-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Philippe

It used to be possible to download a demonstration version of ORCA
from Erasmus University in Rotterdam, but they are redesigning their
site and it does not seem to be available at the moment.

> I don't know ORCA, but what I know is that
>
> - if I have been able to store in a fully structured way a whole
> echographic or endoscopic report
> - if from these data I can elaborate a natural langage report with
> a sufficient quality for the doctor to sign it (in more than 80
> hospitals now) 
>
> I would be very stupid if I can't find what I want in these datas
> for decision support purposes.

Even highly structured data is not always suitable for all forms of
analysis. For example, many natural language analysis software
implementations produce highly structured semantic and syntactic
structures. But you can't use them for the kinds of ontological
decision support reasoning supported by, for example, GALEN and
neither have we found a way to map or transform them such that you
can. 

At this point please be assured that I am not suggesting for an
instant that it is a requirement that captured clinical data must be
suitable for description logic based ontological processing, such as
is provided by GALEN or SNOMED CT. It remains not yet proven that
this technology is either useful or ultimately scalable. However, I'm
just raising the point that, if the data captured by Odyssee is
indeed similar to that captured by ORCA, then our experience suggests
that there may be significant problems ahead.

The principle problem we encountered with ORCA was that although
individual nodes in the user interface navigational tree structure
have unique names, the meaning of each node (and often also the links
either side of it) is not constant but changes depending on which
path it finds itself in. The best example of this would be the
terminal qualitative adjectival nodes, like 'soft', whose conceptual
representation is different when preceeded by the prior node 'faeces'
as opposed to the node 'heart murmur'. One signifies a measure of
physical consistency, the other a measurement of audible tonality
and/or volume, and this distinction may be important to certain
decision support rules. We found many other examples. Other problems
encountered included nodes whose presumed semantics included logical
constructs that are not currently supportable (e.g. disjunctions),
and collections of nodes where negation was used inconsistently,
making logical implementation difficult or impossible.

The major problem described, of changeable nodal semantics (and often
underspecified or unspecified link semantics), does not seriously
affect the ability of the user to use the interface, because they
bring the most complex domain, context and language models of all
with them and this allows them to perform the necessary
disambiguation. Similarly, human readable text synthesis from the
structure is possible because the semantics missing in the original
structure can safely remain missing in the generated text. The reader
will bring the missing models with them when they read the text, just
as they do when they read the structure.

Effectively for machine processing, however, where these rich domain
and context models are missing, the true total conceptual meaning of
an individual path may not be *reliably* computed by reference to
concepts assigned to the nodes or links of which a path is
constructed. While it is perhaps possible that the ontological
meaning of *some* paths might be computed from their substructural
nodal mappings to an ontology, we don't know whether it is possible
to automatically identify in advance which paths are the dangerous
ones. 

The alternative situation is that, in order to migrate a recorded
path into an environment in which ontological processing occurs, it
is necessary to enumerate every possible path that the interface
could ever generate and then manually declare the correct mapping for
each individual path to a concept within the external ontology. If
this is the case, then the recorded structure of a path remains
important in its role as part of the unique identifier of that path,
but it is not useful in deciding how that path as a whole relates to
concepts in some remote ontological system.


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dr Jeremy Rogers MRCGP
OpenGALEN

[EMAIL PROTECTED]
http://www.opengalen.org

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 6.5.3 for non-commercial use <http://www.pgp.com>

iQA/AwUBO7i9tlLurXGRoC3KEQImMACg7wzta3UxiUMbCmS1FRqLNhfKziQAoODt
CiLPHhmQldaAG/SPfDqEVMlG
=3Qr3
-----END PGP SIGNATURE-----


Reply via email to