Information Technology Reference
In-Depth Information
this module identifies only
tifies the BP mentions whi
nents also identify HTN re
identified that "ht" is used a
analyzes the context of “ht
tences and then filters out
tences, like "His ht was 16
identified as height and hyp
output from the HTNSystem
tated text, and text boundari
CUIs relevant to HTN concepts. The BP value filter id
ich are higher than user specified BP values. The com
elevant abbreviations based on context. For example,
as abbreviation for both height and hypertension. This fi
t” based on preceding and succeeding word tokens in s
cases that were not HTN. For example, in the two s
5cms" and "He was diagnosed with ht" where “ht” can
pertension respectively using our context-based filters. T
m is written to a CSV file which contains filename, an
ies for each of the identified and inferred HTN mention
den-
mpo-
we
ilter
sen-
sen-
n be
The
nno-
s.
Fig. 2. Sa
ample Ruta Script used to identify BP values
3
Results
The performance metrics o
test set is presented in Tab
gold standard annotations p
nizers. As shown in Table 3
sion. This is mainly due to
mentions and BP values, w
record might include HTN
not be a representation of th
We performed an extens
tified improvement gaps. T
tual information, resolving
of our previous system in extracting HTN information
ble 3. These evaluation results are calculated based on
provided by the 2014 i2b2/UTHealth Shared-Tasks or
3, our previous system had acceptable recall but low pre
the fact that the system identified a large number of H
which are ignored by the annotators of i2b2. For exampl
N information mentioned under family history, which m
he patient's HTN information.
ive manual error analysis on the previous system and id
The gaps for improvement included leveraging on cont
disambiguation in abbreviations and additional diction
n on
the
rga-
eci-
HTN
le, a
may
den-
tex-
nary
Search WWH ::




Custom Search