Hi,
I am using the cTakes 3.2.0 code base and I have been trying to figure out
what would be the proper way to get cTakes to recognize and annotate
mentions of medical devices.
I am using the AggregatePlaintextUMLSProcessor.xml because one of the main
requirements for the annotation is that it needs to to include subject, and
polarity, certainty, etc.
First, I added TUI T074 to the procedureTuis in the LookupDesc_Db.xml. The
output from this provides mostly what I want, except it lumps the devices
as ProcedureMention where I would like them to be distinguished by their
own annotation.
An example annotation created this way was:
<org.apache.ctakes.typesystem.type.textsem.ProcedureMention
_indexed="1" _id="35574" _ref_sofa="3" begin="472" end="481" id="32"
_ref_ontologyConceptArr="35569" typeID="8" segmentID="SIMPLE_SEGMENT"
discoveryTechnique="1" confidence="1.0" polarity="1" uncertainty="0"
conditional="false" generic="false" subject="patient" historyOf="0"/>
I also tried adding code to classify devices as an EntityMention, and that
seemed to work too:
<org.apache.ctakes.typesystem.type.textsem.EntityMention _indexed="1"
_id="35574" _ref_sofa="3" begin="472" end="481" id="32"
_ref_ontologyConceptArr="35569" typeID="8" segmentID="SIMPLE_SEGMENT"
discoveryTechnique="1" confidence="1.0" polarity="1" uncertainty="0"
conditional="false" generic="false" subject="patient" historyOf="0"/>
Again, that doesn't give devices their own unique annotation so I looked
further. Exploring the typesystem, I noticed the following types in
TypeSystem.xml:
org.apache.ctakes.typesystem.type.refsem.ProcedureDevice
org.apache.ctakes.typesystem.type.textsem.ProcedureDeviceModifier
These seemed like the closest defined types to what I would expect so I
thought I would see if using them would generate what I wanted. I modified
the code to generate these annotations and the result was as follows:
<org.apache.ctakes.typesystem.type.textsem.ProcedureDeviceModifier
_indexed="1" _id="35587" _ref_sofa="3" begin="472" end="481" id="32"
typeID="0" segmentID="SIMPLE_SEGMENT" discoveryTechnique="0"
confidence="0.0" polarity="0" uncertainty="0" conditional="false"
generic="false" subject="patient" historyOf="0"
_ref_normalizedForm="35574"/>
<org.apache.ctakes.typesystem.type.refsem.ProcedureDevice _id="35574"
id="0" _ref_ontologyConcept="35542" discoveryTechnique="1" confidence="0.0"
conditional="false" generic="false" polarity="0" uncertainty="0"
historyOf="0"/>
The problem with this approach was that confidence, polarity, and
uncertainty did not get filled in. I tried adding these to the inputs and
outputs of the AssertionMiniPipelineAnalysisEngine, but that didn't seem to
have any effect. Perhaps I didn't do it right? or maybe it isn't even the
right pipeline component to try to modify?
Since ProcedureDevice and ProcedureDeviceModifer have different supertypes
than ProcedureMention, I also tried creating a new type in
TypeSystem.xml:
<typeDescription>
<name>org.apache.ctakes.typesystem.type.textsem.DeviceMention</name>
<supertypeName>org.apache.ctakes.typesystem.type.textsem.IdentifiedAnnotation</supertypeName>
<features>
<featureDescription>
<name>entity</name>
<description/>
<rangeTypeName>org.apache.ctakes.typesystem.type.refsem.Entity</rangeTypeName>
</featureDescription>
</features>
</typeDescription>
Modifying the code to create this type, the result was:
<org.apache.ctakes.typesystem.type.textsem.DeviceMention _indexed="1"
_id="35574" _ref_sofa="3" begin="472" end="481" id="32"
_ref_ontologyConceptArr="35569" typeID="8" segmentID="SIMPLE_SEGMENT"
discoveryTechnique="1" confidence="0.0" polarity="0" uncertainty="0"
conditional="false" generic="false" subject="patient" historyOf="0"/>
Again the problem here is that confidence, polarity and uncertainty are not
filled in.
So, I am left wondering:
1) Which of the methods I tried would be the best "cTakes way"?
2) What do I need to modify to get confidence, polarity and uncertainty to
be filled in?
Thanks,
Bruce
