Information Technology Reference
In-Depth Information
Fig. 2.
Part of the tree resulting of parsing the ADL file shown in Fig 1
Mapping Through Metathesaurus and Metamap.
Our approach sequen-
tially uses different lexical matching techniques [14] until some mapping is rea-
ched. We initially applied an
Exact search
and, in case of no results we tried
approximate techniques:
Truncate searches
,
Normalized word
searches and
Meta-
Map
(see Algorithm 1). Through this, we try to achieve a high recall without
excessively penalizing the precision. Other types of searches, such as
Normalize
String
or
Approximate Match
, have been discarded because we have checked that
these worsen the results.
The next step involves getting the SNOMED CT concepts associated to the
recovered UMLS concepts. Furthermore, the algorithm filters the SNOMED CT
concepts belonging to the hierarchy
Observable Entity
, as in the work, the se-
lected archetypes register observations. Finally, if two candidate SNOMED CT
concepts share the same parent concept, this is added to the list of candidates.
This optimization slightly increases the recall.
Algorithm 1.
Archetype Root Node Mapping using Metathesaurus and
MetaMap
FindSnomedConceptsForRootNode
(Node rootNode)
begin
1
terms ←− { rootNode.text, rootNode.text
+“
observable
, archetypeName}
2
candidateCUIs ←−
FindByExactMatchingInMetathesaurus(
terms
)
3
if
|candidateCUIs|
=0
then
4
candidateCUIs ←−
FindByTruncateInMetathesaurus(
terms
)
5
if
|candidateCUIs|
=0
then
6
candidateCUIs ←−
FindByNormWordAndMetaMap(
terms
)
7
candidateSnomedConcepts ←−
getSnomedConcepts(
candidateCUIs
)
8
candidateSnomedConcepts ←−
addSharedFatherConcept(
candidateCUIs
)
9
return
candidateSnomedConcepts
10
Direct Mapping to SNOMED CT.
Our approach sequentially combines two
basic types of matching methods classically used in ontology matching [15] in
order to produce the alignment. First, we applied a “named-based method” that
Search WWH ::
Custom Search