ctakes-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Chen, Pei" <Pei.C...@childrens.harvard.edu>
Subject RE: Type System updates
Date Tue, 22 Jul 2014 14:50:14 GMT
+1 on James' point(s) below

> -----Original Message-----
> From: Masanz, James J. [mailto:Masanz.James@mayo.edu]
> Sent: Monday, July 21, 2014 8:13 PM
> To: 'dev@ctakes.apache.org'
> Cc: 'melissa.tharp@me.com'; 'wendy.chapman@utah.edu';
> 'swu@apache.org'
> Subject: RE: Type System updates
> 
> 1) It's not clear if you are saying a downstream user needs some of those
> elements, or is just pointing out the discrepancy. If the elements are
> needed, then I suggest opening a JIRA item listing what it needed and the
> community would decide how to handle that.  As it is we have some
> elements defined that we don't have a component or algorithm to set/fill-in
> and then people wonder why they are there.  So I suggest we don't add
> things until they are needed/requested for a specific use.
> 
> 2) If a named entity doesn't fit into existing types, and there is code to
> populate such a type(s), I think we should add a new type (once we
> understand it fully).
> 
> II & III) I think you should go for it.
> 
> 
> -----Original Message-----
> From: Wu, Stephen T., Ph.D.
> Sent: Thursday, July 17, 2014 2:30 PM
> To: dev@ctakes.apache.org
> Cc: melissa.tharp@me.com; wendy.chapman@utah.edu; swu@apache.org
> Subject: Type System updates
> 
> The type system needs some updating.  Here are some reasons:
> 
> 1. The refsem and textsem namespaces were created to reflect the
> Secondary Use Clinical Element Models (CEMs) defined in the SHARP project.
> However, later modification to those models are not reflected in the cTAKES
> type system.
> 
> 2. Wendy Chapman, Melissa Tharp, et al (CC'ed here) have done some
> quantitative studies showing that physicians do not easily categorize their
> named entities of interest into cTAKES types.  For example, even if we could
> pick up values (as James mentioned, we don't pick up many), how would you
> categorize an Apgar score?  That kind of thing is not exactly a Procedure, Lab,
> or SignSymptom -- at least, physicians don't seem to think so.
> 
> 3. We have been using the type system for a while and it might be due time
> for some ground-up modifications (though I do think this is more of an
> ongoing task).
> 
> 
> The courses of action we can take are aligned somewhat to these different
> reasons.
> 
> I. Try to reconcile the CEMs with the Type System.  Here is a diff, put
> together by Melissa Tharp:
>         http://bit.ly/WkqCPa
> I feel like this will be quite complicated, especially given the differences
> between Assertion and SignSymptom.  Also, if we add everything from the
> CEMs, that significantly adds to the Modifiers that we have to create to
> house those types.  Each attribute of a CEM may require its own processing
> and evaluation (i.e., you might need a dedicated analysis engine just to
> discover DiseaseDisorder:severity), but in practice there may be too many
> options of types and attributes.
> 
> II. Follow the work being done on physician-validated types.  Melissa might
> be able to put together another document with the differences between
> their resulting types (Schema Ontology) and our Type System.  We could
> then use this to update the types with what physicians are actually looking
> for.
> 
> III. Solicit user/developer-initiated changes, to be made at the same time as
> one or both of the above.
> 
> What does everyone think?
> 
> stephen
> P.S. Please use swu@apache.org or stw4@alumni.duke.edu for future
> correspondence!

Mime
View raw message