RE: [MM] use case

Dear Jacco,

Here is an updated use case. Suzanne will send an example image and the RDF
metadata file shortly.

Kind regards
Jane

------------------------------------------------------

Images constitute a primary data source in medicine - originating from
diagnostic technologies (ultrasound, magnetic, fluorographic, radiological),
surgical procedures, pathology (e.g. light and electron microscopy) and
research laboratories. 

For example, researchers at the Institute for Molecular Bioscience of the
University of Queensland, are studying 2D cross-sections of pancreatic cells
in oder to try to create a 3D image of a cell that will enhance
understanding of cellular processes [1]. To this end, pancreatic cells are
cut into 400-nm-thick slices and each slice is studied using electron
tomography techniques. 

Currently molecular biologists have to manually segment and annotate the
images of these slices by drawing lines around and labelling each cellular
component. Hundreds of these 2D images need to be annotated and then
combined to generate a single, high resolution three-dimensional
reconstruction of a pancreatic cell. The significant cellular components
that need to be identified include: the golgi apparatus, endoplasmatic
reticulum, mitochondria, ribosomes, and different types of vesicles. A new
high-throughput microscope will soon be employed that is capable of
producing even larger numbers of images, heightening the need for automatic
segmentation and annotation.

At DSTC/ITEE, we have developed a novel Rules-By-Example (RBE) [2] approach
that enables semi-automatic annotation of image regions from
automatically-extracted low-level features. The RBE prototype enables
experts to define rules specific to their domain, which map particular
combinations of low-level visual features (colour, texture, shape, size
etc.) to high-level semantic terms defined in their domain ontology. The
system assists users to construct rules such as: IF [(color is like this )
AND (texture is like this ) AND (shape is like this )] THEN (the object is a
ribosome) through a Graphical User Interface that provides palettes of
colours, shape-drawing tools and examples of textures. More specifically we
are employing:

-  the Matlab Image Processing Toolbox to extract low level features which
are then represented using terms defined in an MPEG-7 ontology;
-  the Medical Subject Headings thesaurus (MeSH) for labelling cellular
components;
-  RuleML to represent the semantic inferencing rules that relate
combinations of low-level visual features to cellular components.

The semi-automated semantic annotation of the cross-sectional images of
pancreatic cells, enables rapid reconstruction of 3D cellular models and
sophisticated querying over large image sets, in terms familiar to molecular
biologists. In the long term, such tools are expediting scientists'
understanding of cellular processes, in order to improve drug design.

[1] Molecular Cell Biology at the IMB
http://www.imb.uq.edu.au/index.html?page=12015&pid=11668

[2] S. Little and J. Hunter "Rules-By-Example - a Novel Approach to Semantic
Indexing and Querying of Images" 3rd International Semantic Web Conference
(ISWC2004). Hiroshima, Japan, November 2004.
http://maenad.dstc.edu.au/papers/2005/hollink-kcap05.pdf

[3] L. Hollink, S. Little, and J. Hunter. "Evaluating the Application of
Semantic Inferencing Rules to Image Annotation" Proceedings of the Third
International Conference on Knowledge Capture, KCAP05. Banff, Canada.
October 2005. http://maenad.dstc.edu.au/papers/2004/iswc2004-rbe.pdf


> -----Original Message-----
> From: Jacco van Ossenbruggen [mailto:Jacco.van.Ossenbruggen@cwi.nl] 
> Sent: Friday, 21 October 2005 5:23 PM
> To: Jane Hunter; 'swbp'
> Subject: Re: [MM] use case
> 
> 
> Jane,
> 
> Thanks for the use case.  I like the topic, but have two main 
> comments.
> 
> - I think the description is a bit too broad, covering related issues 
> from bioinformatics and image data mining.
> I would like to stick to the issues that relate directly to image 
> metadata and annotation, and phrase it a way that it describes a 
> concrete problem for which current (SemWeb) technologies have 
> (at least) 
> a partial answer.
> - We like to link to example solutions for each use case.  So I would 
> like to have one or more medical images with associated RDF or OWL 
> metadata, and a description about how and with what tools these 
> annotation have been made, and what other software can be 
> used to take 
> advantage of the metadata. 
> The goal is to provide technical insight ("how do the angle brackets 
> look like")  but also some insight in what the costs are and what the 
> pay off is.
> 
> Do you think you have concrete examples from one of your research 
> projects that we can use to explain how SemWeb helps in solving this 
> type of use cases?
> 
> Jacco
> 
> Jane Hunter wrote:
> 
> >Will this do?
> >
> >----------------------------------------------
> >Images constitute a primary data source in medicine. Images 
> originate 
> >from diagnostic technologies (ultrasound, magnetic, fluorographic, 
> >radiological), surgical procedures, pathology (e.g. light 
> and electron 
> >microscopy) and the research laboratory. Moving images are 
> important in 
> >diagnostic radiology and cardiology, and are being 
> investigated in the 
> >treatment of psychosis. Medical training relies increasingly 
> on images, 
> >with the "visible human" and physiome projects as key examples. With 
> >the spread of telemedicine, images can be transmitted for remote 
> >collaborative interpretation.
> >
> >Medical image bioinformatics refers to the technologies, standards, 
> >protocols, algorithms, software, data structures, analyses 
> and systems 
> >by which the rich information in an electronic image can be 
> integrated 
> >into the broader knowledge fabric of molecular bioinformatics, 
> >including associated computational tools and database resources. The 
> >aim being to improve diagnostic systems for clinicians. Key 
> issues and 
> >challenges in medical image bioinformatics include:
> >
> >*   Automated storage, indexing, object recognition, 
> segmentation, semantic
> >labelling and markup of images. This requires automated 
> workflows, and 
> >tools for fast, efficient metadata capture and generation, 3D 
> >reconstruction of images and representation of non-visual 
> output from 
> >imagers, development of domain-specific ontologies, and 
> generation of 
> >semantic annotations and inferencing rules.
> >
> >*   Information integration and correlation. Because the 
> pertinent data are
> >too large and diverse for human correlation, knowledge 
> management and 
> >mining services are required that can efficiently organise, 
> discover, 
> >pre-process, correlate, reason-over and integrate medical imagery 
> >(X-ray, CT, MRI) with molecular sequence, structure, SNP, haplotype, 
> >expression microarray, pathway and other bioinformatic databases, 
> >medical and biomedical literature, and patient and other healthcare 
> >records.
> >
> >*   Innovative presentation and visualisation interfaces to assist
> >clinicians with decision-making. These must enable sophisticated 
> >search, browse and data exploration, fast similarity searches, 
> >statistical analysis, computational modelling, hypothesis 
> formulation, 
> >and multi-modal visualisation over large data volumes.
> >------------------------------------------------
> >
> >
> >  
> >
>

Received on Tuesday, 15 November 2005 05:23:27 UTC