Semantic Relationships and Medical Bibliographic Retrieval: A Preliminary Assessment

1
Semantic Relationships and Medical Bibliographic
RetrievalA Preliminary Assessment

• Perry L. Miller, Kenneth W.Barwick,Jon S.Morrow,
  Seth M.Powsner, and Caroline A.Riely
• Presented by Bindhu

2
Overview

• Whether semantic relationships between
  bibliographic terms effectively partition the
  clinical literature?
• Role in computer based bibliographic retrieval

3
Introduction

• Computer based medical bibliographic retrieval
  useful in patient care
• Different approaches
• Statistical frequency of query terms
  in
• paper as indicator of
  relevancy
• Semantics integrating meaning of
• concepts in query in
  retrieval.

4
Do semantic relationships between terms
partition the clinical literature?

• Abstracts from MEDLINE relating pair of clinical
  terms in the domain of liver diseases.
• Bibliographic terms from 4 categories
• diseases
• treatments
• tests
• patient characteristics
• Different types of relationship among categories

5
Diseases-disease relationships

• Disease X might cause or predispose to disease Y
• Disease X might affect the outcome or prognosis
  of disease Y
• Disease X might be mistaken for disease Y
• Disease X might exacerbate disease Y
• The relationship between two diseases may not
  be same all papers which discuss both diseases.

6

7
How can the papers be partitioned?

• Each paper in the bibliographic database must be
  coded to indicate the terms and semantic
  relationship between those terms
• Rule out papers where no relationship applied
• Different relationships
• - retrieve only the subset of papers
  that user interested in.

8
Coding of semantic relationships

• MEDLINE uses Medical Subject Heading(MeSH) -
  mostly nouns
• Use Verbs and prepositions to show relationships
• Binary relationships
• Disease X might cause or predispose to
  
• disease Y
• Treatment X treats disease Y
• N-ary relationships
• Tests A, B, and C are compared for
  diagnosing
• disease Y
• can be decomposed

9
Ex1Semantic relationships useless

• Hepatic Resection and Liver Neoplasms
• MEDLINE abstract from 40 papers
• All papers discussed same relationship
• Use of hepatic resection treats liver neoplasms

10
Ex 2Semantic relationships useful

• Diagnostic Ultrasound(US) and Liver Neoplasms(LN)
  36 papers
• 11- US for diagnosing LN
• 12-compared US with other diagnostics
• 6-US facilitate treatment of LN
• 1- US facilitate procedure - facilitate
  treatment
• of LN
• 2- US facilitate a test( fine needle biopsy)
  to
• diagnose LN
• 3- US detect secondary effects of LN
• 2- US assess the effects of treatment of LN
• 2-US to screen for LN

11
Is semantic relationship able to partition
Clinical literature ?

• 10 lists of abstracts
• In each list (20-40 abstracts) two MeSH terms
  representing 4 types of relationships
• - disease-disease
• - disease- treatment
• - disease-test
• - disease- patient characteristics
• Analyzed by liver disease expert and
  physician/computational linguist

12
(No Transcript)
13
Assessment

• On average 3.9 relationship between each pair of
  terms
• Semantic relationships between bibliographic
  terms effectively partition the clinical
  literature for computer based bibliographic
  retrieval
• Depend on nature of terms and of the domain
• Factors that interfere
• Few relationships
• Literature focuses on subset of
  relationships

14
Using semantic relationships in computer- Based
Bibliographic retrieval system

• Strategies
• Allow user to choose from all semantic
  relationships before retrieval
• Let the system first retrieve all papers relating
  to the chosen search terms and then allow the
  user to choose
• .

15
Conclusion

• Semantic relationships between bibliographic
  terms partition the clinical literature for
  computer based bibliographic retrieval
• But the effectiveness depend on nature of terms
  involved and of the domain

16
Questions?