UCSB Bio-imaging Infrastructure

1
UCSB Bio-imaging Infrastructure
Center for Bioimaging Informatics
www.bioimage.ucsb.edu Supported by NSF ITR
0331697

• December 2006

2
Projects

• Bisque/OME
• In use at UCSB
• Bisquik
• Next generation system (in construction)

3
Motivation

• Analysis creates knowledge
• Image analysis
• Querying
• Mining
• Available Storage systems are growing
• Available Processing increasing

4
Challenges

• Diverse users
• UCSB Neurosciences institute
• Retinal detachment Fischer et al
• Microtubule dynamics Feinstein and Wilson
• CMU Murphy Lab
• Flybase
• Computable Plant
• Dataset challenges
• Multiple types and collection techniques
• Complex images (5D) and metadata
• Acquisition and management of large sets
• Security of private data
• Sharing of experimental data

5
Challenges

• Metadata
• Collection
• Organization
• Sharing/interpretation
• Dataset management
• Personal collections
• Multiple organizations
• Privacy and security
• Analysis design
• Analysis integration

6
Block System diagram
Image Analysis
Browse Search Content metadata
Collection Analysis
Interactive Enhancement
Semantic Analysis
Knowledge Discovery
Image and metadata capture
DB Storage
Ground truth Collection
7
BISQUEBio-Image Semantic Query User Environment

• Current dataset collection
• BISQUE functionality
• Data management
• Ground truth acquisition
• Analysis
• Architecture

8
Current collections

• UCSB Retinal Objects (confocal, EM)
• UCSB Microtubule Objects(light, AFM)

Type Current Backlog Rate/y Expected 4Yrs 2D Images
Retinal EM 500 22000 500 23,000 23K
Retinal confocal P 3000 600 2400 10,000 10K
Retinal confocal Z 600 14000 12000 10,000 50K
Microtubule light 3000 2500 2500 13,000 400K
Microtubule AFM 0 500 1200 5000 30K
Flybase 0 125K 0 125K 125K

• Total estimated size in TBs and growing for 1 Lab
• Complexity and analysis are the main issues

9
Data management capabilities

• Digital notebook
• Direct import process
• Reconfigurable (confocal, 4 x microtubule, etc)
• Web
• Web access and browsing
• Organize images and metadata
• Data sharing environment
• Search by metadata or content
• Integrated analysis

10
Screenshots (Digital Notebook)
11
Data management capabilities

• Digital notebook
• Capture experimental/image parameters
• Direct Import process
• Reconfigurable (confocal, 4 x microtubule, etc)
• Web
• Web access and browsing
• Organize images and metadata
• Data sharing environment
• Search by metadata or content
• Integrated analysis

12
Screenshots (browsing)
Dataset Browsing
13
Screenshots (search)
14
Screenshots (search)
15
Screenshots (search similar)
16
Screenshots (search similar)
17
Screenshots (personal collection)
Data sharing
18
Screenshots (5D Viewer)
19
Screenshot (5D with tracks)
20
Image analysis

• Integrated image analysis
• Image enhancement
• Cell counter (ImageJ)
• Quantify microtubule dynamics
• Microtubule tracker (manual, automatic)
• Track identification
• Integration in progress
• Segmentation and classification
• Modeling microtubule dynamics
• Relevance feedback improving search

21
Screenshots (Image Analysis)
Image J Cell counter
22
System description overview

• Current dataset collection
• Current functionality
• Data management
• Ground truth acquisition
• Analysis
• Architecture

23
Hardware/software infrastructure

• Hardware
• 16 node Cluster
• dual Intel Xeon 3GHZ
• Gigabit network switch
• 2 TB Storage
• Software
• Bisque
• OME (Apache, Postgresql)
• Linux

24
Bisque ArchitectureBio-Image Semantic Query User
Environment
BISQUE
Research in biology
Distributed computing cluster
image
Digital Notebook
XML
metadata
features
Image/ metadata server
search
WEB
analysis
Cell counter lug-in for ImageJ
Cell counter lug-in for ImageJ
external
internal
Cell counter lug-in for ImageJ
XML
Cell counter plug-in for ImageJ
Research in image processing and indexing
25
OME Base

• Open Microscopy Environment
• OME is an open source software project to
  develop a database-driven system for the
  quantitative analysis of biological images. OME
  is a collaborative effort among academic labs and
  a number of commercial entities.
• Provides base for image/metadata storage and
  analysis integration
• Boston Sorger Lab
• Baltimore Goldberg Lab
• Dundee Swedlow Lab
• Madison LOCI

26
Bisque extensions

• Ongoing extensions with OME
• Content based search
• Analysis integration with OME
• Segmentation
• Cell Counting
• MT identification and Analysis
• Etc.
• Front end dataset and analysis support
• Schema additions for image types and analysis
• (Uncertainty modeling and queries)

27
Bisque

• Built 1st generation w/ gt5000 5-D images
• Integrated several useful analyses
• Being used internally
• External Interest
• Immediate future
• Continued development
• analysis integration
• Dataset integration
• Remote deployments
• Integration with other projects
• Flybase Integration of schema/datasets in
  progress
• Computableplant.org

28
Projects

• Bisque/OME
• In use at UCSB
• http//hammer.ece.ucsb.edu/bisque
• guest/bioimage
• Bisquik
• Next generation system (in construction)

29
Bisque/OME lessons learned

• Getting the correct metadata is hard
• Correct may need to change or be reinterpreted
• Sql schema difficult to change
• Getting the right name is difficult
• Different terms used in different labs make data
  integration painful
• Analysis integration needs to be easy
• Researcher balk at learning complex software
• Users expect a rich experience
• Google, flickr, etc have raised the bar

30
New project Bisquik

• DoughDB Flexible metadata
• Ontology support
• Rich user experience
• Web 2.0 Ajax, SVG
• Web based tools DN, Graphical Annotator
• Programming toolkit
• Smaller is better
• Distributed data store
• Support for large scale computation
• Expected early 2007

31
Block Components
Metadata Annotation
DoughDB
Metadata Renderer
Permission
Web UI
Ontology
Blob/Image Server
Remote Access
Analysis Engine
32
Tagging Screen Shot
33
Motivation

• Current metadata model is inflexible
• Adding new experimental images requires
• Changes to Digital notebook
• Changes to Bisque interface
• Changes to OME/postgres
• Shouldnt this be easier?
• Find images tagged with rod-opsin
• Create a region and specify the object
• Add experimental metadata to this dataset

34
Bisquik requirements

• Add new tag/value pairs to any db object
• (Foo,2)
• (visible-cell, rod)
• Allow multiple tags with same value
• (visible-cell, rod)
• (visible-cell, muller)
• Support fine-grained tag permission/visibility
• Tags have creators and access control
• Support update semantics preserve history
• Timestamp tags
• No deletes (except under certain conditions)

35
Bisquik DoughDB
OID4
OID1
image OID1
Feature f1



Name GH1020
Foo 2
pixels OID2 OID3





OID2
image OID1
data Server//
Pixel-type raw


36
Bisquik queries

• Find objects (images) with tag foo
• Select a.foo
• Find images with name GV100
• Select a.name gv100
• Find images with cellcount lt 100
• Select a.cellcount lt 100
• Find images with a region similar to r1 based on
  feature f1
• Select r.image i and l2(r.f1, r1 ) lt 10

37
DoughDB key features

• No classes or types only tag/value pairs
• Open ended data model
• Pair values have owner and acl
• Preserves history of annotations
• SQL like query language
• Simple keyword queries

38
Bisquik ontology support

• ontology is a data model that represents a
  domain and is used to reason about the objects in
  that domain and the relations between them
• Unstructured tag/value
• Great for taggers
• Unhappy searchers
• Different labs use different terms for the same
  object.

39
Bisquik ontology support

• Dictionary of terms and relations
• Require (or strongly suggest) that tags and value
  are defined before use
• Drop into ontology editor when new values and
  tags are encountered.
• Integrated into search system
• Permit (or offer) alias part-of related-to
  searches

40
Analysis Engine

• Analysis Components
• Analysis Applications
• Glue language python
• Component connection library
• Memory for single node
• MPI for cluster based execution
• Execution engine
• Automatic component placement
• Resources and efficient communication

41
Analysis Engine

• Interfaced with DoughDB
• All input/output are Pairs or objects
• Application executions recorded for data
  provenance

42
Bisquik interface

• http//oib.ece.ucsb.edu/bisquik
• Some bisque functionality
• Flickr-like interface for region tagging
• Pair tagging and keyword tagging
• Metadata renderers
• Textual (tag)
• Graphical (regions, geometry, graph data)
• Analysis oriented (histogram)

43
Region Tagging screenshot
44
Bisquik Metadata Annotation

• Unified offline (Digital Notebook) and online
  manipulation.
• Easy to build annotation forms
• Allow schema modification in field
• Permit annotation templates to be shared between
  DN and Bisquik
• Graphical geometry annotator

45
Blob Image server

• Extensible server for write-once objects
• Pixels, Features
• Pluggable transforms/operations
• Thumbnails, slices
• pixel transforms (watermarks)
• Graphical metadata renderers

46
Remote Access

• All basic services are web accessible
• Soap and WSDL
• DoughDB, Image server,
• DoughDB pairs have unique 64 bit IDs
• Split between machine ID/Pair ID
• All pairs are addressable
• Query engine processes foreign pairs

47
Bisquik new hardware

• 10 TB disk mirrored array (20TB)
• Image Server
• Database (backup) server
• 8-16 Query/Analysis nodes

Image Server
DB/Backup Server
10TB
10TB
48
Status

• Web UI
• Uploading, Tagging, simple searches
• DoughDB
• Single node storage and queries
• Analysis Engine
• In design

49
Conclusion

• Bisque Team August, Jiejun, Melissa,Kris
• Bisquik Team
• Interface August Jiejun Jaechok
• Analysis Engine Kris, Mellisa, Dmitry
• Ontology David(summer), Verleen(summer), Kris
• DoughDB Kris,Vebjorn

50
Biowall
51
Biowall

• 5x4 1600x1200 displays
• (8000x4800 pixels)
• 38 MegaPixels
• 1 Head node 5 slaves
• Distributed multihead X (X proxy)
• New simplified viewer
• OpenEV image server
• Local client
• Web/Database client
• 3D visualizations
• Working with very large screen areas