Transcript Modeling A Deformable Mouse Brain Atlas Using Subdivision

A Geometric Database of Gene
Expression Data for the Mouse Brain
Tao Ju, Joe Warren
Rice University
Overview
• Genes are blueprints for creating proteins
• For given tissue, only a subset of genes are
generating proteins (expressed)
• New laboratory method for determining
which genes are being expressed (Eichele)
• Collect expression data over mouse brain
for all 20K genes in mouse genome
• Build database of gene expression data
Gene Expression Database
• Collect gene expression data for small number
of cross-sections
• Bring 2D cross-sections into 3D alignment
using principal component analysis
• Deform 3D brain atlas onto aligned crosssections to account for anatomical deviations
• Analyze and compare gene expressions via
mapping to standard brain atlas
The Standard Mouse Brain
• 15 anatomical regions spread over 11 saggital crosssections (from lateral to medial)
Deformable Modeling
• Anatomical deviation between mouse brains
– Need to deform standard atlas onto each brain
• Most deformable models are based on a uniform
grid
– “Brain Warping”, by Arthur W. Toga
• Our contribution: subdivision meshes
Subdivision Mesh as Brain Atlas
• Subdivision through splitting and averaging
• Boundaries of anatomical regions modeled by
crease curves
• Intersection of three or more regions modeled
by crease vertices
Demo : Fitting a Mesh
Advantages of Subdivision Meshes
• Subdivision meshes are easy to fit to image
– Simple manual drag-and-drop of control net
– Fast automatic fitting methods
• Anatomical regions isolated as sub-meshes
• Expression data stored as extra coordinate
on refined meshes
– Allows fast, accurate comparison of data
– Multi-resolution structure improves efficiency
Automatic Textual Annotation
• Previously, biologist examined and
manually tagged each anatomical region
with pattern and strength of gene expression
– Pattern: scattered, regional, ubiquitous
– Strength: -, +, ++, +++
• Now, apply filter to determine pattern and
strength of expression over sub-mesh
corresponding to anatomical region
Comparison of Expression Data
•
Search for an image with the most similar
expression pattern to a given target :
1. Build summaries in each quad at each subdivision
level using Haar wavelet
2. Sort all images by comparing at the coarsest
subdivision level into a priority queue
3. Compare the first image with the target at a finer
subdivision level and update the queue, until it is
already at the finest level (i.e., a match is found)
•
Requires monotonic (convex) norm
–
L1, Chi-square, etc.
Geometric Searches
• Let the user define
a target expression
pattern from:
– Preset values,
– Existing genes.
• Selectable distance
norm and number
of matches.
Demo: Searching the Database
Accessing Database via the Web
• Database of gene expression data and
deformed atlases
– currently 1207 images from 110 genes.
• Web server: www.geneatlas.org
–
–
–
–
Uploading and viewing gene images.
Fitting standard atlases (using Java Applet).
Graphical interface for searching gene images.
Automatic annotation.
• It’s all online!
Conclusion
• Subdivision meshes for anatomic modeling:
– Flexible control allows easy deformation.
– Crease points (curves) allows accurate
modeling of region boundaries.
– Enables fast and accurate comparison between
images on the multi-level grid structure.
Future Work
• Construction of a full 3D deformable atlas
of the mouse brain based on hexahedral
subdivision meshes.
• Algorithms for efficient and accurate fitting
of the 3D atlas onto cross-section images.
• Enhancement of the searching engine to
accept more complicated queries.
Collaborators
• Baylor College of Medicine
– Gregor Eichele, Christina Thaller, Wah Chiu,
James Carson
• Rice University
– David Scott
• University of Houston
– Ioannis Kakadiaris