Linkage of image databases with other forms of biological knowledge
Curtis Langlotz (Leader)
Allan Johnson (Co-leader)
Software Design Metaphor
Definitions and context
3 "Use cases"
What linkages are required
Types of Imaging Data
Clinical images (humans)
Small animal images
Histology and pathology
3D chemical structure
Other microscopy (confocal, EM, etc.)
In situ hybridization
Non-Imaging Biological Data
Thesauri, lexicons, vocabulary, ontology
Medical literature, entrez databases
Physiologic models and biochemical pathways
Electronic medical records
Protein classification and function
Normal human and visible human
Use Case A:
From a Clinical Radiographic Image
What is the diagnosis given imaging features?
What are the radiographic features to look for, given a specific diagnosis?
What are the most salient items from the electronic medical record?
What is optimal patient management/treatment?
What is the molecular biology of the disease?
What is the relevant medical literature?
What are relevant areas for further research?
Use Case B:
From an image of a tumor model in a small animal
What is the connection to the human model?
What is the connection to gene expression?
What correlative information is available from other images (histology, EM, histochemistry)?
What is the gene expression and distribution in humans?
Use Case C:
Based on a population of images of a specific disease
More information on the disease or phenotype?
Similar images from other databases/tumors?
What human populations (as identified by geography, genetic markers, body shape, etc) are susceptible?
Use Case Requirements
Feature extraction: pixels features, concept (eg, putamen, mass effect)
Common ontology/thesaurus of data elements, features, and concepts
Problem solving workflow
Automatic retrieval of similar images based on texture, shape, position.
The BIRN Model
Clinical Images connect to Concept Assignment connects to Other Biological Knowledge which may be in a wrapper; and Concept Assignment connects to UMLS
Database Linkage Model
Clinical images connect to Database semantics connect to Ontology/thesaurus while Other biological knowledge connects to another Database semantics and thus to the Ontology/thesaurus
Barriers to Linkage (1)
No common and complete imaging data elements
None of the databases talk the same language: different identifiers for the same concept
None of the radiologists talk the same language: different meanings for the same concept
People and computers talk different languages: (eg, "mass effect")
Barriers to Linkage (2)
Semantics (and APIs) of many existing databases are unpublished--reverse engineering required
No thesaurus/ontology for non-clinical data (genes, proteins), internal creation and maintenance of a thesaurus required
Support efforts development and curation of imaging common data elements
Support efforts to enhance richness of CDEs with description logic and representative images
Support harmonized thesaurus/ontology of non-imaging data elements (UMLS for gene and protein data)
Support development of tools to retrieve by image content