image query (iq) project update building queries one question mark at a time march, 2009
TRANSCRIPT
Image Query (IQ) Project Update
Building queries one question mark at a time
March, 2009
Daniel Rubin M.D., M.S.Assistant Professor of Radiology
Research Scientist, Center for Biomedical Informatics Research
Stanford University
David S. Channin, MDAssociate Professor of Radiology
Northwestern University
Joel Saltz, MD, PhD and OSU TeamProfessor and Chair
Biomedical InformaticsThe Ohio State University
Brenda Young, BAAmerican College of Radiology
Imaging Network (ACRIN)
Presenters
Image Query (IQ) Project
• Researchers need to intuitively search caBIG resources• Holy Grail: Query across any and all models across any and
all grid services for those models• Current caBIG infrastructure does not focus on query
• Imaging Information: NCIA Model and in AIM Model• NCIA Model models some DICOM images meta-data
• Middleware provides access to images via this model
• AIM models annotation and markup data• Need to be able to query and retrieve image (NCIA Model) and
image annotation (AIM Model) data by many different criteria
Some info is in NCIA Model(DICOM metadata)
Some info is in AIM Model
Anatomic Entity: Left Lung (Radlex:1326)
Anatomic Entity: Upper lobe of left lung (RID1327
Observation: Mass (RID:3874)
Characteristic: Microlobulated margin (RID5712)
Geometric Shape: Polyline
2D coordinates: {(x,y),(x,y)….}
Calculation: Largest diameter
result: 2.8 cm
Why do we need Image Query?
• Retrieve data in from caGrid Services
• Use Case: query NCIA
• Retrieve data accessible via caGrid to a DICOM Workstation
• From NCIA images (DICOM header attributes)• From “NCAA” (AIM metadata)
Overview of Query Formulation Project
• Purpose: Create query formulation/execution engine for images on caGrid
• Will show: Phase II Plans and initial work• New developments: selected a use case for
query based on LIDC and NCIA• Biggest challenges: Limited budget for Phase II;
need to scope project• Plans: Will create working demonstration of query
formulation/execution on caGrids
IQ Phase II
• Develop Working Prototype
• Working prototype of query federation and execution components of the IQ Tool
• Simple GUI for user to construct a cross-domain query• Targets AIM and NCIA data services
• Retrieves the relevant image metadata and associated annotations
• Will use high-performance data transport, and leverage role-based authorization
• Will leverage the In Vivo Imaging Middleware and the caGrid federated query processing infrastructure
QF Project Deliverables
• GUI for users to create queries intuitively
• Query Formulation Engine translating user query to a DCQL query that runs on caGrid
• Query Execution Engine that processes the query and retrieves images
• Demonstrate concrete use case using NCIA data and AIM Grid Service
Project Tasks
• Collect/define use cases: Focus on current data in NCIA; selected one particular use case for querying NCIA data
• Define a single canonical query graph structure: Initially support just one graph
• Develop GUI for users to select query attributes: to enable users to specify the query attributes
• DCQL Query of DICOM header and AIM data (DCQL = query language for caGRID)
• Execute DCQL on NCIA and AIM Data Services: The AIM Data Service (“NCAA”) will store AIM image metadata and annotations
• Send retrieved AIM & DICOM objects to user
A Miracle Occurs….
Use Case: NCIA and LIDC
• Query:
“Find all images with slice thickness <= 1.5mm showing lung nodules < 3 cm diameter and that have a spiculation rating of at least 4”
• Query Parameters:
• Slice thickness <=1.5 mm (DICOM header)• Lung nodules (AIM)• Size < 3 cm in diameter (AIM)• Radiologists’ subjective rating spiculation observation
characteristic >= 4 (AIM)• NB: This query requires searching two federated resources
• NCIA Model data in NCIA (from DICOM header)• AIM Model data in Annotation Archive
Federated Query
caGrid Selected
Search Expression
Search Expression Detail
Example Canonical Query Graph
Lung Nodule
DICOM Image
Spiculation
“<= 1.5 mm”
OBSERVATION
DISEASE
CHARACTERISTIC
DISEASE IMAGING
has SOPInstanceUID
has ImagingObservation
has ImagingObservation Characteristic
DICOM Image
has SliceThickness
has Rating
>=3
has Size
>=4
“Find all images with slice thickness <= 1.5mm showing lung nodules < 3 cm diameter and that have a spiculation rating of at least 4”
Ontology Driven Query Process
• Query is constructed in the Query Formulation UI using semantically meaningful and ontology anchored concepts
• Query is represented as SPARQL and submitted to the DCQL translator, along with the target Data Service URLs
• DCQL translator generates DCQL from SPARQL, and return it to Query Formulation UI
• DCQL is executed using a local instance of Federated Query Processing (FQP) Engine
• FQP Engine converts DCQL into CQL queries and coordinates the execution of the CQL queries against the target Data Services
• FQP Engine retrieves the results (images and annotations) and return to the Query Formulation UI
• Query Formulation UI displays the retrieved images and annotations
Ontology Driven Query Process
Query Translation
and Processing
Query TranslationQuery Translation
Query
Execution (FQP)
Query
Execution (FQP)
Grid DICOMGrid DICOM
Grid AIMGrid AIM
DCQL
User InputUser Input
DCQL
AIM/
DICOM
CQL
CQL
SPARQL
OntologiesOntologies
Query Execution Engine
Once a query is formulated as an ontology-based query graph, this query must be translated in such a way so that it can be executed on the
diverse caBIG data sources. This is done in the Query Execution Engine
Summary of Phase II
Create prototype Image Query tools and infrastructure•Application ontology
• Represent kinds of information that users seek and granular data fields actually contained in various image-related databases
•User interface
• Allow users to select data elements and data element values and combine them with Boolean operators
•Query engine
• Execute the query that is formulated by the UI and application ontology
Questions?