Cell Segmentation Using Coupled Level Sets and ... - Semantic Scholar
Cell Segmentation Using Coupled Level Sets and Graph-Vertex Coloring
Sumit K. Nath, Kannappan Palaniappan, Filiz Bunyak Dept. of Computer Science, Univ. of Missouri, Columbia, MO
9th MICCAI Copenhagen, Denmark Oct 1-6, 2006
Cell Segmentation Using Coupled Level Sets and Graph-Vertex Coloring • • • • • • •
Motivation - cell tracking in high throughput imaging Source of the segmentation problem Spatial coupling using level sets Graph-vertex coloring Topological control Fragmentation vs absorption Experimental results for epithelial cell tracking
Tracking of Cancer Cells, Stem Cells, Leukocytes, Bioprinting 40,000 frames, 1280x1024, 12-bit 1000’s of cells per frame 100+ GB per experiment About 20 sec per frame processing
Li et al & Kanade, IEEE MMBIA, 2006
Althoff & Gustavsson, Sweden Mukherjee, Ray, Acton, IEEE TIP, 2004
Cell Cycle Analysis Using Fluorescence Microscopy HTS Genome wide RNAi screening 21,000 Drosophila genes 384 plate wells 400,000 to 1,000,000 images per screen Interphase
Prophase
Metaphase
Anaphase
Zhou & Wong, IEEE Signal Proc, 2006
High-throughput Cell Imaging Wound Closure Assays
Lauffenburger, Biophysical Letters, 2006
MitoCheck: High-throughput RNAi-based Phenotype Screening 384 plate wells, time-lapse GFP Images 1300 x 1024 1356 microarray spots per 30min 100’s GB to terabytes of imagery
Neumann, et al & Ellenberg, Nature Methods, 2006
Spatiotemporal Segmentation Challenge - Touching & Overlapping Cells • • • •
Cells are distinct objects prevent cell merging Inaccurate counting of cells Inaccurate assessment cell-to-cell communication Close by moving cells can touch and separate numerous times many false cell split (ie mitotic) events during tracking
•
Other potential approaches usually lead to over segmentation or fragmentation and need supervision for reliable performance – – – –
•
Thresholding Correlation matching Shape analysis Morphological processing
Explicit and implicit active contours powerful framework for cell segmentation
Active Contours without Edges (Multiphase) Level Set Method for Cell Tracking
Bunyak et al, IEEE ISBI, 2006
Which Type of Active Contour? • •
Explicit active contours (ie parametric snakes) using Langrangian formulation can preserve a known initial topology Cannot adapt to complex shape changes ie cell splits, apoptosis Coupled (region-based) parametric active contours, Zimmer and Olivo-Marin, 2005
• • • •
Implicit active contours solved using Eulerian formulation can adapt to complex topological changes and detecting new objects or disappearances during tracking Multiple level sets allow for multiple cell types but zero-sets can intersect and overlap or leave vacuums in labeling Multiphase level sets use logN level sets Computationally prohibitive Coupled geometric active contours, Zhang, Zimmer, Olivo-Marin, IEEE ISBI 2004
Spatial Coupling Using N-Level Sets
Spatial Coupling Using Level Sets •
Require flexibility of implicit level sets methods for cell cycle analysis
•
Extension of single level set framework – N-level sets, one per cell (ie connected component) – Spatial coupling between level sets to prevent cell merges
•
Need efficient computational implementation – N-level sets would need PDE solver for hundreds to thousands of evolving curves each potentially taking thousands of iterations – Spatial coupling terms are O(N2) so a thousand cells would need a million interaction terms
Graph Vertex Coloring •
Why? Spatial coupling needs to be local interactions only Reduce the number of level sets from N to k
Sumit K. Nath, Kannappan Palaniappan, Filiz Bunyak Dept. of Computer Science, Univ. of Missouri, Columbia, MO
9th MICCAI Copenhagen, Denmark Oct 1-6, 2006
Cell Segmentation Using Coupled Level Sets and Graph-Vertex Coloring • • • • • • •
Motivation - cell tracking in high throughput imaging Source of the segmentation problem Spatial coupling using level sets Graph-vertex coloring Topological control Fragmentation vs absorption Experimental results for epithelial cell tracking
Tracking of Cancer Cells, Stem Cells, Leukocytes, Bioprinting 40,000 frames, 1280x1024, 12-bit 1000’s of cells per frame 100+ GB per experiment About 20 sec per frame processing
Li et al & Kanade, IEEE MMBIA, 2006
Althoff & Gustavsson, Sweden Mukherjee, Ray, Acton, IEEE TIP, 2004
Cell Cycle Analysis Using Fluorescence Microscopy HTS Genome wide RNAi screening 21,000 Drosophila genes 384 plate wells 400,000 to 1,000,000 images per screen Interphase
Prophase
Metaphase
Anaphase
Zhou & Wong, IEEE Signal Proc, 2006
High-throughput Cell Imaging Wound Closure Assays
Lauffenburger, Biophysical Letters, 2006
MitoCheck: High-throughput RNAi-based Phenotype Screening 384 plate wells, time-lapse GFP Images 1300 x 1024 1356 microarray spots per 30min 100’s GB to terabytes of imagery
Neumann, et al & Ellenberg, Nature Methods, 2006
Spatiotemporal Segmentation Challenge - Touching & Overlapping Cells • • • •
Cells are distinct objects prevent cell merging Inaccurate counting of cells Inaccurate assessment cell-to-cell communication Close by moving cells can touch and separate numerous times many false cell split (ie mitotic) events during tracking
•
Other potential approaches usually lead to over segmentation or fragmentation and need supervision for reliable performance – – – –
•
Thresholding Correlation matching Shape analysis Morphological processing
Explicit and implicit active contours powerful framework for cell segmentation
Active Contours without Edges (Multiphase) Level Set Method for Cell Tracking
Bunyak et al, IEEE ISBI, 2006
Which Type of Active Contour? • •
Explicit active contours (ie parametric snakes) using Langrangian formulation can preserve a known initial topology Cannot adapt to complex shape changes ie cell splits, apoptosis Coupled (region-based) parametric active contours, Zimmer and Olivo-Marin, 2005
• • • •
Implicit active contours solved using Eulerian formulation can adapt to complex topological changes and detecting new objects or disappearances during tracking Multiple level sets allow for multiple cell types but zero-sets can intersect and overlap or leave vacuums in labeling Multiphase level sets use logN level sets Computationally prohibitive Coupled geometric active contours, Zhang, Zimmer, Olivo-Marin, IEEE ISBI 2004
Spatial Coupling Using N-Level Sets
Spatial Coupling Using Level Sets •
Require flexibility of implicit level sets methods for cell cycle analysis
•
Extension of single level set framework – N-level sets, one per cell (ie connected component) – Spatial coupling between level sets to prevent cell merges
•
Need efficient computational implementation – N-level sets would need PDE solver for hundreds to thousands of evolving curves each potentially taking thousands of iterations – Spatial coupling terms are O(N2) so a thousand cells would need a million interaction terms
Graph Vertex Coloring •
Why? Spatial coupling needs to be local interactions only Reduce the number of level sets from N to k
