the role of metadata in querying grid-resident medical images · ist mmgps 16/12/2003...

The Role of MetaData in Querying Grid-Resident Medical Images

Tamás HauerUniversity of West England – CERN

IST Workshop on Metadata Management in Grid and P2P Systems16th December 2003

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 2/24

Outline

The MammoGrid projectMeta-Data and domain ontology (flexibility)Meta-Data, services and grid (openness)Query negotiatorOutlook

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 3/24

Outline

The MammoGrid projectMeta-Data and domain ontologyMeta-Data, services and gridQuery negotiatorOutlook

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 4/24

The MammoGrid Project

A grids solution for mammographyEU-funded, 2002-2005– CERN, Mirada Solutions– Universities: UWE, Oxford, Pisa, Sassari– Hospitals: Addenbrookes (Cambridge), Policlinico Universitario, Udine

Proof-of-concept R&D:– Grid + medicine– Pan-European distributed X-ray image database– Possible target areas:

• Education• Quality control• Epidemiology research• Standardized screening• Communication aid

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 5/24

GRID

Federated System SolutionHospital Italy

Healthcare Institute

University Database

Hospital UK

Shared meta-dataAnalysis-specific data

•Knowledge is stored alongside data•Active (meta-)objects manage various versions of data and algorithms•Small network bandwidth required

Clinician’s Workstations

QueryResult

LocalQuery

LocalAnalysis

LocalAnalysis

LocalAnalysis

LocalAnalysis

Massively distributed dataAND distributed analyses

LocalQuery

LocalQuery

LocalQuery

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 6/24

Outline

The MammoGrid project

Meta-Data and domain ontologyMeta-Data, services and gridQuery negotiatorOutlook

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 7/24

The Medical Domain

Requirements:– User community:

• process-oriented rather than information-oriented• Heterogeneous (technician, radiologist, researcher,…)

– Research area hard-to-predict requirements– Regional differences– Constantly changing knowledge-base

Flexibility, Extensibility is top priorityManagement of domain information

Flexibility, Extensibility is top priorityManagement of domain information

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 8/24

The Medical Domain

Entities– Patients, Physicians, Hospitals, Equipment, …– Medical History (Drugs, Surgery, Pregnancy,…)– Studies, Images, Pathology results, …– Annotation, Diagnosis, Treatment, …– Treatment protocol, screening, … (workflow)– Epidemiological study, research, …

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 9/24

The Medical Domain

Patient X

History At Present PerformedStudy

Mother’s breast cancer

Three pregnancies

On Hormone Replacement

Weight=60kg

Smoking

Four Mammograms

Annotation: lump in right breast

Diagnosis:Possible cancer…

… Treatment:Lumpectomy

Define semantics so that elements are queriable

Bridge discrepancies between different standards

Associations between pixel data of digitized images and structured information

Extension to other modalities

CAD, “Find-one-like-it”

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 10/24

The Medical Domain

Typical workflow:

1. Register patient, scan image, add pathology data2. Run SMF to bring image(s) to standardized format (possibly

automatic)3. Annotate image (define location, nature, size, etc… of

finding(s) )4. Run CAD5. Define relationship between findings (temporal, spatial and

pathological)6. Write assessment of image, series, study. Associate to

findings

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 11/24

The Medical DomainImage Centric Approach:

•Digitized images (e.g. raw file)•Semistructured information about the image•Associate non-derived information (make it useful and interesting)

ImageDerived Data:Size, colormap…CAD, SMF

Associated Data:Patient, Study, Equipment, …Related ImagesAnnotation, Diagnosis, …

Meta-Data:Semantics of derived Entities

Meta-Data:

Storage format, …

Meta-Data:

Semantics of the associated Entities

Meta-Data: semantics of derivations and relationships

•Describe the image format•Define “classes” of derived and associated information•Describe the relationships between the entities (of a record, of records)

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 12/24

Outline

The MammoGrid projectMeta-Data and domain ontology

Meta-Data, services and gridQuery negotiatorOutlook

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 13/24

Meta-Data, services and grid

Requirements:– Data heavy (20MB/image) – storage, bandwidth– CPU heavy (image processing: CAD, SMF,…)– Geographically distributed, inhomogeneous,…– “Plug-and-play” discovery– Scalability, adaptability– Confidentiality

•Service-oriented architecture•Use grid-middleware

•Service-oriented architecture•Use grid-middleware

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 14/24

Cambridge Hospital

Udine Hospital

Oxford University

CERN

Meta-Data services and grid

Udine Hospital:•I store mammograms•You can talk to me via SOAP or DICOM•I can run algorithms (CAD) for you•You need to use XXX authentication format (so that I can verify your site and country)•I authenticate myself with XXX or YYY format•You can subscribe to …

Meta-Data describes service definition

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 15/24

Cambridge Hospital

Udine Hospital

Oxford University

CERN

Meta-Data services and grid

Oxford University:•I do not have storage•I understand SOAP only•I authenticate myself with XXX or YYY format

Meta-Data describes service

definition

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 16/24

Cambridge Hospital

Udine Hospital

Oxford University

CERN

Meta-Data services and grid

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 17/24

Meta-Data services and grid

Grid metadata defines service ochestration mechanisms.

Follows OGSA specification

Implements OGSI:•Service handles•Lifetime management•Represents agreements

Grid data services:

Data VirtualizationGrid Data Interfaces

•DataDescription•DataAccess•DataFactory•DataManagement

Implemented by Data Services

•Flexible: plug-and-play system: each participant is responsible for managing its own metadata and can change its service description on-the-fly. •Extensible: New sites can seamlessly join.•Domain and Service ontology are defined independently

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 18/24

Outline

The MammoGrid projectMeta-Data and domain ontology (flexibility)Meta-Data, services and grid (openness)

Query negotiatorOutlook

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 19/24

Query Negotiator

Issues– Data location, discovery– Query caching– Bandwidth, network, etc.– Do I need to run an algorithm

for this?

??

What belongs to the negotiator?

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 20/24

Query NegotiatorClient

QueryTranslator

Query Handler

Result Handler

User’s Terms&

Domain MetaData

Local Database

Stored QueryDatabase

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 21/24

Query NegotiatorClient

QueryTranslator

QueryAnalyzer

LocalQuery Handler

RemoteQuery Handler

Result Handler

User’s Terms&

Domain MetaData

Local Database

Stored QueryDatabase

GRIDGRID

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 22/24

Query NegotiatorQuery

Translator

QueryAnalyzer

LocalQuery

Handler

RemoteQuery

Handler

Result Handler

QueryTranslator

QueryAnalyzer

LocalQuery

Handler

RemoteQuery

Handler

Result Handler

GRID

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 23/24

Outline

The MammoGrid projectMeta-Data and domain ontology (flexibility)Meta-Data, services and grid (openness)Query negotiator

Outlook

IST MMGPS 16/12/2003 tamas.hauer@cern.ch 24/24

Outlook

Within the Mammogrid project we focus on three distinct areas where active metadata management is usefulThese provide key features such as flexibility, extensibility and efficiencyThe three meta-data management facilities work in an orchestrated fashion to define the overall characteristics of the Mammogrid system.Later it will be interesting to define interfacing two grid-enabled projects (Mammogrid and eDiamond?)To that end it might turn out that there is room for a further meta-layer which defines (and unifies) our three (and possibly more) meta-data concepts.

Thank You

Tamas.Hauer@cern.ch