PowerPointPrsentation

1
SummIt-BMT Summarize It in Bone Marrow
Transplantation
An interim report
Brigitte Endres-Niggemeyer Sigrid
Jauris-Heipke Margit Becher Elisabeth Wansorra
Clemens Jürgens
2
Core project goals
Knowledge-based summarization following competent
human practice
Fast user- and demand-oriented knowledge supply
from the web for physicians in bone marrow
transplantation (BMT)
3
Overview of the talk
Scientific background and system overview
Brigitte Endres-Niggemeyer A life science
approach to the SummIt ontology Sigrid
Jauris-Heipke Technical system, user
interface Margit Becher Summarizing,
agents Elisabeth Wansorra Medical
background, scenarios, system demo
Clemens Jürgens
4
SummIt-BMT overview
5
Scientific background
Ontology engineering Thesaurus construction
Bone marrow transplantation
Information seeking
Language and knowledge processing
Web technologies
Information retrieval
ltPubmedArticlegt ltMedlineCitation Owner"NLM"
Status"Completed"gt ltPMIDgt12803126lt/PMIDgt ltDateCre
atedgt ltYeargt2003lt/Yeargt
6
Empirical foundation of summarization agents 54
working processes of abstracters/indexers
Definition of thinking strategies, reconstruction
of working steps, realization of agents
Relevant-call TextToProposition
7
Empirical foundation of the ontology Methods
from thesaurus construction and corpus linguistics
Concepts and assertions are grounded on
occurrences in 15 BMT articles. The ontology is a
grounded theory of BMT.
8
Empirical foundation of user scenarios more than
150 queries of BMT physicians
12 original queries
Refractory virus infections Pat. with resistant
anal herpes and restricted kidney function. Can
Foscarnet ointment achieve efficent / measurable
systemic levels? Systemic fusarium
infection. Most efficient therapy?
and so on
9
Aims until 2005
Automatic summarizing according to human
principles more and smarter agents, practical
applicability
Fieldable BMT ontology with reliable and flexible
content
Computerized ontology maintenance workbench,
database solution and machine learning for
ontology update
Evaluation
Fast and safe web system
Keeping up with progess in communication
technologies
10
Long-term aims
More languages Spanish, Japanese
Multimedia text, grafics and image information
Portability at least one additional domain
A cognitive robot web companion
E-commerce distribution
.....
11
Summit-Ontology PCR (polymerase chain reaction)
12
Ontology

• Represents the most important concepts of the
  field bone marrow transplantation and their
  semantic relations
• Consists of
• concepts
• First-order statements

13
Ontology (concepts)

• Source corpus of 15 publications in the field of
  bmt (bone marrow transplantation)
• Currently there are about 4700 concepts
• concepts are connected via is-a-relations with
  their hypernyms

14
Polymerase chain reaction lower concepts
15
Concepts)
16
First order statements

• First order statements are derived by
  formalisation of sentences found in these 15
  publications
• A sentence is equivalent to a context (currently
  about 2500 contexts)
• The contexts are formed by predicates, currently
  about 4800 predicates
• Right now we are developing unifiers, about 1/3
  of the predicates are unified using 358 unifiers

17
Example
cDNA sequences are amplified using specific PCR
primers for each gene.
18
Context
19
Predicate
20
Predicate with unifier
21
Unifier
22
Predicate with unifier
23
User Interface, Technical System Realisation
24
A system-oriented Query Interface PubMed
25
Design Principles
User centered requirements
DIN EN ISO 9241-10 - Ergonomic requirements for
office work with visual terminals - suitability
for the task - self descriptiveness -
controllability - conformity with user
expectations - error tolerance
Design Gerrit Fichtner
26
A Scenario
24
27
A framework for scenarios
Analysis Which gui-controls will be needed ?
choices
simple textfields
textfields with add. specifications
multiple textfields
28
JavaScript class of a simple text field
Uml einfügen!
function tfEinfach(b, klassetxt,head) this.b
b // Bereich auf der
Seite this.klassetxt 'top.' klassetxt
this.head head // Ueberschrift this.fr
ageZ false // Methoden this.fzSetzen
fzSetzen // fz umsetzen this.write
e_writeTf // Schreiben this.auswert
e_auswert // Auswerten
29
Technical system platform

• Hardware Apple Xserve G4
• Software
• Apache Webserver
• Tomcat as Servlet-Engine
• for result-presentation Apache Cocoon
• XML-DB Apache Xindice
• FDG Parser (Linux)
• some helper programs, i.e. pdfToText

30
The Search
The user input will be send as XML-string to
Summit-Servlet ltscenario
name"Antiinfective therapy"gt ltblock
name"Patient"/gt ltblock
name"Infection"gt ltfield
type"input" question"false"gt
ltinputgtpneumonialt/inputgt
lt/fieldgt lt/blockgt
ltblock name"Infective agent"/gt
ltblock name"Drug"/gt ltblock
name"Contraindicated drugs"gt
ltfield type"input" question"true"gt
ltinputgtpenicillinlt/inputgt
lt/fieldgt lt/blockgt
lt/scenariogt
31
The Search
32
The Search - Result Presentation
33
Summarization agents
34
Summarization in the SummIt-BMT system
35
SummIt-BMT overview
The summarization agents of SummIt-BMT
36
SummIt-BMT overview
Context
37
Redundancy
Redundancy deletes doubles in the propositions
that have been deemed relevant.
38
The text interpretation group
The text interpretation group is the implemented
analogon of the empirical summarization strategy
relevant-call.
It currently consists of three agents
TextToProposition, EmbedPropositions, and
Question.
Together they implement the function of
relevant-call supported by reading strategies,
as far as needed in a web environment.
The implemented version of relevant-call performs
task-driven text interpretation and concomitant
relevance assessment.
39
TextToProposition
... has the practical advantage of weekly
administration
40
EmbedPropositions
text-based candidate proposition
corresponding ontology proposition
41
Question
42
Upcoming agent upgrades
43
Medical background System demonstration
44
Clinical case
Immunosuppression e Infection risk e Opportunist
ic agents e Antibiotic therapy e Allergy
e Resistence

• BMT - patient
• Pneumonia
• Haemophilus influenzae
• ??? Ciprofloxacin ???
• Penicillin allergy
• Trimethoprim resistence

45
Choose a scenario
46
Scenario families
47
Scenario - Antiinfective therapy
48
Check your input
49
Search started
50
Search string generation
therapy bone marrow transplantation
Infectioninfection type
pneumonia
haemophilus influenzae
Infective agent
Drugfind
ciprofloxacin
51
Search string generation
("haemophilus influenzae" OR "Pfeiffers
bacillus" OR "H FLU" OR "H. influenzae" OR
"hemophilus influenzae") AND (pneumonia OR
pneumonitis OR "lung inflammation" OR "pulmonary
inflammation") AND (ciprofloxacin OR cipro OR
ciproflovacin OR ciproloxacin) AND bone marrow
transplantation AND therapy
52
Search string generation
Attempt 2 therapy bone marrow
transplantation Infection Drugfind
Attempt 1 therapy bone marrow
transplantation Infection Infective
agent Drugfind
Attempt 3 therapy Infection
53
Downloading documents
54
Entering text passage retrieval
55
Scenario interpretation
suffer(patient, pneumonia) diseaseCharacteristic
(pneumonia, local) cause(haemophilus
influenzae, pneumonia) isTreated(,pneumonia,
ciprofloxacin) disregard(penicillin) disregard(t
rimethoprim)
suffer(patient, Infection) diseaseCharacteristi
c(Infection, Infectioninfection
type) cause(Infective agent, Infection) isTrea
ted(,Infection, Drugfind) disregard(Contra
indicated drugs)
56
(No Transcript)
57
(No Transcript)
58
(No Transcript)
59
(No Transcript)
60
(No Transcript)
61
Lots of work ahead!