Ellisman Laboratory

Ellisman Laboratory
3D Microscopy Data
IGP
• Electron tomography
• Large data sets
Heterogeneous collection of
data/scientific questions
Goal: segmentation &
visualization of cells and
subcellular structures
NCRR
EMT Data
IGP
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Projection
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Manual Segmentation
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Reconstruction (FBP)
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Direct Volume Rendering
NCRR
EMT Data
IGP
Gap Junctions
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Actin Fibers
NCRR
Ellisman Collaboration Activities
Meetings at NCMIR
IGP
• February 2006: Discuss segmentation app. ideas
• July 2006: Preview segmentation app. prototype,
interview users, discuss research ideas
Software development
• Segmentation / visualization application
Collaborative activities
• Visualize gold-labeled data
• Segmentation of spiny dendrite data (ongoing)
• Segmentation of mitochondria data (ongoing)
NCRR
Visualization
IGP
Visualizing Gold Particles in EMT Data
NCRR
Segmentation Application
Motivation
IGP
• Concrete progress on Ellisman
collaboration
• Designed in consultation with NCMIR staff
• Generalize functionality
– Wide range of applications within the CIBC and
elsewhere
Design strategy
• Comprehensive 3D interface
– Photoshop look and feel, layers
• Tools for manual interaction
• Hooks to wide variety of ITK algorithms
NCRR
Segmentation Application
IGP
Image
Data
Stack of Image
Volumes
ITK File I/O
Manual
Input
Contouring and
Painting
Layered
Slice
Visualization
Select Layers
Set Parameters
ITK Filters
Teem Filters
Custom Filters
3D Visualization
Resampling
And Streaming
Processed layer
returned to stack
as new layer
NCRR
Segmentation Application
Demo – 5 mins
IGP
NCRR
Capecchi Laboratory
Impact
• Genetics, development, cancer
• Expand the scope of questions
IGP
Eviscerated Mouse Paw
Challenges
• High-throughput
• User-assisted/automated tools
• Quantitative microCT
Approach
• Processing pipeline
• Progression of tools
NCRR
Capecchi Collaboration Activites
IGP
Regular meetings with Capecchi lab
staff and students
Segmentations of wild-type and
homozygous (Hox d11) mutant
forelimbs
Paper in progress: length
comparison (digital) of mutant and
wild-type forelimb bones
NCRR
Capecchi Collaboration Activities
Shape analysis algorithms
IGP
• Paper accepted: Workshop on Mathematical
Foundations of Computational Anatomy 2006
Shape analysis library
• Particle system and shape analysis code for ITK
(supported locally by CIBC and NAMIC)
• Integrate into CIBC Segmentation application
Shape analysis of hoxd11 vs. wild-type mouse
forelimb bones
NCRR
Capecchi Collaboration Activities
IGP
Image-based
Meta
0.69
P1
0.78
P2
0.79
P3
1.08
Original
0.63
0.79
0.63
0.99
Preliminary results for digit 2 from the image-based bone-length
comparison study.
Expressed as the ratio of the average length of the mutant bone to the
average length of the wild type bone.
Normalized with respect to the length of an individual specimen's
humerus.
NCRR
Capecchi Collaboration Activities
IGP
Wild Type
Homozygous Mutant
Segmentations of wild type and homozygous mutant
bones of the second digit of the left mouse forelimb.
NCRR
Capecchi Collaboration Activities
IGP
Length measurement of the mouse humerus using SCIRun.
NCRR
Capecchi Collaboration Plans
IGP
A shape analysis pipeline for mouse
phenotyping
• Open-source particle system library (Winter 2006)
• Further shape analysis research (ongoing)
Specific scientific results for skeletal
phenotyping with the Capecchi Laboratory
• Finish forelimb segmentations (Fall 2006)
• Bone length comparison paper (Fall 2006)
• Shape study of mouse bones (ongoing)
NCRR
Related Project: Bridge
Deconvolved
Original
Reconstruction and visualization of microtubules
from FLM
IGP
Visualization of Block Traversal of Tubuli
NCRR
Related Projects: Chien,Marc
Tracking of axons in SBFEM
IGP
Deformable tiling and registration
of serial section TEM
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
QuickTime™ and a
TIFF (Uncompressed) decompressor
are needed to see this picture.
QuickTime™ and a
TIFF (Uncompressed) decompressor
are needed to see this picture.
Papers to appear in MICCAI 2006 Workshop on Microscopic Image Analysis
NCRR
IGP
NCRR
Image and Geometry Processing
Highlights of ongoing work
IGP
• Geometry processing
• Shape analysis
• Visualization/segmentation
NCRR
Geometry Processing/Modeling
IGP
NCRR
MRI Head Segmentation
IGP
Applications
• Bioelectrical fields
• Neuroscience, neurology, psychiatry
State of the art
•
•
•
•
Parametric statistics (EM)
Inhomogeneity correction
MRF (regularization)
Atlas-based prior (registration)
Our goals
• Improve robustness/accuracy -> usability
• Robust, ubiquitous implementations
• Integrate into processing pipeline
NCRR
Segmentation Research
Neighborhood statistics
IGP
• Manifolds in high-dimensional
spaces
• Filtering, compression,
segmentation, visualization
Strategy
• Nonparametric density
estimation
• Engineering issues
NCRR
Image Filtering
IGP
Reduce entropy – PDF of image neighborhoods
Process that learns underlying image structure (UINTA)
• Awate, Whitaker, CVPR, 2005
• Awate, Whitaker, PAMI 2005
Include MRI noise model for a posteriori estimation
• Awate, Whitaker, IPMI 2005
NCRR
MRI Head Segmentation
IGP
Tasdizen, Awate, Foster, Whitaker, MICCAI 2005
(under review)
GM Classification Performance vs Noise Level
97
95
93
91
89
87
Proposed
Leemput
85
0%
MRI Input
GM, WM, CSF Seg.
1%
3%
5%
7%
9%
Comparison: SOTA–EM w/MRFs & Atlas
(Leemput et al.)
Collab: Makeig-Worrell, Wolters, Warfield, McIntyre
NCRR
Cell Segmentation
Nonparametric density estimation – image neighorhoods
IGP
• Texture
Partition image to reduce in-class entropy
Random initialization
Fast (nonlocal) level-set method
Collab: NCMIR, Marc
NCRR
Modeling DW-MRI
IGP
David Tuch, MGH
• Q-Ball Imaging
Computational microanatomy
• Reproduce DW-MRI measurements
through simulation
Impact
• DW MRI for the study of neurological
disorders (e.g. dementia)
• Scientific/clinical inferences
• Image acquisition
Challenges
• Generating realistic anatomies
• Controling relevant parameters
• Generating sufficient statistics
Peters et al., 2002
– Scale
NCRR
Preliminary Results
IGP
Strategy
• Generate statistically accurate
geometry
• Continuum model
• Parameters from literature (radii, gratios, etc.)
FA = 0.44
FA = 0.55
FA = 0.73
NCRR
Summary
Motivation and Philosophy
Research and Development
Integration and Applications
Progress
IGP
NCRR
IGP
NCRR