1. Bioimage database architecture and infrastructure 2005, Bio-ITR, UCSB

2. Overview Current system (UCSB) –Status of collection –Capabilities –Architecture Current system (CMU) –Status of collection –Capabilities –Architecture Joint system under development –Capabilities –Architecture Future –Layered databases –Distributed databases

3. Current collection Retinal –Confocal microscope –EM (Electron micrograph) TypeCurrentBacklogRate/yExpected 4YrsTotal size Retinal EM6001900050020,00020GB Retinal confocal P3000500240010,00010GB Retinal confocal Z0140001200010,00065GB Microtubule light30002500 13,00012GB Microtubule AFM20001200500015GB Microtubule DIC002.7M10M10TB Microtubule –Light –Atomic Force Microscopy –DIC/Nomarski

4. Current capabilities Import process Image and meta storage Web access and browsing Limited access by content

5. Screenshots (browsing)

6. Screenshots (search)

7. Screenshot (metadata edit)

8. Screenshot (retina meta)

9. Current architecture Metadata Database implementation Front end implementation Image import API Software and hardware infrastructure

10. Metadata Standard (image types, parameters) –File, size, type, tiff data, channel info, etc. Retinal –Visible cells –Antibody labeling –Experimental conditions –Researcher Microtubule –Track (hand captured) AFM –Machine parameters Metadata sources –Researcher –Annotated excel files –Proprietary image formats

11. Database implementation MySql First generation schema –image parameters File, size, type, tiff data, etc. –Metadata Experimenter, condition, antibodies, tissue, notes, etc.

12. Front end Apache, Php, Javascript Import proprietary image types Browse images Search by metadata Search by similarity Multi user and release protection

13. Image and metadata import Excel parser for metadata Image import library –Image Format API and C/C++ library for database and client applications were developed. –Currently supported proprietary image formats: Metamorph Stack, Fluoview TIFF, BioRad PIC, PSIA TIFF, Nanoscope, + common: JPEG, TIFF, BMP, PNG…

14. Hardware and software infrastructure Hardware –Dell Server with dual Intel Xeon cpu at 2.4Ghz –140GB scsi hard drive set up as RAID 1 –Gigabit network switch Software –Linux, version Fedora 2 –Apache Web server with PHP, PERL and graphical modules –MySQL Database server

15. Overview Current system (UCSB) –Status of collection –Capabilities –Architecture Current system (CMU) –Status of collection –Capabilities –Architecture Joint system under development –Capabilities –Architecture Future –Layered databases –Distributed databases

16. Overview Current system (UCSB) –Status of collection –Capabilities –Architecture Current system (CMU) –Status of collection –Capabilities –Architecture Joint system under development –Capabilities –Architecture Future –Layered databases –Distributed databases

17. Motivation Common schema between UCSB and CMU Support greater functionality –Analysis and interpretation tools –Ground truth –Semantics –Uncertainty –Complex features and distance metrics MPEG-7 features Other features –Querying and relevance feedback

18. Capabilities Image and metadata storage Web access and browsing Access and search by content Import/Export –Streamlined XML import/export for external tools Schema extensions –Image5d, semantic, uncertainty, analysis Image processing modules and tools

19. Infrastructure – Interchange XML Unified interchange XML format is being developed for database feeding and extraction procedures, external client application interaction and database intercommunication. DB XML External clients Image library External DB interchange Import/export remote access Ground truth tools Image processing tools

20. Ground truth acquisition tools Image processing and infrastructure teams are developing universal “ground truth” collection tools able to retrieve data from data-base and feed user defined information back to the database. The main communication vehicle is XML interchange format. At the current stage stand alone tools are being developed and tested that later on will be grouped in the universal application able to communicate directly to the data-base. +

21. Image processing API Fast development of image processing tools concentrated on problem solving. API provides simple access to multi-channel image and mask information. Allows progress output, acquisition of user defined parameters and automatically created filter preview. Example of API usage: Noise removal for Fluoview images result noise input

22. Semantic data modules Integration of current research in automatic image analysis: –Cell identification –Layer detection –Cell counting –Microtubule detection and tracking –Microtubule dynamicity and global characterization

23. Modeling uncertainty Uncertain identification/analysis –Simple probability (e.g., 0.8) –“Is this a rod bipolar cell?” Imprecise location/extent/count –90% accuracy in cell count –Line segment (single or sequence), polygon Identified by a sequence of points Each point Gaussian Store mean x, mean y, and standard deviation –Circle Center  Gaussian point, as above Radius  mean r and standard deviation

24. Schema Image5d Analysis and interpretation tools –Quantitative data generation –Semantic Labeling Experimental description Shape and geometry Domain knowledge –Ground truth –Semantic objects Uncertainty Features and distance metrics MPEG-7 features Other features Querying and relevance feedback

25. Schema (image5d) 5d images Image is a set of bit- planes Group planes by which dimensions vary Permits –Multiple formats –Caching

26. Schema (semantic objects) Capture semantics Capture uncertainty Type of object : confidence Position of object: Gaussian domain

27. Schema (analysis and features) Capture provenance Support type checking Support feature substitution

28. Hardware and software components Hardware requirements –Same as original system Software –Postgresql backend –JSP / JSF front end Migrate php/javascript current code into components

29. Architecture Web Page UI Generation View MenuTable Semantic Interface DB Storage Image Cell Dynamic JSF Components Programmable Image API Model API Object Relational (Postgresql) HTML XML

30. Overview Current system (UCSB) –Status of collection –Capabilities –Architecture Current system (CMU) –Status of collection –Capabilities –Architecture Joint system under development –Capabilities –Architecture Future –Layered databases –Integration with other databases BIRN OME metadata and schema exchange

31. Layered database Overlay model (interpretation) on image (raw) data Multiple interpretations of data URI references between databases Pro: Logical distinction, multiple interpretations, flexible implementation

32. BIRN (Biomedical Informatics Research Network) Goals: –Link multiple databases with different schemas, maintained at different research institutions 19 universities, 26 research groups Current collection –Three test beds centered around brain imaging of human neurological disorders and associated animal models: Functional BIRN Morphometry BIRN Mouse BIRN

33. Integration with BIRN Databases at UCSB/CMU Centers can be integrated into the BIRN federation UCSB/CMU infrastructure supports –Extensive metadata for images –Standard XML interchange format for 5d images –Computational tools to refine data Web based visualization and analysis tools We need to: –Translate UCSB/CMU Schema to F-logic (Knowledge-based mediation) –Link UCSB/CMU dataset to UMLS (Unified Medical Language System) ontology –Reference a common spatial framework Standard atlas coordinate system, e.g., SMART Atlas

34. OME Open Microscopy Environment –a set of software that interacts with a database to manage images, image meta data, image analysis and analysis results Designed to perform as a local system Integration with OME –Adapt OME XML image interchange mechanism –Adapt the database oriented modular analysis approach of OME

35. Conclusion Built prototype and collected ~4000 images –Being used internally Concurrent work on 2 nd generation system –Image loading –Integration of tools –New front end