Download Negation Detection in Medical Documents Using Syntactical Methods

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
C
Diplomarbeitspräsentationen der Fakultät für Informatik
Masterstudium:
Medizinische Informatik
Negation Detection in Medical Documents
Using Syntactical Methods
Technische Universität Wien
Institute of Software Technology & Interactive Systems
Arbeitsbereich: Information and Software Engineering
Betreuerinnen: Ao. Univ.-Prof. Mag. Dr. Silvia Miksch
und Mag. Dr. Katharina Kaiser
Stefan Gindl
Abstract
Objective
In clinical practice guidelines (CPGs) the medical information is stored in a narrative way. A large
part of this information occurs in a negated form. The detection of negations in CPGs is an important
task since it helps medical personnel to identify not occurring symptoms and diseases as well as
treatment actions that should not be accomplished. We developed algorithms capable of Negation
Detection in this kind of medical documents. According to our results, we are convinced that the
involvement of syntactical methods can improve Negation Detection, not only in medical writings but
also in arbitrary narrative texts.
Development, implementation, and evaluation of algorithms capable of the detection of negated phrases in
CPGs.
Background
Methodology
MetaMap Transfer (MMTx) used for text processing:
 MMTx maps phrases to the UMLS Metathesaurus
 MMTx provides methods for the processing of natural language (see Fig. below), such as part-of-speech
(POS) tagging
● Development of methods using the English syntax to detect negations
●
Document processing with MMTx
CPGs represent the currently best known knowledge within a certain medical topic. As state-of-theart knowledge they support practitioners in the decision for or against medical treatments, and they
serve as learning aids for students as well as information material for concerned patients. Due to this
intended purpose, CPGs are written in narrative speech. Thus, an automated processing is difficult,
but nevertheless desirable. Negation Detection is an important task for automated processing of
CPGs. It helps to accelerate the decision-making process and can support medical staff in the care
for patients.
The functionality of MMTx allows a document analysis starting with the splitting of documents into sections and
sentences in these sections. Subsequently, an assignment of part-of-speech tags allows the text analysis on
phrase level.
NegHunter
The Intra-Phrase Triggered Negation
All negations are assigned a certain negation class depending on their syntactical characteristics. We
differentiate five different classes:
In this type of negation the negation trigger is contained in the phrase. We use following rule:
The Adverbial Negation
An adverbial trigger negates a noun phrase via a verb. Passive and active voice are distinguished to
allocate the negated phrase. For example, the tenses simple present active and simple past active
are characterized by following combination:
<Adverbial Negation Active> ::= <Active Negation Trigger> <Negated Phrases>
<Negated Phrases> ::= [<Noun Phrase>] [<Noun Phrase>] [<Noun Phrase>]
Document
In active and passive voice we used the following rules to identify triggers:
<Active Negation Trigger> ::= (<Verb Phrase> <Adverb Trigger> <Verb Phrase>)
| (<Be Verb> <Adverb Trigger> <Verb Phrase -ing>)
| (<Verb Phrase> <Adverb Trigger> <Verb Phrase> <Verb Phrase>)
<Verb Phrase -ing> ::= <Verb Phrase> “ing”
<Passive Negation Trigger> ::= (<Be Verb> <Adverb Trigger> <Verb Phrase>)
| (<Verb Phrase> <Adverb Trigger> <Be Verb> <Verb Phrase>)
| (<Be Verb> <Adverb Trigger> <Be Verb> <Verb Phrase>)
| (<Verb Phrase> <Adverb Trigger> <Verb Phrase> <Be Verb> <Verb Phrase>)
“Evidence obtained from at least one well-designed study without randomisation.”
The Prepositional Negation
A prepositional phrase follows the negation trigger. This prepositional phrase is considered negated. The
detection pattern is:
<Prepositional Negation> ::= (“lack” | “absence” | “free” | “freedom” | “none”) <Prepositional Phrase>
For example:
“Patients with good performance status, … , and the absence of systemic disease.”
The Adjective Negation
Here, adjectives work as negation triggers. The rule for this negation is:
<Adjective Negation> ::= <Noun Phrase> [<Not Noun Phrases>] <Be Verb> <Adjective Phrase>
The following sentence shows an example:
“Recommendation indicates at least fair evidence that the service is ineffective.”
The Verb Negation
This type of negation distinguishes itself by the usage of verbs as negation triggers. According to the suffix of
the verb we determine, whether the sentence is written in active or passive voice. The rules are:
<Adverb Trigger> ::= “not” | “never”
1
Negation triggers are highlighted in yellow, negated phrases in green.
Phrase
For example:
The preceding three noun phrases are considered negated. Examples for these rules are:
“Guideline developers do not recommend chemotherapy.” (active voice)1
“Chemotherapy is not recommended.” (passive voice)
Sentence
<Intra-Phrase Triggered Negation> ::= (“no” | “without” | “ineffective” | “inappropriate”) <Noun Phrase>
Three noun phrases succeeding this combination are considered negated. The rule for the passive
voice of these both tenses is:
<Adverbial Negation Passive> ::= <Negated Phrases> <Passive Negation Trigger>
Section
<Verb Negation Active> ::= (“lack” | “lacked” | “deny” | “denied”) [<Not Noun Phrases>] <Noun Phrase>
<Verb Negation Passive> ::= <Noun Phrase> [<Not Noun Phrases>] (“lacking” | “denying”)
The following sentence contains an example in the passive voice:
“Information on final patient outcomes was also lacking.”
Evaluation
We evaluated 16 CPGs which totally contained 731 negated phrases in 642 sentences. The figure
below shows the relative occurrence of each negation class to the overall number of detected
negations:
Results
NegHunter's strength lies in the detection of the following negation types:
● Intra-phrase triggered negation
● Prepositional negation
● Adjective negation
We achieved lower results in the more complex classes:
● Adverbial negation
● Verb negation
The following table shows the exact results:
For the evaluation of our method we used recall and precision, which are statistical parameters
commonly used in Information Retrieval fields. As Natural Language Processing is a part of
Information Retrieval, these parameters also apply for our purposes. Recall and precision are
calculated in the following way:
Conclusion
Syntactical Negation Detection in CPGs leads to promising results. The usage of the English grammar enables
us to identify negated phrases even in CPGs, which are due to their natural language style very difficult to treat.
To achieve this, we use syntactical characteristics of the language like tenses and the active and passive voice.
The future outline goes into the additional employment of semantic features. With this, we are able to refine our
searching procedures. Consider the following:
We did not treat the infection.
We did not detect the infection.
Both sentences express a differing matter of fact. With the usage of a semantic approach adapted to search
requirements the identification of needed information can be accelerated and improved.
Kontakt: [email protected]