Implementing an RDF Schema for Pathology Images, From the Association for Pathology Informatics

1
Implementing an RDF Schema for Pathology Images,
From the Association for Pathology Informatics

• Jules J. Berman, Ph.D., M.D.
• APIII, Pittsburgh, PA
• Monday, September 10, 2007
• 730 am 830 am

2

• Pathology images have no value unless they are
  annotated with information that describes the
  image.

3

• Important descriptors of an image might include
• File information
• Image capture information
• Image format information
• Specimen information
• Patient information
• Pathology information
• Region of interest information

4

• The API (Association for Pathology Informatics)
  wants to provide anyone using pathology image
  data with optional methods for annotating any
  kind of pathology image, in any image format they
  prefer.
• We did not want to create yet another new
  standard that obligates people to use a
  particular image format.
• Yet, we want to provide methods that could be
  understood by colleagues using existing, free
  standards for specifying data.

5

• From 2004-2007, the API sponsored LDIP, the
  Laboratory Digital Imaging Project, which
  consisted of API members and imaging software
  developers.
• The original purpose of LDIP was to develop a
  new, open data specification for pathology
  images.
• LDIP had monthly conference calls, and the
  minutes of their discussions are available for
  anyone to review.

6
In 2007, after much discussion, the API Council
determined that there were, in existence,
adequate methods for annotating images. LDIP was
dissolved, and the API Council accepted the
primary goal of providing the field of pathology
informatics with a document that describes
available open annotation methods. As a
secondary goal, the API would provide a very
short RDF Schema that would permit those who
prefer RDF annotations to type their metadata
under general classes and properties that have
particular relevance to pathologists (more about
this later).
7
A technical white paper (by Jules Berman and
Bill Moore) that contains detailed methods for
annotating images is published today at
www.julesberman.info/rdfimage.pdf This paper is
distributed under an open source license, and can
be downloaded, copied, re-distributed, and even
re-posted at other web sites.
8
The paper describes methods for 6 levels
(organized by increasing difficulty and
complexity) of image annotation. The methods use
existing standards (including RDF, jpeg, exif,
Dublin Core, XML Schema, W3C Semantic Image
Annotation) and do not create any new standards,
just one new very short RDF Schema document.

9
Level 1. Simply composing a free-text
description of your image and any other
information you'd like to add, such as your name,
and adding the information as a Comment field in
the header of the image file. The Comment will
not alter the binary content of the image or the
visual form of the image. When the file is
copied, it will retain the header comment, and
anyone receiving the image can read what you've
added, using a simple Perl or Ruby script
provided in the document, or using a simple
extraction program prepared in any preferred
programming language.
10
Level 2. Insert the Dublin Core file descriptors
into your Comment. The Dublin Core is basic
information designed by librarians to provide a
minimal set of data to describe the contents of
an electronic document. When the file is copied,
it will retain the Dublin Core metadata, and
anyone receiving the image can read what you've
added, using a simple Perl or Ruby program
provided in the document, or using a simple
extraction program prepared in any preferred
programming language.
11
(No Transcript)
12
Level 3. Insert an RDF (Resource Description
Framework) document into your image file. The
RDF document can be extracted, and the triples in
the document can be extracted and integrated with
other data.
13
(No Transcript)
14
All data can be specified using RDF, developed by
the W3C. RDF files are collections of
statements expressed as data triples ltidentified
subjectgtltmetadatagtltdatagt Jules Berman blood
glucose level 85 Mary Smith eye color
brown Samuel Rice eye color blue Jules
Berman eye color brown When you bind a
key/value pair to a specified object, you're
moving from the realm of data structure (i.e.,
XML) into the realm of data meaning.
15
RDF permits data to be merged between different
files
Medical file Jules Berman blood glucose
level 85 Mary Smith eye color
brown Samuel Rice eye color blue Jules
Berman eye color brown
Merged Jules Berman database Jules Berman
blood glucose level 85 Jules Berman eye
color brown Jules Berman hat size 9
Hat file Sally Frann hat size 8 Jules
Berman hat size 9 Fred Garfield hat size
9 Fred Garfield hat_type bowler
16
Level 4. Insert your image into an RDF document.
The image can be extracted from the RDF
document.
17
Level 5. Point to your image file from an RDF
document. The RDF document and the image file
(for example jpeg) can be separate documents
linked by URLs.
18
Level 6. Break up your annotative data and your
image binaries into multiple documents that can
be pointed from any of the files and that can
exclude or include RDF or image binary data as
desired. The RDF data can be distributed into
multiple documents, and each RDF document may
point to more than one image file.
19
By annotating our images, we can ensure that the
image conveys meaning and value By using RDF, we
can ensure that the individual triples can be
integrated with heterogeneous data sources beyond
those of images. By using pre-existing
international standards for describing data, we
attain interoperability and avoid the confusion
and complexity that occurs whenever a new
standard is created. See www.julesberman.info/
rdfimage.pdf