Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Johan MontagnatI3S, CNRS
http://www.i3s.unice.fr/~johan/
Introduction
MICCAI tutorial on
grid services for medical image processing and
registration
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 2
Agenda
Grids, a tool for compute and data-intensive medical imaging applicationsJohan MontagnatMammography analysis on gridsMike BradyBiomedical Informatics Research NetworkRon KikinisIXI e-Science projectDerek HillCoffee breakThe case of medical image registrationDaniel RueckertRound table: our vision of grid technology for medical image registration, trends and problems, opened discussion, call for participationX. Pennec, D. Hill, J. MontagnatLunch
9h
9h30
9h50
10h10
10h3010h50
11h20
12h20
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 3
What are grids?
The myth: providing unlimited computing power by letting the user transparently access to infinite Internet resources.
Early grid adoptersData storage: napsterComputation: SETI@homeInformation: web
The infamouselectrical network analogy
Client-server technology}
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 4
What are grids?
Assembling resourcesStorageComputingNetwork
Federating usersLarge scale user communities (Virtual Organizations)Ease exchanges
Pushing standardsCommunication protocolsData representation and formatsComputation control languages
The real grid potential is in sharing resources, data, and knowledge.
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 5
New bottles around old wine?
Scalability/Extensibility problemsLoad balancingParallelizationDecentralizationFault toleranceSecurity...
Distributed computing has addressed these kind of problems for decades!
New wineInteroperability high level interfaces, standardsAuthentication and autorization certificates and policies
Peer-to-Peer (P2P) or agent technologies matureAlternative to the client-server approach
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 6
The new factors empowering grids
Faster networksGrowth of network bandwidth vs growth of computing/storage: cheapest data exchanges
Cheap PCsSuccessful development of cluster computingSome degree of standardization of hardware and software
Standardization bodiesW3C http://www.w3c.org/ (HTTP(S), *ML, SOAP, WSDL...)OASIS http://www.oasis-open.org/ (WS, security...)Global Grid Forum http://www.ggf.org/ (OGSA, WS-RF...)...
Grid marketingIndustry adoption
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 7
All is becoming gridified
Overuse of “grids”grid = parallelismgrid = cluster computinggrid = just name itgrid or gridS?
Over-expectationsin grids deployment schedulein grids capabilitiesin transparency of grids from the user point of view
PCs clusters are:cheap... but administrators are notable to deal with embarrassingly parallel applications
Supercomputers and dedicated systems have their own playground
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 8
From PCs to supercomputers
Several kind of grids:network of computersnetwork of clustersnetwork of supercomputers
Single PC capabilitiespowerful, yet limited capabilities of a each hostgeneral purpose network connection
Parallelismtrade-off between splitting and network overhead
Supercomputerslowest communications overhead, large splitting
Global GridNetwork of clusters and supercomputersResource brokering among available resources
Medical imagingcommunity
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 9
Grids are emerging technologiesSome components available
User authenticationHigh performance data transferData managers with replication and metadataResource brokering...
Still a lot to be donePerformance issues in scalabilityHigh level representation of dataTransparency of the underlying infrastructure...
Medical data processing is very complexThis community has one of the most challenging requirements listConfidentiality of data and security requirementsParallel processing, interactive jobs, emergency situations...
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 10
SemanticsToo much information kills information
To find some information on the web: googleTry “Mike Brady” keyword...
Semantic webCommon framework that allows data to be shared and reused across applicationsGive explicit meaning to informationResource Description Framework
Semantic gridsPropagate to grids what may become the web
Ontologies, OWLFormally describe the meaning of terminology used in a domainEnable processing of information rather than just delivery to humans
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 11
Real grid enabled applications
Very few decentralized, scalable algorithmssuch as Domain Name Service
How does Google backup the web?6 centers20000 PCsLoad distribution
Embarrassingly parallel applicationsEasily distributed computationsJust bring in more CPU and bandwidth
More to learn from P2P and agent technologies
Lot done in the field of data managementWhat about computations, knowledge extraction?
Client/Server
Client/Server
Client/Server
Client/Server
Client/Server
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 12
Application areas
High Energy PhysicsPb of data generated per sec as the output of the Large Hadron Collider
AstronomyAstronomy images
Earth ObservationSatellite images
GeophysicsSismic data revealing underground structure
Aeronautics, industryFinite Element Modeling, etc.
...
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 13
Sharing to get stronger
Unlike electrical power, adding computers is not equivalent to add volts
Distributed computing conjecture : a parallel machine performance is lower than the sum of individual processors performance
Humans are selfishEvery user want to use other's resources without seeing others using its resources
The real grid potential is in sharingdatasets: to create virtual databases, esp. for rare dataalgorithms: to foster reusability and comparisonsresources: to face computation picksenforce standards
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 14
Medical data on grids
Data storage and archival10 Tb of medical images/hospital/yearNeed for long term archival (20 to 70 years)
DatasetsLarge scale data setsStatistics, epidemiologyRare diseasesPersonalized atlases construction
Data representationMedical imagesMetadataOntologies
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 15
Security of medical applications
Authentication and AuthorizationCertificate authorities issuing certificate pairsAsymmetric certificate based authenticationAuthorization policies enforcement
Data access control at individual levelPhysiciansPatientsResearchers
DelegationGranting access rights
Encryption for data storage and transferBest effort privacy protectionKey-based data encryption
owner
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 16
Biomedical computing
Embarrassingly parallel applications
databases processingbioinformatics
Parallel computationscostly processings
Interactive computationsresources reservationuser supervision and validation
Emergency situationsresources preemption
Algorithms warehouse
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 17
Standardization
Medical data representationFile level: DICOM, Analyse...Database level: Data storage and mediatorsInter-sites: ontologies
Algorithms reusabilityImage formatsInput/outputParameters
Unique opportunity to enforce standards
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 18
Medical algorithms assessment
Sharing dataCommon datasets
Sharing algorithmsTesting others algorithms
Sharing proceduresCommon test suites
Sharing computing resourcesLarger assessment studies
Empowering testing and comparisons
Seq1 > dcscdssdcsdcdsc bscdsbcbjbfvbfvbvfbvbvbhvbhsvbhdvbhfdbvfd
Seq2 > bvdfvfdvhbdfvb bhvdsvbhvbhdvrefghefgdscgdfgcsdycgdkcsqkc
…
Seqn > bvdfvfdvhbdfvb bhvdsvbhvbhdvrefghefgdscgdfgcsdycgdkcsqkchdsqhfduhdhdhqedezhhezldhezhfehflezfzejfv
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 19
Grid infrastructures in Europe
EGEEhttp://www.eu-egee.org/27 countries, 70 partnersProduction platform8000 PCs to be deployed in 2 yearsLCG2 middleware (Globus+DataGrid based)
DEISAhttp://www.deisa.org/consortium of supercomputing centersBuild a distributed terascale computing facilityTight coupling of supercomputing centers by high bandwidth networks
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 20
References and related events
HealthGrid associationhttp://www.healthgrid.org/HealthGrid conference (next event in Oxford, 2005)
BIRN: Biomedical Informatics Research Network http://www.nbirn.net/
TeraGrid: http://www.teragrid.org/
UK e-Science: http://www.rcuk.ac.uk/escience/ myGrid: http://www.mygrid.org.uk/, In silico biologyIXI: http://www.ixi.org.uk/, medical imaging
French ACI-GRID, http://www-sop.inria.fr/aci/grid/public/
Mammogrid, http://mammogrid.vitamib.com/
GEMSS, http://www.ccrl-nece.de/gemss/
Crossgrid, http://www.eu-crossgrid.org/
...
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 21
Conclusions
Grids are emerging technologiesEarly adoptersEvolving field
Yet, they are addressing medical imaging challenges
Standards and tools arisingSome successful applications deployedThe tool creates the need: new applications will emerge from grid infrastructures
Need to federate the biomedical communityStrengthen the communityDeploy large scale infrastructures
Need standardsTo enable flexibility and interoperability
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 22
Meeting follow-up
http://www.i3s.unice.fr/~johan/miccai/
MICCAI tutorial on grid services for medical image processing and registration Johan Montagnat 23
Agenda
Grids, a tool for compute and data-intensive medical imaging applicationsJohan MontagnatMammography analysis on gridsMike BradyBiomedical Informatics Research NetworkRon KikinisIXI e-Science projectDerek HillCoffee breakThe case of medical image registrationDaniel RueckertRound table: our vision of grid technology for medical image registration, trends and problems, opened discussion, call for participationX. Pennec, D. Hill, J. MontagnatLunch
9h
9h30
9h50
10h10
10h3010h50
11h20
12h20