Difference between revisions of "Denecke and Bernauer ARTIFICIAL INTELLIGENCE IN MEDICINE 2007"

From Cohen Courses
Jump to navigationJump to search
(Created page with '== Citation == Ciaohua Zhou et al. 2006. Approaches to Text Mining for Clinical Medical Records. In Proceedings of the 2006 ACM symposium on Applied computing, 235-239. == Onli…')
 
 
(8 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
== Citation ==
 
== Citation ==
  
Ciaohua Zhou et al. 2006. Approaches to Text Mining for Clinical Medical Records. In Proceedings of the 2006 ACM symposium on Applied computing, 235-239.
+
Kerstin Denecke and Jochen Bernauer. 2007. Extracting Specific Medical Data Using Semantic Structures. Artificial Intelligence in Medicine, LNCS 2007 Volume 4594/2007, 257-264.
  
 
== Online version ==
 
== Online version ==
  
[http://delivery.acm.org/10.1145/1150000/1141330/p235-zhou.pdf?key1=1141330&key2=0111885821&coll=GUIDE&dl=&CFID=106800160&CFTOKEN=10813007 ACM portal]
+
[http://www.springerlink.com/content/n858219p72rm3105/fulltext.pdf SpringerLink]
  
 
== Summary ==
 
== Summary ==
  
The [[Category::paper]] presents a MEDical Information Extraction (MedIE) system, which extracts patient information from free-text clinical records.  
+
The [[Category::paper]] presents a medical [[AddressesProblem::Information Extraction]] system which is based on automatic generation of semantic structures for free-text. The dataset used is a gold standard comprising 50 discharge summaries from a surgical department in a hospital. The purpose of the system is to fill in three templates defined below.
  
They divided their extraction job into three tasks below.
+
[[File:templates.png]]
* Extraction of medical terms
 
* [[AddressesProblem::Relation Extraction]] 
 
**In this paper, relation extraction means extraction of associated medical concepts. For example, 'Blood pressure' and '144/90' are associated terms in the sentence, "Blood pressure is 144/90".
 
* [[AddressesProblem::Text Classification]]
 
**For example, a patient can be classified as a former smoker, a current smoker, or a non-smoker
 
  
Their approaches are:
+
Current(2007) approaches to medical NLP are often limited to a certain medical domain and their construction is fairly complicated. In addition, they are limited to a certain language, especially in English. So, the system in this paper is designed to overcome these limitations.
* An ontology-based approach for extracting medical terms of interest
 
**They used Unified Medical Language System (UMLS). About terms that are not defined in UMLS, they predicted categories of some terms using sentence structures.
 
* A graph-based approach which uses the parsing result of link-grammar parser for [[AddressesProblem::Relation Extraction]]
 
**Notable things in their approach are three. First, they included the processing of negation. Second, when the parser fails, they used a pattern-based approach. Lastly, they replaced multi-word terms with placeholders because the parser did not process the terms.
 
* an NLP-based feature extraction method coupled with an ID3-based [[AddressesProblem::Decision Tree Learning]] for [[AddressesProblem::Text Classification]]
 
  
 +
The key method of this system the authors suggest is [[UsesMethod::building semantic structures]] from texts.
 +
A conceptual graph-like representation is produced for each sentence of a text, and then these semantic structures are exploited to extract information.
  
This approach was fairly successful mostly showing over 80% of precision and recall. However, the system was tested on the data written by only a clinician, which means that the style of free-text records was consistent. Nevertheless, the research is worth in that they applied various IE techniques to the free-text clinical records, explain about the problems they encountered.
+
The system is composed of three major modules: a text-preprocessor, a semantic interpreter, and an extractor.
 +
In the text-preprocessor, sections are detected and classifies, and texts are parsed. This results in tagged texts with dates, quantities and dosage specifications, and expressions with special meaning like "exclusion of" and "evidence of". The semantic interpreter produces a semantic representation consisting of concepts and relations from each sentence. In the extractor, according to manually defined rules, information is extracted from the text itself, and its semantic representation in order to fill in templates.
  
== Related papers ==
+
[[File:semanticRep.png]]
  
An interesting follow-up paper is [[RelatedPaper::Denecke and Bernauer AIME 2007]] which uses semantic structures to extract medical information.
+
 
 +
The advantage of this system is that the components that might be adopted for processing texts in another language because it represents texts as semantic structures.
 +
The system achieved 81-95% precision and 83-97% recall of performance on template filling tasks. The main reasons of error were unknown words, wrong paragraph detection, missing trigger words, and processing needs additional knowledge.

Latest revision as of 16:06, 21 October 2010

Citation

Kerstin Denecke and Jochen Bernauer. 2007. Extracting Specific Medical Data Using Semantic Structures. Artificial Intelligence in Medicine, LNCS 2007 Volume 4594/2007, 257-264.

Online version

SpringerLink

Summary

The paper presents a medical Information Extraction system which is based on automatic generation of semantic structures for free-text. The dataset used is a gold standard comprising 50 discharge summaries from a surgical department in a hospital. The purpose of the system is to fill in three templates defined below.

Templates.png

Current(2007) approaches to medical NLP are often limited to a certain medical domain and their construction is fairly complicated. In addition, they are limited to a certain language, especially in English. So, the system in this paper is designed to overcome these limitations.

The key method of this system the authors suggest is building semantic structures from texts. A conceptual graph-like representation is produced for each sentence of a text, and then these semantic structures are exploited to extract information.

The system is composed of three major modules: a text-preprocessor, a semantic interpreter, and an extractor. In the text-preprocessor, sections are detected and classifies, and texts are parsed. This results in tagged texts with dates, quantities and dosage specifications, and expressions with special meaning like "exclusion of" and "evidence of". The semantic interpreter produces a semantic representation consisting of concepts and relations from each sentence. In the extractor, according to manually defined rules, information is extracted from the text itself, and its semantic representation in order to fill in templates.

SemanticRep.png


The advantage of this system is that the components that might be adopted for processing texts in another language because it represents texts as semantic structures. The system achieved 81-95% precision and 83-97% recall of performance on template filling tasks. The main reasons of error were unknown words, wrong paragraph detection, missing trigger words, and processing needs additional knowledge.