This document discusses using spreading activation, a connectionist method, to recommend related concepts from medical ontologies. It describes a Java tool called ONTOSPREAD that implements spreading activation in 3 stages over a graph. The tool can use constraints like context and can recommend related symptoms and diseases from ontologies like Open GALEN and SNOMED CT when given initial medical concepts. Evaluation shows it works like the human brain but still requires domain experts for configuration.

1. Application of the Spreading Activation
Technique for Recommending
Concepts of well-known Ontologies in
Medical Systems
José María Álvarez Rodríguez
WESO-Universidad de Oviedo
http://purl.org/weso/
First International Workshop on Semantic Applied Technologies on Biomedical
Informatics (SATBI 2011)
In conjunction with ACM International Conference on Bioinformatics and
Computational Biology (ACM-BCB)
Chicago, IL, U.S.A. August 1-3, 2011

2. Introduction
Psycho linguistics
and
Semantic priming

3. Retrieve data as
brain can do
...a connectionist
method

4. New information
realm...
Semantic Technologies
Linked Data...
Different scopes...
E-Health, E-procurement,
etc.

5. E-Health Sector
Need to automate
processes
combine and synthesize complex
related pieces of information
facilitate access to clinical information
identify patterns within the patient data
…

6. How to select
related concepts
(symptoms &
diseases)?

7. Tagging of EHR
CDSS

8. Graph Exploration
Document Retrieval
Information Retrieval
Annotation & Tagging
Recommending engines
Semantic Search Engines

9. Open GALEN
“Ontologies” are considered
categories: 23,141 concepts
950 relations…
to facilitate clinical apps &
statistics

10. SNOMED CT
765,000 active English-language
descriptions
830,000 logically-defining
relationships…
Data retrieve & analysis,
tagging…

11. Summary...
Concepts
Relations
Apps to retrieve, analyze,
annotate
…a huge amount of
medical data

12. GRAPH
EXPLORATION?

13. Spreading Activation
(SA)
3 Stages
Activation function
Calculation of activation value
Constrained SA
Stop condition

14. 3 Stages
Preadjustment
Initial nodes, weights of relations, set
functions…
Execution
Spread the activation value, graph
exploration…
Postadjustment
Rank nodes…

15. Variables & Constants
/
E
E
t

16. Activation function
Calculation of activation
value

17. Graphically Activation
function

18. Constrained SA
Distance
Nodes far from an activate node should be penalized
Path
Path of activation built by the algorithm
Fan out
Nodes highly connected should not be representative
Activation-threshold
Spread nodes have an activation value = threshold

19. Stop Condition
There is no node to spread
with an activation value
Min Activation Value
Threshold

20. ONTOSPREAD
http://code.google.com/p/ontospread/
API Java
Extensible (intensive use of design patterns)
Add new constraints to SA
Context, time, output degradation …
3 Tools
library, test module graphical
debugger

21. New constraints
1) Context of activation
2) Min activation value*
3) Max Min spread
concepts
4) Time of execution
5) Degradation Functions

22. Degradation Functions
1) Generic
2) Distance-Based
3) Beats-based (k number of iterations)

23. Converging Paths
Reward

24. Implementation
1) Set of activated nodes
2) Set of spread nodes
3) Activation value

26. ONTOSPREAD in
Action

27. ^ Z

28. Methodology
1. Select well-known ontologies
2. Define a set of initial concepts
3. Specify the weight of relations
4. Combine restrictions
5. Select degradation function
6. Add reward function
7. Test repeat!

29. Open GALEN
1. Ontology
2. #Advanced-BreastCancer
NAMEDSymptom
3. Default value 1.0
4. Constrained SA + New restrictions
5. H1 H2
6. No | Yes

31. SNOMED-CT
1. Ontology
2. #Articular cartilage of lunate
#Articular tissue sample
3. Default value 1.0
4. Constrained SA + New restrictions
5. H1 H2
6. No | Yes

33. Use Cases
BOPA Project
(semantic searh engine of legal documents)
http://www.w3.org/2001/sw/sweo/public/UseCases/CTIC/
10ders Project
(recommendation of public procurement notices)
http://purl.org/weso/moldeas/

34. Evaluation
1. Close to the human brain
behaviour
2. Configurable extensible
framework
3. Flexibility scalability
..but...
A domain-expert is still needed

35. Further Steps
1. Automatic configuration of
the algorithm
2. Development of Map-
Reduce version of SA (to
be published)

36. Application of the Spreading Activation
Technique for Recommending
Concepts of well-known Ontologies in
Medical Systems
José María Álvarez Rodríguez
WESO-Universidad de Oviedo
http://purl.org/weso/
First International Workshop on Semantic Applied Technologies on Biomedical
Informatics (SATBI 2011)
In conjunction with ACM International Conference on Bioinformatics and
Computational Biology (ACM-BCB)
Chicago, IL, U.S.A. August 1-3, 2011