4
[130]. A study specific atlas was then built using linear, unbiased diffeomorphic, and symmetric diffeomorphic registrations of skull−stripped DTIs as detailed in [84]. We next performed interactive fiber tractography for the bilateral cingulum, fornix, and
3http://www.nitrc.org/projects/autoseg/
Fornix Uncinate Body Crus Fimbria Columns Temporal Frontal Cingulum Body Posterior Anterior A B C D Posterior Frontal
Figure 3.2: Fiber tracts analyzed and anatomical definitions. A. Fiber bundles of interest are represented in their anatomical location in a glass brain, with the cingulum in blue, the fornix in pink, and the uncinate in green. B−D display the individual fiber bundles with locations along the tract labeled to clarify findings. These fibers are colored by the parameterized addresses along each tract (arclength). B. Cingulum, C. Fornix, and D. Uncinate.
uncinate on the Final DTI Atlas in 3D Slicer. See Figure 3.2 for anatomical orienta- tion of each tract investigated. We executed gross fiber bundle cleaning in 3D Slicer utilizing ROIs, followed by fine scale clustering based processing in FiberViewerLight5
. In order to analyze only the properties of the main fiber bundle, cortical projections were removed from the cingulum bundle using the cutter option in FiberViewerLight. A fiber origin plane was then manually placed for each fiber bundle for tract spatial parameterization, i.e. defining the corresponding arclengths per fiber [139]. Diffusion properties along each tract were extracted for each individual participant employing the final transform computed during atlas building using DTIAtlasFiberAnalyzer6. Within DTIAtlasFiberAnalyzer, these diffusion parameters are then plotted as diffusion pro- files along the length of each tract, and FA profiles were visually quality controlled for participant profiles with poor correlation to the atlas FA profile. An example of FA profile plots for all participants is displayed in Figure 3.3, which depicts the left and right fornices. Diffusion parameter measurements were then statistically analyzed via Functional Analysis of Diffusion Tensor Tract Statistics (FADTTS)7 [126]. Local
5
http://www.nitrc.org/projects/fvlight
6http://www.nitrc.org/projects/dti_tract_stat/
−60 −40 −20 0 20 40 60 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 arclength FA fornixR FA NonSmokers Smokers T M T M M T A Non-Smokers Smokers 0.6 0.4 0.2 0 −60 −40 −20 0 20 40 60 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 arclength FA fornixL FA NonSmokers Smokers Non-Smokers Smokers M T 0 0.6 0.4 0.2 0.8 B
Figure 3.3: Participant fornix fractional anisotropy (FA) profiles. Measurements of FA along the tract for all participants in the left (A) and right (B) fornices. FA profiles are plotted in the middle panels, with nonsmokers indicated in blue, and smokers indicated in red. The color bar on the x axis of the FA profile plots corresponds to the location along the left or right fornix, as depicted in the outer panels. M−medial, T−temporal.
p−values and betas computed with FADTTS were plotted onto the fiber bundle and visualized in 3D Slicer or ShapePopulationViewer8.
Tractography ROIs
Region of interest label maps for use in the tractography of each tract were based on tract anatomy. Label maps for each tract were manually constructed to be inclusive, but specific for individual tracts. Left and right tract label maps were created at the same time in Slicer, using unique label identifiers. Simultaneous label creation aids in assuring the bilateral consistency of the label map extent for each tract.
Tractography algorithm
Tracts were seeded in 3D Slicer using label maps as described above. Tracks between 10mm and 800mm were computed with a seed spacing of 0.5mm, stopping FA value of 0.18, stopping curvature of 0.7mm, and step length of 0.5mm. Tracts were then cleaned utilizing negative and positive ROIs until only the fibers of the desired tract remained.
8