Exploring Connectivity of the Brain’s White Matter with Dynamic Queries Presented by: Eugene (Austin) Stoudenmire 14 Feb 2007 Anthony Sherbondy, David Akers, Rachel Mackenzie, Robert Dougherty, and Brian Wandell IEEE Transactions on Visualization and Computer Graphics, V11, No 4, July/August 2005
Problem New technology emerged –Diffusion Tensor Imaging (DTI) –White matter connections, i.e. fiber tracts, can now be measured Need to take advantage of it Requires better visualization
We Care Better visualization would –Assist research –Interactive
Approach Combine types of data –Anatomical – White – DTI –Functional – Gray – fMRI Functional Magnetic Resonance Imaging Precompute Query Interface –Pictoral –Labeled –Ranges
DTI Diffusion Tensor Imaging New Technology Measures white matter pathways Estimates water molecule diffusion –Water diffuses lengthwise along axons –Diffusion direction nerve fiber orientation
One Method of DTI Visualization MR Tractography Traces principle direction of diffusion Connects points into fiber tracts Fiber tracts = pathways Anatomical connections between endpoints of the pathways are implied Therefore, implied white matter structure
These Pathways Not individual nerves Not Bundles But something Abstract, white matter route “possibilities”
fMRI Functional Magnetic Res Imaging Correlate activity Suggests gray matter connections
The Combination Take the MR Tractography data Precompute paths, statistical properties Interactive manipulation –Regions of interest – Box / Ellipsoid –Path properties – Length / Curvature Combine with fMRI –Search for anatomical paths that might connect functionally-defined regions Saves time over existing approaches
Query Interface
Query Interface – Partial Blowup
Acqusition DTI & fMRI
Subject Neurologically Normal Male Human 35
DTI Eight 3-minute whole brain scans –Averaged –38 axial slices –2 x 2 x 3 mm voxels 8-minute high res anat images –1 x 1 x 1 mm voxel Coregistered DTI resampled to 2 mm
fMRI obliquely oriented slices 2 x 2 x 3 mm voxel Registered with anatomy Mapped to cortical surface mesh
Precomputation
Fractional Anisotropy (FA) Diffusion orientation ratio 0 = spherical = gray matter 0.5 = linear or planar ellipsoid 1 = very linear Uses –Algorithm termination criteria –Queries –Navigational aid
Approaches Typical –Interactively trace pathways Authors’ –Precompute pathways –Over entire white matter –Then let software “prune”
Cortical Surface Classified white matter Semi-manually – neuroscientist Marching-Cubes -> t-mesh Smoothed Kept both 230,000 vertices
Precomputation Statistical properties Length Avg FA Avg Curvature Tractography Algorithm
Implementation
Path Rendering Lines vs streamtubes (for speed) Pathways – luminance offset Groups of pathways – hue –User defined hue –Virtual staining Queries modified – stains remain
Hardware/Software Visualization C++ ToolKit (VTK) RAPID –Fast VOI / Path Intersection Comp –80K-120K paths/sec (w/SGI RE) –Allowed MB for 26K 20KB/path 160MB for cortical meshes
Sequential Dynamic Queries
All 13,000 Pathways
Length > 4 cm
Through VOI 1
Through VOI 1 AND (2 or 3)
Volumes of Interest Surface-constrained
VOI on Cortical Surface
Same VOI, Smoothed Surface
Validation of Known Pathways
Occipital Lobe
Occipital to Right Frontal Lobe
Occipital to Left Frontal Lobe
Occipital to R & L, w/Context
Forming Hypotheses
Known and Unknown Paths
Algorithm Comparison STT – Streamlines Tracking Techniques Vs TEND – Tensor Deflection
STT (blue) vs TEND (yellow)
Exploration of Connections Between Functional Areas
fMRI Areas Colormapped
VOI Placement
Surface Removed Paths Visible
VOI Adjusted Different Paths
Evaluation Types of functions –Validation of known pathways –Hypothesis generation Time to explore – 10 minutes for significant exploration Speed – Interactive rates Interface – Interactive queries
Alternative Methods
Diffusion tensor visualization
White Matter Algorithms Streamlines Tracking Techniques Fiber Assg thru Cont Tracking Tensor-deflection
Filters Length Average linear anisotropy Regions of interest
Conclusion Multiple data types (DTI & fMRI) New visualization interface Interactive queries Hypothesis generation & testing
Next Steps Real work Multiple subjects Normal to abnormal Acquisition technology Path tracing algorithms
Question Is there any reason for tools such as this to be validated?
Question If validated this early on, wouldn’t every change pretty much negate the validation?
Question Should there be some kind of benchmark to use to measure these applications against?