1 / 41

Sharing models for multi- modality image simulation in VIP

Sharing models for multi- modality image simulation in VIP . Bernard Gibaud 1 , Germain Forestier 2 1 IRISA, puis LTSI , U1099 Inserm, Rennes 2 MIPS, Université de Haute-Alsace, Mulhouse. Acknowledgements. Hugues Benoit- Cattin Frédéric Cervenansky Patrick Clarysse

dalit
Download Presentation

Sharing models for multi- modality image simulation in VIP

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. VIP Workshop, December 14, 2012, Lyon (France) Sharing models for multi-modality image simulation in VIP Bernard Gibaud1, Germain Forestier2 1IRISA, puis LTSI, U1099 Inserm, Rennes 2 MIPS, Université de Haute-Alsace, Mulhouse

  2. VIP Workshop, December 14, 2012, Lyon (France) Acknowledgements • Hugues Benoit-Cattin • Frédéric Cervenansky • Patrick Clarysse • Denis Friboulet • Alban Gaignard • Tristan Glatard • Patrick Hugonnard • Carole Lartizien • Hervé Liebgott • Johan Montagnat • Joachim Tabary

  3. VIP Workshop, December 14, 2012, Lyon (France) Overview • Introduction – Goal and scope • Part 1. Methodology • Part 2. Result: OntoVipontology • Part 3. Result: Utilizationin VIP platform • Discussion and conclusion

  4. VIP Workshop, December 14, 2012, Lyon (France) Introduction Goal and scope

  5. http://www.creatis.insa-lyon.fr/vip Context Object model repository Simulated data repository metadata models metadata images Thorax Simulation tools repository US Himan body MRI Heart MR: SIMRI CT: Sindbad PET: Sorteo US: FIELD CT Anatomical and physiological models PET Simulated data Grids et clusters metadata Simul.

  6. VIP Workshop, December 14, 2012, Lyon (France) Goal • Define a commonvocabulary to describeshared information within the VIP platform • Content of the modelsused in simulation • Nature and provenance of simulated data • Use of ontologies • Common vocabulary • Formaldefinitions of entities and relationships, enablingreasoning

  7. VIP Workshop, December 14, 2012, Lyon (France) Scope • Models: • Data files, modeling a 3D scene • eitherobjects(voxelmaps, meshes) • and / or maps of values taken by physicalparameters • Simulated data • Non-reconstructed and Reconstructed data • Simulation actions BrainWeb model Simulated T1weighted images

  8. VIP Workshop, December 14, 2012, Lyon (France) Specificconstraints • A commonframeworkfor all imagingmodalities • X-Ray imaging, e.g., radiographs and computedtomography (CT) • Magneticresonanceimaging (MRI) • Ultrasound (US) • Positron Emission Tomography (PET)

  9. VIP Workshop, December 14, 2012, Lyon (France) Part I - Methodology

  10. VIP Workshop, December 14, 2012, Lyon (France) MethodologySpecial concerns • Reuseexisting ontologies as far as possible • not re-invent the wheel • facilitate future interoperation with biological modelling software • Createontologies whennecessary • ensureadequate documentation • Integrateall components in a consistent whole • use of consistent integration framework

  11. VIP Workshop, December 14, 2012, Lyon (France) Reuse of existing ontologies • FMA (865) • Mouse pathology (494) • PATO* (84) • RadLex (81) • RadLex (49) • RadLex (189) • ChEBI (255) • Anatomy • Pathology • Physicalqualities • Contrast agents • Radiopharmaceuticals • Foreign bodies • Atoms * PATO: PhenotypicAttribute and Trait Ontology

  12. VIP Workshop, December 14, 2012, Lyon (France) Creation of ontologies • Domains • Models(medical image simulation objectmodelsormodels) • Simulated data (simulated data) • Simulation actions(medical image simulation) (Extensive reuse of OntoNeuroLOG) • Tworepresentations • OntoSpecdocuments • OWLimplementation

  13. VIP Workshop, December 14, 2012, Lyon (France) Extract of an OntoSpec document Subsumption relation (« is a ») Essential property / existential restriction

  14. VIP Workshop, December 14, 2012, Lyon (France) Common integrationframework • One single application ontology (calledOntoVIP) • one Foundationalontology, i.e. DOLCE • SeveralCoreontologies • e.g. Actions, Artefacts, Participation roles, Agentives, Discourse message acts, I.E.C. (inscriptions-expressions-conceptualizations) • SeveralDomain ontologies particular (Masolo et al., 2003) perdurant quality abstract endurant physical quality temporal quality region physical object non-physical object event stative state time interval process

  15. VIP Workshop, December 14, 2012, Lyon (France) proposition (non-physical endurant) physical endurant quality is a is a is a physical object physical quality representational object refers to has for quality

  16. VIP Workshop, December 14, 2012, Lyon (France) conceptual action representational object representational object is a is a is a medical image simulation object model medical image simulation simulated data is a data of has for result

  17. VIP Workshop, December 14, 2012, Lyon (France) Part 2. Result: OntoVIPontology

  18. VIP Workshop, December 14, 2012, Lyon (France) OntoVIP : major features • Organization in model layers • Relation to physicalproperties (physicalparameter) • Relation to time (time point, instant) • Taxonomy of models • Taxonomy of medical image simulation • Taxonomy of simulated data • Provenance

  19. VIP Workshop, December 14, 2012, Lyon (France) Organization in layers (1/3) • Layers (model layers) are representationalobjectsthataimatcharacterizing the contents of a model • Values layersdepict the 3D spatial distribution of a parameter • e.g. the map of proton density values • Object layersdepict the 3D distribution of objects • Composed of objectlayerparts • Eachassociated to a label value 3 2 1

  20. VIP Workshop, December 14, 2012, Lyon (France) Organization in layers(2/3) • Values layer • Object layer

  21. VIP Workshop, December 14, 2012, Lyon (France) Organization in layers(3/3)examples anat-layer01 rdf:typeanatomical-object-layer anat-layer01 is-stored-in-file‘anat.nii’ anat-layer01 has-for-proper-partanat-layer-part01 anat-layer-part01 rdf:typeanatomical-layer-part anat-layer-part01 refers-togrey-matter01 grey-matter01rdf:typefma:grey-matter anat-layer-part01 has-for-label-in-model ‘2’ … 2

  22. VIP Workshop, December 14, 2012, Lyon (France) Relation to physicalpropertiesexamples • In the case of values layers… values-layer01 rdf:typephysical-parameter-values-layer values-layer01 refers-toproton-density01 proton-density01rdf:typeproton-density • In the case of objectlayers … • mathematical distribution of physicalquality anat-layer-part01 has-for-physical-parameter-distributionmath-distrib01 math-distrib01 rdf:typegaussian-distribution math-distrib01 has-for-meanmean01 math-distrib01 has-for-standard-deviationstd01 math-distrib01 refers-toproton-density01 proton-density01rdf:typeproton-density

  23. VIP Workshop, December 14, 2012, Lyon (France) Relation to time • Models’ components (such as model layers) refer to discretetime intervals • Twoscales are considered • Time points depict « long intervals », i.e. betweenseveralimagingprocedures (e.g. days, months, years) • whereasinstantsdepict « short intervals » within an imagingprocedure (e.g. seconds, minutes)

  24. VIP Workshop, December 14, 2012, Lyon (France) Taxonomy of models denotes « is a » relationship denotes disjoint classes

  25. VIP Workshop, December 14, 2012, Lyon (France) Taxonomy of simulated data

  26. VIP Workshop, December 14, 2012, Lyon (France) Simulation (actions) • Simulation (actions) • Relation to simulated data • Examples simulation01rdf:typePET-simulation simulation01has-for-resultsimul-data01 simul-data01 rdf:typePET-sinogram

  27. VIP Workshop, December 14, 2012, Lyon (France) Provenance of simulated dataexamples • Relation betweensimulated data and models simul-data01 rdf:typeMR-image simul-data01 derives-from-modelmodel01 model01rdf:type MR-compatible-model • Relation betweensimulated data and parameters and parameter sets simul-data01 derives-from-parameter-set par-set01 par-set01 rdf:typeparameter-set simul-data01 derives-from-parameterparam01 param01 rdf:typesimulation-parameter param01 rdf:typeecho-time-value-information

  28. VIP Workshop, December 14, 2012, Lyon (France) OntoVIPmodular structure OntoVIP VIP-simulated-data VIP-simulation OntoNeuroLOG components VIP-model VIP-foreign-body VIP-biological-object VIP-contrast-agent VIP- radio-pharma-ceuticals VIP-Pathological-anatomical-object-quality VIP-patho logical-object VIP-atom

  29. VIP Workshop, December 14, 2012, Lyon (France) Part 3. Result: Utilizationin the VIP platform

  30. VIP Workshop, December 14, 2012, Lyon (France) Model annotation

  31. VIP Workshop, December 14, 2012, Lyon (France) Model annotation • Ontology-basedreasoningat model upload • Semi-automaticcreation of model layersbased on the nature of referredobjects • e.g. anatomical structure assigned to an anatomicalobject layer • Retrieval of ontologyterms(ranksearchingusinglucene) • Inference of modality compatibility • e.g. a model isMR-simulation-compatible • Control of consistency • Automaticcreation of model.rdf annotation file • It becomes a part of the model • Possibility of upload of an annotated model

  32. VIP Workshop, December 14, 2012, Lyon (France) Search and reuse of models • Modelscanbesearched for based on model contents

  33. VIP Workshop, December 14, 2012, Lyon (France) Annotation of simulated data 1/3(Alan GaignardPhDwork, I3S) • GOAL: automatingthe semantic annotation of simulated data, based on: • detailed provenance data provided by the execution manager (open provenance model) • existingknowledge about models and simulation workflow components (represented as domain-specificinferencerules)

  34. VIP Workshop, December 14, 2012, Lyon (France) Annotation of simulateddata 2/3Example: PET sinogramgeneratedwith SORTEO wf-execution sinogramrdf:typePET-sinogram sinogramis-a-result-of wf-execution wf-executionrdf:typePET-simulation sinogramderives-from-modelphantom sinogramderives-from-parameter-set protocol protocolrdf:typeparameter-set

  35. VIP Workshop, December 14, 2012, Lyon (France) Annotation of simulated data 3/3Example: MR image generatedwithSimuBloch • Provenance data isbrowsablefrom the portal • An alternative to the (relatively ad-hoc) inferencerulesisprovided by conceptualworkflows(PhDwork of Nadia Cerezo, I3S), whichdescribe actions at a more abstract level (work in progress)

  36. VIP Workshop, December 14, 2012, Lyon (France) Part 4. Discussion

  37. VIP Workshop, December 14, 2012, Lyon (France) Reuse of existing ontologies • GnMost termscouldbefoundfromexistingresources: FMA, RadLex, PATO, MPATH, ChEBI • Set of termsselectedbased on a reasonableguess • Hard to seewhetheritissufficient • Selected corpus canbeextendedquiteeasily (vSPARQLquerybased on a set of terms) • Althoughpotentiallyinteresting for reasoning, « Part-whole » relations not retrievedfrom FMA, yet • Need for feedback fromactual use

  38. VIP Workshop, December 14, 2012, Lyon (France) Ontology of models • Organisation in model layersverysatifactory • Intuitive, natural relation to physicalparameters • All objectives not met • Hard to integrate all simulators in a single consistent framework • Physiological and Physiopathologicalprocesses not annoted, yet • e.g., no means to express that a model models ‘breathing’ or ‘heartmovement’  Wouldstillneedsubstantialwork

  39. VIP Workshop, December 14, 2012, Lyon (France) Deployment and adoption • Successful use of ontologies in the platform • Theycanbeused in similarprojects • Wecarefullyavoidedembedding in the ontologies specificconstraints of ourplatform • Stilllimited exploitation of embeddedknowledge • Search of models • Inference of compatiblity to modalities and/or specificsimulators • Exploitation of provenance data for assistingusers in choosing simulation parameters  Possibility to extend the platform’scapabilitybased on users’ feedback and wishes

  40. VIP Workshop, December 14, 2012, Lyon (France) Conclusion • The development of the OntoVIPprojectrequired to find a compromise betweensomewhatcontradictory contraints • Providing a vocabularytailored to the deployment of the VIP platform • Providing a generalenoughontologythatcould support similarneeds in the field of medical image simulation • Only feedback fromactual use (in VIP and elsewhere) willallow us to improveit.

  41. VIP Workshop, December 14, 2012, Lyon (France) Representation of pathologicalentitiespathological-anatomical-object hippo01 rdf:typepathological-anatomical-object hippo01 rdf:typefma:hippocampus hippo01 has-for-qualityqual01 qual01 rdf:typedecreased-volume

More Related