THE RETRIEVAL PROCESS IN THE SAFRS SYSTEM WITH
THE CASE-BASED REASONING APPROACH
Souad Demigha
Centre de Recherche en Informatique (C.R.I), Université de Paris 1 – Panthéon-Sorbonne
90 rue de Tolbiac 75634 Paris Cedex 13, France
Keywords: Retrieval process, SAFRS, case-based reasoning, training, MAP, intention, strategy.
Abstract: The paper presents the retrieval process in the SAFRS system (system supporting the training of
radiologists-senologists) with the case-based reasoning approach (CBR, which is adopted to represent the
experience of expert radiologists-senologists under the form of cases) and modelized with the MAP concept.
The retrieval process relies on a procedure of case-based reasoning for retrieval of similar cases formalized
using a MAP, a re-use methodology named the retrieval MAP. The model of the MAP is an intentional
representation system. It is based on concepts of intention and strategy. The concept of intention (or a goal)
aims to capture the objective to be achieved. A strategy is the manner an intention is achieved. The retrieval
process with the MAP is a multi-step/multi-algorithm process, which permits to retrieve similar cases in
various modes and strategies. It is achieved according to three complex strategies: global strategy (or global
retrieval strategy), elementary strategy (or elementary retrieval strategy) and mixed strategy (or mixed
retrieval strategy).
1 INTRODUCTION
The SAFRS system (Système d’Aide à la Formation
des Radiologues-Sénologues) is a training system in
the domain of radiology-senology. It aims at
capitalizing and reusing the experience of
radiologists-senologists in order to enable junior
radiologists-senologists to have access to and learn
from the experience of experts. Experts’ experience
is represented as knowledge; both product
knowledge (mammographies and associated
diagnoses…) and process knowledge (heuristics) are
considered. While the product is the result to be
achieved, the process is the way the result is
achieved. The paper presents the retrieval process in
the SAFRS system with the case-based reasoning
approach (CBR) and which is modelized with the
MAP concept representing the process knowledge.
The case-based reasoning is adopted to represent the
experience of expert radiologists-senologists as
cases (Aamodt and al, 1994). This allows to obtain
an oriented-object model with the UML formalism
(Unified Modelling Language) structured as cases.
The retrieval process is divided into three hierarchic
levels: the first level (the case) is a patient at
different intervals of treatment (time). A case may
comprise several successive senologic episodes. The
second level (the sub-case) is one senologic episode
(clinical examination, image reading, radiological
interpretation, and anatomo-pathological
examination) for a given patient. The third level (the
sub-sub-case) represents one phase of a senologic
episode for a given patient (clinical examination OR
image reading…), (Demigha and Prat, 2004).
The retrieval process using the MAP is a multi-
step/multi-algorithm process, which permits to
retrieve similar cases in various modes and
strategies. It is achieved according to three complex
strategies: global strategy (or global retrieval
strategy), elementary strategy (or elementary
retrieval strategy) and mixed strategy (or mixed
retrieval strategy). The global strategy allows for
retrieval at the global level, i.e. the case. The
retrieval process starts at the sub-sub-case: it
represents one phase of a senologic episode for a
given patient (clinical examination OR image
reading OR radiological interpretation OR anatomo-
pathological examination), then we go to the
intermediate level (the sub-case: it is one senologic
episode for a given patient), until cases of interest in
468
Demigha S. (2007).
THE RETRIEVAL PROCESS IN THE SAFRS SYSTEM WITH THE CASE-BASED REASONING APPROACH.
In Proceedings of the Ninth International Conference on Enterprise Information Systems - AIDSS, pages 468-473
DOI: 10.5220/0002354904680473
Copyright
c
 SciTePress

the treatment of the new case (target case) are found;
finally we aggregate at the case level. Elementary
strategy allows combining one to three phases of the
radiological process. The mixed strategy aims to go
back and start from the elementary level (the sub-
case) until finding cases of interest in the treatment
of the new case (target case).
The paper structure is as follows:
- Section 2 provides in details the retrieval process
with the concept of the MAP.
- Section 3 provides the evaluation of the retrieval
process with the concept of the MAP.
- Section 4 is the conclusion with further research
works in progress.
2 THE RETRIEVAL PROCESS IN
THE SAFRS SYSTEM
In the SAFRS system, the retrieval process is
modelled by a MAP called the retrieval MAP (see
Figure 1), (Demigha, 2005).
2.1 The Model of the MAP
The model of the MAP is an intentional
representation system. It is based on concepts of
intention and strategy. It includes one or several
sections. A section is based on two concepts:
intention (or goal) and strategy. The concept of
intention aims to capture the objective to be
achieved at one time of a process. A strategy is the
manner to achieve an intention. A section is an
aggregation of two types of intentions: a source
intention, a target intention and a strategy as well.
A MAP is represented by a graph oriented and
labelled. Intentions represent nodes and strategies
represent the arcs. A section is then represented by
two nodes linked by an arrow. A section must be
selected when it is initialized. The selection of
sections is based on directives. A directive includes
a signature and a body. A signature represents the
visible part of the directive. A signature is defined
by the couple <(intention), situation>. A body
defines the followed step in order to satisfy the
intention captured in the signature. A directive
includes two types of directives: strategic directive
and tactical ones: (1) the strategic directive
represents a strategic view of the multi-step
development based on a set of intentions and
strategies. It is represented by a MAP and a set of
associated directives and (2) the tactical directive
has a three-structure. It is composed of three other
directives (context: a context represents the
development of a process by a hierarchy of
contexts): plan, selection (the selection of several
alternative sub-directives) and executable. A plan
directive corresponds to a complex problem
decomposed into a set of sub-problems. The
execution of the composed directives is defined by a
graph. The nodes of the graph are directives
(components of the plan). Arcs (previous links)
represent arranged or parallel transitions between
directives. A selection directive corresponds to a
situation that necessitates the exploration of different
possibilities. An executable directive corresponds to
an intention which can be characterized by an action
of the product transformation or an action of
selection of an other directive.
2.2 Similarities
In the senology case representation model, cases are
collections of objects, each of which it is described
by a set of attribute-value pairs. The structure of an
object is described by an object class that defines the
set of attributes together with a type (set of possible
values or sub-objects) for each attribute. Object
classes are arranged in a class hierarchy, that is, a
tree in which sub-classes inherit attributes as well as
their definition from the current class.
We define a hierarchy of attributes types. New
types are defined by building sub-types of the
existing elementary types shown in Table 1. They
differ in their usability: A type may be used in an
immediate or derived type. While immediate types
cover the whole range of possible values of a type,
derived types are restricted in their range by defining
an enumeration of elements of their elementary
types or, in case of numeric types, by specifying an
interval (Bergmann and Althoff, 1998).
Table 1: Elementary types in the SAFRS system.
Type Usability
Integer Immediate and derived
Float Immediate and derived
Date Immediate and derived
Boolean Immediate only
String Immediate and derived
Enumeration Immediate and derived
Ordered Enumeration Derived only
Text Derived only
The approach we have chosen to determine
similarities is to establish a comparison between
attributes (attribute by attribute), then to each
attribute corresponds a comparison measure, it is a
local similarity measure, it determines a similarity
between two attribute values, and for each object we
determine a global similarity measure which
THE RETRIEVAL PROCESS IN THE SAFRS SYSTEM WITH THE CASE-BASED REASONING APPROACH
469

determines the similarity between two objects (or
between the case and the query) based on the local
similarity of the belonging attributes. The local
similarity measure allows to compare any two types
values. It returns a numeric value from the interval
[0..1]. This value is further used in the computation
of a global similarity.
2.3 The Retrieval Process
The retrieval process is, with the MAP, a multi-
step/multi-algorithm process, which permits to
retrieve similar cases in various modes. The retrieval
MAP of the SAFRS system represented on the graph
(Figure 1) defines, besides the two intentions ‘to
start’ and ‘to stop’, two major intentions for the
retrieval process achievement ‘to elaborate the new
case’ and ‘to retrieve similar cases’.
Figure 1: The retrieval MAP.
The intention ‘to elaborate the new case’ is
achieved according to two strategies: ‘by
preparation’ and ‘by creation’, contained in sections
C1 and C2, respectively. It consists in describing a
new case (new case to be diagnosed). From the new
problem, the case is elaborated from the general
knowledge, while keeping only relevant information.
If the case description is complete, then the intention
‘to elaborate the new case’ has the same meaning as
the creation (strategy ‘by creation’) or the
preparation (strategy ‘by preparation’) of the case,
else the creation phase is simplified.
The intention ‘to retrieve similar cases’ is
achieved according to three complex strategies:
‘global strategy’ (or global retrieval strategy),
‘elementary strategy’(or elementary retrieval
strategy) and ‘mixed strategy’ (or mixed retrieval
strategy) that are contained in sections C3, C4 and
C5, respectively. The ‘global strategy’ included in
section C3 allows for retrieval at the global level,
i.e. the case. The retrieval process starts at the sub-
sub-case level, then we go to the intermediate level,
the sub-case, and finally it ends to aggregate at the
case level. ‘Elementary strategy’ included in section
C4 allows to combine one to three phases, i.e. the
sub-sub-case (‘image-reading’, ‘radiological
interpretation’ and ‘anatomo-pathological
examination’) of the senological process. The ‘mixed
strategy’ included in section C5 allows to combine
the first two strategies (global and elementary ones).
It aims to go back and start from the elementary
level (sub-case), until it finds cases of interest in the
treatment of the new case (target case). The
‘abandonment strategy’ included in section C6
allows the case expert to abandon his/her retrieval
process for the new case, before starting the retrieval
when he/she makes mistakes in his/her reasoning,
thus allowing him/her to start again the retrieval
process, without starting from the very beginning,
i.e. from the source intention ‘to start’ of the MAP.
Once the case expert has carried out the retrieval
process, i.e. he/she succeeded or failed in searching
an interesting case for solving the new problem,
he/she has got four possibilities to treat this new
case: the ‘reuse strategy’ included in section C7
allows to revise the validity of retrieved solution,
which is retained for the goal problem (new problem
to solve). The ‘revise strategy’ included in section
C8 allows to revise the case according to three steps:
to revise it ‘by test’, ‘by correction’ and finally ‘by
validation’. The ‘retained strategy’ included in
section C9 allows to integrate to the case base the
new solved problem, if the latter confers novel
abilities to the system. The strategy ‘by retrieval
failure’
included in section C10 allows to send back
a negative result from the case base to the case
expert when no case could be identified as similar
enough to the target case (new case). Finally, the last
strategy ‘by abandonment strategy’ included in
section C11 allows the case expert to abandon the
retrieval of similar cases if he/she deems it
necessary, even after the overall process is achieved.
We detailed in the three following sub-sections
the three main strategies for the retrieval process.
2.3.1 The Global Strategy
The global retrieval strategy consists in retrieving
the case in its totality. The Figure 2 models this
strategy. Indeed, it is a plan directive: <(new case),
to research the similar cases by global strategy>
composed of a hierarchy of plans which contains
three contexts: plan, selection and executable. The
plan directive DRI
3
proposes three sub directives:
- DRI
3.1
: <(new case), to calculate similarities at
the sub-sub-case level>* ; (* means an iterative
form); -DRI
3.2
: <(sub-sub-cases selected), to
to start
C
3
.
B
y
g
l
o
b
a
l
s
t
r
a
t
e
g
y
C
4
.
B
y
e
l
e
m
e
n
t
a
r
y
s
t
r
a
t
e
g
y
C
5
.
B
y
m
i
x
t
s
t
r
at
eg
y
to stop
to retrieve
similar cases
C7. By reuse
strategy
C10. By
retrieval
failure
C9. By retain
strategy
C8. By
revise
strategy
to elaborate
the new case
C1. By creation
C2. By
preparation
C
6
.
B
y
a
b
an
d
o
n
m
e
n
t
C11. By
abandonment
ICEIS 2007 - International Conference on Enterprise Information Systems
470

calculate similarities at the sub-case level>* ; -
DRI
3.3:
<(sub-cases selected), to calculate
similarities at the case level>*.
Figure 2: The global strategy (a plan directive: hierarchy
of plan contexts).
As shown on Figure 2, the plan directive DRI
3.1
:
<(new case), to calculate similarities between sub-
sub cases>* proposes two plan contexts for the
realization of its intention: - DRI
3.1.1
: <(new case), to
retrieve similar cases by subsumption>* ; - DRI
3.1.2
:
<(new case), to retrieve similar cases by
similarity>*.
The subsumption is a mechanism of discrimination.
The directive DRI
3.1.1
is performed by the execution
of two plan contexts:
- DRI
3.1.1
: <(new case), to retrieve similar cases by
subsumption>*. The intention ‘to research by
subsumption’ is performed via two executable
contexts: DRI
3.1.1.1
: <(index new case), to match the
new case index with the abstract case>* and
DRI
3.1.1.2
: <(set of indices), to evaluate the
subsumption>*. To evaluate the subsumption
consists of browsing a net of indices where, at each
node, cases are selected by taking into account the
subsumption criterion.
For facilitating the retrieval process, the case is
abstract in order to extract indices. The abstraction is
aimed to divide the problem descriptors of the input
into two classes: the relevant descriptors (useful) and
the non-relevant descriptors (not useful) or noises.
The abstraction consists in eliminating noises.
- DRI
3.1.2
: <(new case), to select a sub-set of
relevant cases>*.
The intention ‘to select a sub-set of relevant cases’
eliminates the very distant cases and selects a set of
cases that are suitable for the target problem. It
implies that cases are organized in a classification
hierarchy according to relevant characteristics. The
selection of these characteristics determines the
capability to retrieve the ‘best’ cases.
After restricting the research space, the case
author performs a more specific comparison
between the target problem and each source case
previously selected by discrimination ‘by
subsumption’ with the plan directive ‘by similarity’:
DRI
3.1.2
: <(New case), to retrieve similar cases by
similarity>* .
The directive DRI
3.1.2
: <(new case), to retrieve
similar cases by similarity>* is performed by two
plan contexts: DRI
3.1.2.1
: <(set of index), to research
by similarity>* and DRI
3.12.2
: <(set of similar cases),
to select the most similar case>*.
- ‘To research by similarity’ (to research similar
cases) performs a comparison more specific between
the target problem and the source case previously
selected by discrimination. This comparison
necessitates a two by two comparison of cases,
attribute by attribute. This directive proposes two
plan directives for the realization of its intention
DRI
3.1.2.1.1
: <(selected cases), to match selected
cases and the new case>* and DRI
3.1.2.1.2
:
<(matched cases), to evaluate the similarity>*.
- The intention ‘to match selected cases and the
new case’: the matching process compares two by
two characteristics of cases. In most systems, the
matching is performed on characteristics of cases: it
is a global matching (global similarity by attribute
weighting at a local similarity level).
- The intention ‘to evaluate the similarity’: a
similarity measure is used in order to arrange source
cases by decreasing the similarity with the target
case. The evaluation is performed by considering
common characteristics; each one has a significant
importance level (weight) of the role that each
element of a problem plays in the reuse of elements
of the solution. The similarity evaluation is assumed
to depict the facility of the reuse of a source case.
- ‘To select the most similar case’: the solution
of cases having the best ‘score’ is selected for the
target problem. The directive plan DRI
3.1.2.1.1
:
<(selected cases), to match selected cases and the
new case>* proposes two selection alternatives to
complete the retrieval process: DRI
3.1.2.1.1.1
<(selected cases), to calculate similarities between
attributes>* and DRI
3.1.2.1.1.2
<(selected cases), to
calculate similarities between objects>*. These
directives allow the computation of similarity
measures between attribute-values (a local similarity
measure) and objects (global similarity measure)
(sub-sub-case, sub-case and case).
to retrieve
similar cases
to elaborate
the new case
C3. By global
strategy
DRI
3
<(New case), to retrieve similar cases by global strategy>
DRI
3.1
<(New case), to calculate
similarities between sub-sub-cases>*
DRI
3.2
<(Selected sub-sub-cases ), to
calculate similarities between sub-cases>*
DRI
3.3
<(Selected sub-cases), to
calculate similarities between cases>*
DRI
3.1.1
<(New case), to retrieve
similar cases By subsumption>*
DRI
3.1.2
<(New case), to retrieve
similar cases By similarity>*
DRI
3.1.1
<(New case), to
research By
subsumption>*
DRI
3.1.2
<(New case), Select a
sub-set of useful cases>*
DRI
3.1.2.1
<(Set of index), to
research By similarity>*
DRI
3.12.2
<(Set of similar cases), to
select more similar case>*
DRI
3.1.1.1
<(Index New
case), to match the new
case index with the
abstract case>*
DRI
3.1.1.2
<(Set of index), to
evaluate the subsumption>*
DRI
3.1.2.1.1
<(Selected cases ), to
match selected cases and the new
case>*
DRI
3.1.2.1.2
<(Matched cases), to evaluate
the similarity>*
DRI
3.1.2.1.1.1
<(Selected cases), to
c
alculate similarities between
value-attributes>*
DRI
3.1.2.1.1.2
<(Selected cases), to
calculate similarities between
objets>*
DRI
3.2.1
<(Selected sub-sub-cases),
to calculate similarities between
objets>*
DRI
3.3.1
<(Selected sub-cases), to
calculate similarities between
objets>*
THE RETRIEVAL PROCESS IN THE SAFRS SYSTEM WITH THE CASE-BASED REASONING APPROACH
471

2.3.2 The Elementary Strategy
The objective of elementary strategy (or elementary
retrieval strategy) is to offer the case author various
possibilities to resolve his/her problem. In the
absence of complete information on the new case,
the case author only considers into the case base the
knowledge that resembles new knowledge. The case
author can start the process with some knowledge of
one phase of different phases of a patient x; for
instance, this knowledge is compared with the
knowledge of the new case. The case author selects
knowledge of another phase concerning another
patient y. He/she combines these knowledge and
reiterates the process whenever required to make a
diagnosis. All these fragments, coming from
different phases of different patients or even from a
same patient, combined together (in the case that the
patient had previous reports), make up one solution
of the new problem to solve. The assessment of the
similarity (attributes and objects) is performed in the
same manner as the global strategy and the mixed
strategy.
The intention ‘to calculate similarities between
attribute-values’ of the directive DRI
3.1.2.1.1.1
allows
to use the hierarchy of UML types (Demigha and
Prat, 2004). Indeed, according to various types of
attributes, a similarity measure is selected.
The two other main sub-directives of the DRI
3
:
DRI
3.2:
<(selected sub-sub-cases), to calculate
similarities between sub-cases and DRI
3.3
:
<(selected sub-cases), to calculate similarities
between cases>* are executable contexts and thus
are not factorized.
- The second sub-directive of the directive
DRI
3.2
: <(selected sub-sub-cases), to calculate
similarities between sub-cases>* is a plan directive
including one context plan: DRI
3.2.1:
<(selected sub-
sub-cases), to calculate similarities between
objects>*.
- The third sub-directive of the directive DRI
3.3
:
<(selected sub-cases), to calculate similarities
between cases >* is a plan directive including one
context plan: DRI
3.3.1
: <(selected sub-cases), to
calculate similarities between objects>*.
Figure 3 models this strategy. Indeed, it is a plan
directive: <(new case), to research similar cases by
elementary strategy> composed of a hierarchy of
plans which include three contexts: plan, selection
and executable.
The plan directive DRI
4
proposes three principal
sub-directives: - DRI
4.1
: <(new sub-sub-case image
reading phase), to calculate similarities at the image
reading phase>* ;- DRI
4..2
: <(solution part of the
image reading phase), to calculate similarities at the
radiological interpretation level (RI)>* ; - DRI
4..3
:
<(solution part of the RI phase), to calculate
similarities at the anatomo-pathological
examination phase (AE)>*.
2.3.3 The Mixed Strategy
The mixed strategy allows to combine the first two
strategies, the global strategy and the elementary
strategy. For a radiologist (in the case that he/she
combines several knowledge from various sources),
the interest of this strategy lies in picking up
knowledge at the intermediate level, to find again
archives of previous examinations, and thus to
obtain a full knowledge.
Figure 4 models this strategy. Indeed, it is a plan
directive: <(new case), to research similar cases by
mixed strategy> composed with a hierarchy of plans
including three contexts: plan, selection and
executable.
The directive plan DRI
5
proposes five main sub-
directives: - DRI
5.1
: <(new sub-sub-case image
C4. By elementary
strategy
to retrieve
Similar cases
to elaborate
the new case
DRI
4
<(New case), to retrieve similar cases By elementary strategy>
DRI
4.1
<(New sub-sub-case Reading), to
calculate similarities in the image reading
phase>*
DRI
4.2
<(solution part of image reading phase),
to calculate similarities in the radiological
interpretation phase (RI)>*
DRI
4.3
<(solution part of RI phase), to calculate
similarities in the Histological Examination
phase (HA)>*
DRI
4.1.1
<(New sub-sub-case
Image Reading), to retrieve
similar case By subsumption>*
DRI
4.1. 1.1
<(New sub-sub-
case Image Reading, to
research By subsumption>*
DRI
4.1.2
<(New sub-sub-case
Image Reading), to retrieve
similar case By similarity>*
DRI
4.1.1.2
<(New sub-sub-case
Image Reading), to select a sub-set
of useful sub-sub-cases >*
DRI
4.1.1.1.1
<(Index New
sub-sub-case), to match
the new su b-sub-case
index with the abstract
case>*
DRI
4.1.1.1.2
<(Se t of inde x), to
evaluate the subsumption>*
DRI
4.1.2.1
<(Set of index), to
research By similarity>*
DRI
4.1.2.2
<(Set of similar sub-
sub-cases), to select the more
simil ar sub -sub -case>*
DRI
4.1.2.1.1
(Selected cases ), to
match selected cases and the new
sub-sub-case>*
DRI
4.1. 2.1.2
<(Matched sub-sub-cases), to
evaluate the similarity>*
DRI
4.1. 2.1.1.1
<Selected cases), to
calculate similarities between
value-attributes>*
DRI
4.1. 2.1.1.2
<(Selected cases), to
calculate similarities between
objets>*
Selec t the numbe r of phases
Image
Reading
only
RI
only
HE
only
Image Reading
and RI
RI and HE Image Reading, RI and HE
-------
- ------
-
-
-
-
Figure 3: The elementary strategy (a plan directive:
hierarchy of plan contexts).
to elbaorate
the new case
to retrieve
Similar cases
C5. By mixed strategy
DRI5<(New case), to retrieve similar cases By mixed strategy>
DRI
5.1
DRI
5.2
DRI
4.5
DRI
5.1.1
DRI
5.1.2
----------
----------
DRI
5.1.1.1
DRI
5.1.1.2
DRI
5.1.2.1
DRI
5.1.2.2
DRI5
5.1.1.1.1
DRI
5.1.1.1.2
DRI
5.1.2.1.1
DRI
5.1.2.1.2
DRI
5.1.2.1.1.1
DRI
5.1.2.1.1.2
Select thenumberofphases
Image
Reading
only
RI
only
AE
only
Image reading
and RI
RI and
AE
Image reading, RI and AE
---------
Figure 4: The mixed strategy (a plan directive: hierarchy
of plan contexts).
ICEIS 2007 - International Conference on Enterprise Information Systems
472

reading phase), to calculate similarities at the image
reading phase>*; - DRI
5.2
: <(solution part of the
image reading phase), to calculate similarities at the
radiological interpretation phase (RI)>*; - DRI
5.3
:
<(part solution of the RI phase), to calculate
similarities at the anatomo-pathological
examination phase (AE); - DRI
5.4
: <(selected sub-
sub-cases), to calculate similarities at the sub-case
level>*; - DRI
5.5
: <(selected sub-cases), to calculate
similarities at the case level)>.
As shown on Figure 4, the directive DRI
5.1
is a
hierarchy of directives of plan contexts, selection
and executable. This hierarchy has the same course
as the directive DRI
3.1
of the global strategy. We do
not provide details of the steps of calculation of
similarities. Extensive details for the retrieval
process are presented in (Demigha, 2005).
3 DISCUSSION
This paper deals with our retrieval process of the
case-based reasoning training system which we built
based on the MAP model. The latter has several
advantages: 1) as a process meta-model in the
radiologists-senologists modelling approach to their
interpretation, it enables, thanks to the directives, a
fast and simple access to knowledge. Actually, the
MAP offers a hierarchical and structuring approach
using selection and mixed strategies, 2) thanks to
these strategies, the radiologists-senologists can at
the same time have a free and diversified access in
order to browse dynamically the MAP. Selecting a
strategy is made as the realization of the intentions is
carried out. This means that the selection is a
dynamic process and the construction of the paths is
achieved according to the situations that are met
with, 3) most of the time; the radiologist-senologist
does not have enough knowledge in data-processing
and solves randomly his/her daily problems. He/she
has the advantage of adapting to the intentional
reasoning of the MAP. At no point is the radiologist-
senologist (especially the junior radiologist) forced
to carry out a particular intention or to apply a
strategy of realization of particular intention, unless
otherwise required by the situation, 4) the intentional
approach is structuring. Thanks to the intentions, it
makes it possible to synthesize and to abstract the
details in order to concentrate on the most important
at even the priorities, 5) there are various ways of
carrying out the intentions thanks to the strategies;
as a result, the problems can be solved in a flexible
and versatile way and 6) the strategies enabled us to
provide the radiologists-senologists with the
heuristics enabling them to use the best cases. These
heuristics are strongly related to the complex
structure and to the dependence on the cases,
therefore allowing for a lowering of the cost of
adaptation.
4 CONCLUSIONS
In this paper, we have presented the retrieval process
in the SAFRS system with the case-based reasoning
approach using the MAP model. The retrieval
process in the SAFRS system we developed has an
original aspect: this approach relies on the formal
description of the process in an intentional manner.
It is a complex process. It describes in a quite
accurate and detailed fashion the way we retrieve
similar cases according to various modes. The
exploitation of the layered case structure allows for
the search of a similar case by composition of sub
and sub-sub similar cases and by a set of a powerful
similarity measures. Implementation and validation
of the retrieval process will be developed in an other
paper.
REFERENCES
Aamodt, A., Plaza, E., Case-Based Reasoning:
Foundational Issues, Methodological Variations, and
System Approaches, 1994. In AI Communications,
94’ volume 7, number 1.
Bergmann, R, Althoff. K.D, Methodology for Building
CBR Applications, 1998. In M.Lenz, B.Bartsh-Sport,
H.-D. Burkhard, & S. Wess (Eds). Case-Based
Reasoning from Foundations to Applications, Spring-
Verlag, 98’.
Demigha, S., Prat, N., A case-based training system in
radiology-senology. 2004. In Information and
Communication Technologies: from Theory to
Applications (ICTTA) 04’, Damascus, Syria.
Demigha, S., Un système à base de connaissances en
radiologie-sénologie dédié à la formation des
radiologues-sénologues juniors, 2005. Phd thesis in
Informatic, Université de Paris 1 Panthéon-Sorbonne.
THE RETRIEVAL PROCESS IN THE SAFRS SYSTEM WITH THE CASE-BASED REASONING APPROACH
473