Post on 16-Jan-2016
Fuhai Li, Ph.D.BBP-TMHRI, Feb 7 2011
3D Tumor Stem Cell Niche Image Analysis
http://www.methodisthealth.com/icbp
NCI-ICBP CMCD U54 Progress Report
Tumor Stem Cell Niche
J Rosen, BCM
From 2D to 3D
From center grant proposal
DAPI GFP
Dextran Superimposed .44 μm X .44 μm X 1 μm
Specific Aim
Digitize 3D images of tumor stem cell niche & provide quantitative data for
• Tumor modeling• Blood vessel tracing• Nuclei segmentation
What information is expected?Anything we can see, image,
and imagine! E.g.,
• Spatial distributions (in normal and tumor, before and after treatment) of blood vessels, stem cells, stroma, etc.
• Cell morphology, proliferation and death
From John, Tegy, Mei Zhang, Michael T. Lewis
Blood Vessel and Nuclei Segmentation
Blood Vessel Segmentation
(Tracing)
Technical Challenges:
Noise Signal
Broken Points and Vessel Crossing
3D Blood Vessel Image
Maximum Intensity Projection (MIP)
Validation of Available Segmentation Software
http://research.mssm.edu/cnic/tools-ns.html
NeuronStudio
http://research.mssm.edu/cnic/tools-ns.html
V3D
http://penglab.janelia.org/proj/v3d/V3D/About_V3D.html
FARSIGHT
http://www.farsight-toolkit.org/wiki/Main_Page
Multi-scale Filtering and Optimal Linking Based Approach
Multiscale Vessel Enhancement Filtering
2
2
2
, ,
, ,
, ,
I I Ix y x zx
I I IH I y x y zy
I I Iz x z y z
Hessian Matrix
Multiscale means image filtering with Gaussian filters (with different sigmas)The eigenvalues of Hessian matrix tell us where are the vessels
A. Frangi, et. al, MICCAI, 1998
Binary Image of MIP (Original)
Multiscale Vessel Enhancement Filtering
Removal of Debris
Skeleton Extraction
T.C. Lee, et. al, GVGIP, 1994
Binary
3D View
Optimal Linking by Integer Programming (ongoing)
1
, ,i i
i i
p p ip p
v p q I p
argmax ,all possible broken pairs
v p q
Cost function:
,p q can only be linked to at most to once!subject to:
Fitness termsmoothness term
Solving:
Nuclei Segmentation
Technical Challenges:
Uneven Intensity and Cell Clustering
3D View
FARSIGHT
3D Analysis
Flowchart of Nuclei Segmentation
Z-Series 2D Images
Adaptive Thresholding
Distance Transform(Shape Info.)
Seed Image(Non-Maximum Suppression)
LoG Filtering(Intensity Info.)
Level Set Segmentation
2D Image Analysis
TT
Proposed
3D Analysis
Refine & Define Nuclei by Ellipsoid Fitting
Low Z resolution
digital 3D tumor stem cell nichenucleus = {o, a, b, c, α, β}
Ongoing Work
1. Blood vessel smoothing & broken points linking
2. Nuclei ellipsoid fitting, Image quantification & dividing cell identification
3. Process large volumes of image data to derive information for tumor modeling
4. Graphic user interface (GUI) tools for image analysis, visualization & tumor simulation
TT
Normal
Breast
Tissue
From Dr. Michael T. Lewis (John David Acquired)
Acknowledgements
BBP-TMHRI:
Stephen Wong
Xiaofeng Xia
Hong Zhao
Derek Cridebring
BCM:
Michael Lewis
Mei Zhang
Mary Dickinson
John Landua
Tegy Vadakkan
Wei Wei
NIH NCI-ICBP U54 Grant Support