UBICC Journal

Ubiquitous Computing and Communication Journal

Volume 4 · Number 3 · July 2009 · ISSN 1992-8424

Special Issue on ICIT 2009 Conference – Bioinformatics and Image

UBICC Publishers © 2009

Ubiquitous Computing and Communication Journal
Co-Editor Dr. AL-Dahoud Ali

Ubiquitous Computing and

Communication Journal

Book: 2009 Volume 4

Publishing Date: 07-30-2009
Proceedings
ISSN 1992-8424

This work is subjected to copyright. All rights are reserved whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, re-use of
illusions, recitation, broadcasting, reproduction on microfilms or in any other way, and
storage in data banks. Duplication of this publication of parts thereof is permitted only
under the provision of the copyright law 1965, in its current version, and permission of
use must always be obtained from UBICC Publishers. Violations are liable to prosecution
under the copy right law.

UBICC Journal is a part of UBICC Publishers

www.ubicc.org

© UBICC Journal
Printed in South Korea

Typesetting: Camera-ready by author, data conversation by UBICC Publishing Services,

South Korea

UBICC Publishers
 Guest Editor’s Biography
Dr. Al-Dahoud, is a associated professor at Al-Zaytoonah University, Amman, Jordan.
He took his PhD from La Sabianza1/Italy and Kiev Polytechnic/Ukraine, on 1996.
He worked at Al-Zaytoonah University since 1996 until now. He worked as visiting
professor in many universities in Jordan and Middle East, as supervisor of master and
PhD degrees in computer science. He established the ICIT since 2003 and he is the
program chair of ICIT until now. He was the Vice President of the IT committee in the
ministry of youth/Jordan, 2005, 2006. Al-Dahoud was the General Chair of (ICITST-
2008), June 23–28, 2008, Dublin, Ireland (www.icitst.org).
He has directed and led many projects sponsored by NUFFIC/Netherlands:
- The Tailor-made Training 2007 and On-Line Learning & Learning in an Integrated
Virtual Environment" 2008.
His hobby is conference organization, so he participates in the following conferences as
general chair, program chair, session’s organizer or in the publicity committee:
- ICITs, ICITST, ICITNS, DepCos, ICTA, ACITs, IMCL, WSEAS, and AICCSA
Journals Activities: Al-Dahoud worked as Editor in Chief or guest editor or in the Editorial
board of the following Journals:
Journal of Digital Information Management, IAJIT, Journal of Computer Science, Int. J.
Internet Technology and Secured Transactions, and UBICC.
He published many books and journal papers, and participated as speaker in many
conferences worldwide.
 UBICC Journal
Volume 4, Number 3, July 2009

SPECIAL ISSUE ON ICIT 2009 CONFERENCE: BIOINFORMATICS AND IMAGE

618 Testing of program correctness in formal theory

Ivana Berkovic, Branko Markoski, Jovan Setrajcic, Vladimir Brtka, Dalibor
Dobrilovic

628 Detecting metamorphic viruses by using arbitrary length of control flow

graphs and nodes alignment
Essam Al Daoud, Ahid Al-Shbail, Adnan Al-Smadi

634 Reliability optimization using adapted ant colony algorithm under

criticality and cost constraints
Belal Ahmad Ayyoub, Asim Elsheikh

642 A comprehensive quality evaluation system for PACS

Dinu Dragan, Dragan Ivetic

651 A multi-level method for criticality evaluation to provide fault tolerance

in multi-agent systems
Mounira BOUZAHZAH, Ramdane MAAMRI

658 A modified partition fusion technique of multifocus image for improved

image quality
Dheeraj Agrawal, Al-Dahoud Ali, J.Singhai

664 Integrating biomedical ontological ontologies – OBR – SCOLIO Ontology

Vanja Luković, Danijela Milošević, Goran Devedžić
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

TESTING OF PROGRAM CORRECTNES

IN FORMAL THEORY

Ivana Berkovic
University of Novi Sad, Technical Faculty “Mihajlo Pupin”, Zrenjanin, Serbia
berkovic@tf.zr.ac.yu

Branko Markoski
University of Novi Sad, Technical Faculty “Mihajlo Pupin”, Zrenjanin, Serbia
markoni@uns.ns.ac.yu

Jovan Setrajcic
University of Novi Sad, Faculty of Sciences, Novi Sad, Serbia
bora@if.ns.ac.yu

Vladimir Brtka
University of Novi Sad, Technical Faculty “Mihajlo Pupin”, Zrenjanin, Serbia
vbrtka@tf.zr.ac.yu

Dalibor Dobrilovic
University of Novi Sad, Technical Faculty “Mihajlo Pupin”, Zrenjanin, Serbia
ddobrilo@tf.zr.ac.yu

ABSTRACT
Within software’s life cycle, program testing is very important, since quality of
specification demand, design and application must be proven. All definitions related
to program testing are based on the same tendency and that is to give answer to the
question: does the program behave in the requested way? One of oldest and best-
known methods used in constructive testing of smaller programs is the symbolic
program execution. One of ways to prove whether given program is written correctly
is to execute it symbolically. Ramified program may be translated into declarative
shape, i.e. into a clause sequence, and this translation may be automated. Method
comprises of transformation part and resolution part.This work gives the description
of the general frame for the investigation of the problem regarding program
correctness, using the method of resolution invalidation.. It is shown how the rules
of program logic can be used in the automatic resolution procedure. The examples of
the realization on the LP prolog language are given (without limitation to Horn's
clauses and without final failure).. The process of Pascal program execution in the
LP system demonstrator is shown.

Keywords: program correctness, resolution, test information, testing programs

1. INTRODUCTION testers (persons performing tests) in case when test

The program testing is defined as a process of
program execution and comparison of observed
behaviour to behaviour requested. The primary goal
of testing is to find software flaws [1], and
secondary goal is to improve self-confidence in
finds no errors. Conflict between these two goals in
visible when a testing process finds no error. In
absence of other information, this may mean that
the software is either of very high or very poor
quality.
Program testing is, in principle, complicated

UbiCC Journal – Volume 4 No. 3 618

Special Issue on ICIT 2009 Conference - Bioinformatics and Image
process that must be executed as systematically as
possible in order to provide adequate reliability and
quality certificate.
Within software lifespan, program testing is one
of most important activities since fulfillment of
specification requirements, design and application
must be checked out. According to Mantos [2], big
software producers spend about 40% of time for
program testing. In order to test large and
complicated programs, testing must be as
systematic as possible. Therefore, from all testing
methods, only one that must not be applied is ad
hoc testing method, since it cannot verify quality
and correctness regarding the specification,
construction or application. Testing firstly certifies
whether the program performs the job it was
intended to do, and then how it behaves in different
exploitation conditions. Therefore, the key element
in program testing is its specification, since, by
definition, testing must be based on it. Testing
strategy includes a set of activities organized in
well-planned sequence of steps, which finally
confirms (or refutes) fulfillment of required
software quality. Errors are made in all stages of
software development and have a tendency to
expand. A number of errors revealed may rise
during designing and then increase several times
during the coding. According to [3], program-
testing stages cost three to five times more than any
other stages in a software life span.
In large systems, many errors are found at the
beginning of testing process, with visible decline in
error percent during mending the errors in the
software itself. There are several different
approaches to program testing. One of our
approaches is given in [4]. Testing result may not
be predicted in advance. On the basis of testing
results it may be concluded how much more errors
are present in the software.
The usual approach to testing is based on
requests analyse. Specification is being converted
into test items. Apart of the fact that incorrigible
errors may occur in programs, specification
requests are written in much higher level than
testing standards. This means that, during testing,
attention must be paid to much more details than it
is listed in specification itself. Due to lack of time
or money, only parts of the software are being
tested, or the parts listed in specification.
Structural testing method belongs to another
strategy of testing approaches, so-called "white
box"" (some authors call it transparent or glass
box). Criterion of usual "white box" is to execute
every executive statement during the testing and to
write every result during testing in a testing log.
The basic force in all these testings is that complete
code is taken into account during testing, which
makes easier to find errors, even when software
details are unclear or incomplete.
UbiCC Journal – Volume 4 No. 3
According to [5] testing may be descriptive and
prescriptive. In descriptive testing, testing of all test
items is not necessary. Instead, in testing log is
written whether software is hard to test, is it stable
or not, number of bugs, etc... Prescriptive testing
establishes operative steps helping software control,
i.e. dividing complex modules in several more
simple ones. There are several tests of complex
software measurements. Important criterion in
measurement selection is equality (harmony) of
applications. It is popular in commercial software
application because it guarantees to user a certain
level of testing, or possibility of so-called internal
action [6]. There is a strong connection between
complexity and testing, and methodology of
structural testing makes this connection explicit [6].
Firstly, complexity is the basic source of software
errors. This is possible in both abstract and concrete
sense. In abstract sense, complexity above certain
point exceeds ability of the human mind to do an
exact mathematical manipulation. Structural
programming techniques may push these barriers,
but may not remove them completely. Other
factors, listed in [7], claim that when module is
more complex, it is more probable that it contains
an error. In addition, above certain complexity
threshold, probability of the error in the module is
progressively rising. On the basis of this
information, many software purchasers define a
number of cycles (software module cyclicity,
McCabe [8] 1) in order increase total reliability. On
the other hand, complexity may be used directly to
distribute testing attempts in input data by
connecting complexity and number of errors, in
order to aim testing to finding most probable errors
("lever" mechanism, [9]). In structural testing
methodology, this distribution means to precisely
determine number of testing paths needed for every
software module being tested, which exactly is the
cyclic complexity. Other usual criteria of "white
box" testing has important flaw that may be
fulfilled with small number of tests for arbitrary
complexity (using any possible meaning of the
"complexity") [10].
The program correctness demonstration and the
programming of correct programs are two similar
theoretical problems, which are very meaningful in
practice [11]. The first is resolved within the
program analysis and the second within the
program synthesis, although because of the
connection that exists between the program analysis
and the program synthesis it is noticed the
reciprocal interference of the two processes.
Nevertheless, when it is a mater of the automatic
methods that are to prove the correctness and of the
1
McCabe, measure based on a number and
structure of the cycle.
619
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
methods of automatic program synthesis, the
difference between them is evident.
In reference [12] in describes the initial
possibility of automatic synthesis of simple
programs using the resolution procedure of
automatic demonstration of theorems (ADT), more
precisely with the resolution procedure of
deduction of answer to request. The demonstration
that the request that has a form of (∃x)W(x) is the
logical consequence of the axioms that determinate
the predicate W and determinate (elementary)
program operators provides that the variable x in
the response obtains the value that represents the
requested composition of (elementary) operators,
i.e. the requested program. The works of Z. Mann,
observe in detail the problems of program analysis
and synthesis using the resolution procedure of
demonstration and deduction of the response.
The different research tendency is axiomatic
definition of the semantics of the program language
Pascal in the form of specific rules of the program
logic deduction, described in the works [14,15].
Although the concepts of the two mentioned
approaches are different, they have the same
characteristic. It is the deductive system on
predicate language. In fact, it is a mater of
realization in the special predicate computation that
is based on deduction in formal theory. With this,
the problem of program correctness is to be related
to automatic checkup of (existing) demonstrations
regarding mathematical theorems. The two
approaches mentioned above and their
modifications are based on that kind of concept.
2. DESCRIPTION OF METHOD FOR ONE
PASSAGE SYMBOLIC TESTING PROGRAM
The method is based on transformation of given
Pascal program, into a sequence of prologue
clauses, which comprise axiomatic base for
functioning of deductive resolution mechanism in a
BASELOG system [10] . For given Pascal program,
by a single passage through resolution procedure of
BASELOG system, all possible outputs in Pascal
program are obtained in a symbolic shape, together
with paths leading to every one of them. Both parts,
transformation and resolution one, are completely
automated and are naturally attached to each other.
When a resolution part has finished, a sequence of
paths and symbolic outputs is reading out for given
input Pascal program. This is a transformation of
programming structures and programming
operators into a sequence of clauses, being realized
by models depending on concrete programming
language. Automation covers branching IF-THEN
and IF-THEN-ELSE structures, as well as WHILE-
DO and REPEAT – UNTIL cyclic structures,
which may be mutually nested in each other. This
paper gives review of possibilities in work with
UbiCC Journal – Volume 4 No. 3
single-dimension sequences and programs within a
Pascal program. Number of passages through cyclic
structures must be fixed in advance using counter.
During the testing process of given (input) Pascal
program both parts are involved, transformation
and resolution, in a following way: Transformation
part
• ends function by a sequence of clauses, or
• demands forced termination, depending
on input Pascal program.
Impossibility of generating a sequence of clauses in
transformation part points that a Pascal program has
no correct syntax, i.e. that there are mistakes in
syntax or in logical structure (destructive testing).
In this case, since axiomatic base was not
constructed, resolution part is not activated and user
is prompted to mend a Pascal program syntax. In
the case that transformation part finishes function
by generating a sequence of clauses, resolution part
is activated with following possible outcomes:
Ra) ends function giving a list of symbolic
outputs and corresponding Pascal
program routes, or
Rb) ends by message that id could not generate
list of outputs and routes, or
Rc) doesn't end function and demands forced
termination.
Ra) By comparing symbolic outputs and routes
with specification, the user may
• declare a given Pascal program as
correct, if outputs are in
accordance to specification
(constructive testing), or
• if a discrepancy of some symbolic
expression to specification has
been found, this means that there
is a semantic error in a Pascal
program (destructive testing) at the
corresponding route.
Rb) Impossibility to generate a list of symbolic
expressions in resolution part, which means
that there is a logical-structural error in a
Pascal program (destructive testing).
Rc) Too long function or a (unending) cycle
means that there is a logic and/or semantic
error in a Pascal program (destructive
testing).
In this way, by using this method, user may be
assured in correctness of a Pascal program or in
presence of syntax and/or logic-structure semantic
errors. As opposite to present methods of symbolic
testing of the programs, important feature of this
method is single-passage, provided by specific
property of OL – resolution [11] with marked
literals, at which a resolution module in BASELOG
system is founded.
620
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
3. DEDUCTION IN FORMAL THEORY AND
PROGRAM CORRECTNESS
The program verification may lean on
techniques for automatic theorem proving. These
techniques embody principles of deductive
reasoning, same ones that are used by programmers
during program designation. Why not use same
principles in the automatic synthesis system, which
may construct program instead of merely proving
its correctness? Designing the program demands
more originality and more creativity than proving
its correctness, but both tasks demand the same way
of thinking. [13]
Structural programming itself helped the
automatic synthesis of computer programs in the
beginning, establishing principles in program
development on the basis of specification. These
principles should be guidelines for programmers. In
the matter of fact, advocates of structural
programming were very pessimistic regarding
possibility to ever automatize their techniques.
Dijkstra went so far to say that we should not
automatize programming even if we could, since
this would deprive this job from all delight.
Proving program correctness is a theoretical
problem with much practical importance, and is
done within program analyse. Related theoretical
problem is the design of correct programs that is
solved in another way – within program synthesis.
It is evident that these processes are intertwined,
since analysis and synthesis of programs are closely
related. Nevertheless, differences between these
problems are distinct regarding automatic method
of proving program correctness and automatic
method of program synthesis.
If we observe a program, it raises question of
termination and correctness, and if we observe two
programs we have question of equivalence of given
programs. Abstract, i.e. non-interpreted program is
defined using pointed graph. From such a program,
we may obtain partially interpreted program, using
interpretation of function symbols, predicate
symbols and constant symbols. If we interpret free
variables into partially interpreted program, a
realized program is obtained. Function of such a
program is observed using sequence executed.
Realized program, regarded as deterministic, has
one executive sequence, and if it does not exist at
all, it has no executive sequence. On the other hand,
when the program is partially interpreted, we see
several executive sequences. In previously stated
program type, for every predicate interpreted it is
known when it is correct and when not, which
would mean that depending on input variables
different execution paths are possible. Considering
abstract program, we conclude that it has only one
executive sequence, where it is not known whether
predicate P or his negation is correct.
UbiCC Journal – Volume 4 No. 3
According to The basic presumptions of
programming logic are given in [14]. The basic
relation {P}S{Q}is a specification for program S
with following meaning: if predicate P at input is
fulfilled (correct) before execution of program S,
then predicate Q at the output is fulfilled (correct)
after execution of program S. In order to prove
correctness of program S, it is necessary to prove
relation {P}S{Q}, where input values of variables
must fulfill predicate P and output variable values
must fulfill predicate Q. Since it is not proven that
S is terminating, and that this is only presumption,
then we may say that partial correctness of the
program is defined. If it is proven that S terminates
and that relation {P}S{Q} is fulfilled, we say that S
is completely correct. For program design, we use
thus determined notion of correctness.
The basic idea is that program design should be
done simultaneously with proving correctness of
the program for given specifications[15,16]. First
the specification {P}S{Q} is executed with given
prerequisite P and given resultant post condition Q,
and then subspecifications of {Pi}Si{Qi} type are
executed for components Si from which the
program S is built. Special rules of execution
provide proof that fulfillment of relation {P}S{Q}
follows from fulfillment of relations {Pi}Si{Qi} for
component programs Si.
Notice that given rules in [9] are used for
manual design and manual confirmation of
program's correctness, without mention about
possibility of automatic (resolution) confirmation
methods.If we wish to prove correctness of
program S, we must prove relation {P}S{Q}, where
input values of variables must fulfill the formula P
and output values of variables must fulfill the
formula Q. This defines only partial correctness of
program S, since it is assumed that program S
terminates. If we prove that S terminates and that
relation {P}S{Q} is satisfied, we say that S is
totally correct.Thus designated principle of
correctness is used for program designation.
Designation starts from specification {P}S{Q} with
given precondition P and given resulting
postcondition Q.Formula {P}S{Q} is written as
K(P, S, Q), where K is a predicate symbol and
P,S,Q are variables of first-order predicate
calculation.
{Pzy} z := y {P}
we are writing as K(t(P,Z,Y), d(Z,Y), P)...
where t,d are function symbols and P,Z,Y are
variables;
...Rules R(τ):
P1. . {P}S{R} , R⇒Q.
{P}S{Q}.
we write.. K(P,S,R) ∧ Im(R,Q) ⇒ K(P,S,Q)
621
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

where Im (implication) is a predicate symbol, and A(S) = R(τ) ∪A(τ). This means that derivation of
P, S, R, Q are variables; theorem B within theory τ could be replaced with
derivation within special predicate calculus S,
P2 R⇒P, {P}S{Q}
whose own axioms A(S)= R(τ) ∪A(τ).Axioms of
{R}S{Q}. we write Im(R,P) ∧ K(P,S,Q) ⇒ special predicate calculus S are: A(S)= A(τ)
K(R,S,Q) ∪R(τ).We assume that s is a syntax unit whose
P3 {P}S1{R} , {R}S2{Q} (partial) correctness is being proven for certain
input predicate U and output predicate V.
{P}begin S1; S2 end {Q}
Within theory S is being proved
K(P,S1,R) ∧ K(R,S2,Q) ⇒ K(P,s(S1,S2),Q)
where s is a function symbol, and P, S1, S2, R, q ... ⎥⎯ (∃P)(∃Q)K(P,s,Q)
are variables S
where s is a constant for presentation of a given
P4 {P∧B}S1{Q, {P∧~B}S2{Q} program. Program is written in functional notation
{P}if B then S1 else S2{Q} with symbols: s (sequence), d (assigning), ife (if-
then-else), if (if-then), wh (while), ru (repeat-until).
K(k(P,B),S1,Q)∧K(k(P,n(B)),S2,Q) ⇒ To starting set of axioms A(S), negation of
K(P,ife(B,S1,S2),Q) statement is added: Result of negation using
where k, n, ife are function symbols resolution procedure is as follows: /Im(Xθ,Yθ,)∨
Odgovor(Pθ,Qθ), where Xθ,Yθ,Pθ,Qθ are values for
P5 {P∧B}S{Q} , P∧~B ⇒ Q which successful negation To means that for these
values a proof is found. But this does not mean that
{P} if B then S{Q}
given program is partially correct. It is necessary to
K(k(P,B),S,Q) ∧ Im(k(P,n(B)),Q) ⇒ establish that input and output predicates U, V are
K(P,if(B,S),Q) in accordance with Pθ, Qθ, and also that Im (Xθ,Yθ)
is really fulfilled for domain predicates ant
where k, n, if are function symbols
terms.Accordance means confirmation that .. is
P6 {P∧B} S {P } valid. : U ⇒ Pθ, ∧ Qθ ⇒ V) ∧ ( Xθ ⇒ Yθ).there
are two ways to establish accordance: manually or
{P} while B do S {P∧~B} by automatic resolution procedure. Realization of
K(k(P,B),S,P) ⇒ K(P,wh(B,S),k(P,n(B))) these ways is not possible within theory S, but it is
possible within the new theory, which is defined by
where k, n, wh are function symbols predicates and terms which are part of the program
s and input-output predicates U, V. Within this
P7 {P}S{Q} , Q∧~B ⇒ P theory U, P, Q, V, X, Y are not variables, but
formulae with domain variables, domain terms and
{P}repeat S until B {Q∧B} domain predicates.This method concerns derivation
within special predicate calculus based on
K(P,S,Q) ∧ Im(k(Q,n(B)),P) ⇒ deduction within the formal theory. Thus the
K(P,ru(S,B),k(Q,B)) program's correctness problem is associated with
where k, n, ru are function symbols automatic proving of (existing) proofs of
mathematical theorems.
Transcription of other programming logic rules is
The formal theory τ is determined with the
also possible.
formulation of (S(τ), F(τ), A(τ), R(τ)) where S is
the set of symbols (alphabet) of the theory τ, F is
Axiom A(τ):
A1 K(t(P,Z,Y),d(Z.Y),P) the set of formulas (regular words in the alphabet
assigning axiom S), A is the set of axioms of the theory τ (A⊂F), R
Formal theory τ is given by (α(τ), F(τ), A(τ), R(τ)), is the set of rules of execution of the theory τ.
where α is a set of symbols (alphabet) of theory τ, Deduction (proof) of the formula B in the theory τ
F is a set of formulae (correct words in alphabet α), is the final sequence B1, B2, ... , Bn (Bn is B) of
A is a set of axioms for theory τ(A⊂F), R is a set of formulas of this theory, of that kind that for every
derivation rules for theory τ.B is a theorem within element Bi of that sequence it is valid: Bi is axiom,
theory τ if and only if B is possible to derive within or Bi is deducted with the application of some rules
calculus k from set R(τ) ∪A(τ) (k is a first-order of deduction Ri∈R from some preceding elements
predicate calculus).Let S be special predicate
calculus (first-order theory) with it's own axioms

UbiCC Journal – Volume 4 No. 3 622

Special Issue on ICIT 2009 Conference - Bioinformatics and Image
of that sequence. It is said that B is the theorem of
the theory τ and we write ⎯ B [17].
τ resolution rules of derivation.
Suppose S(τ) is a set of symbols of predicate
computation and F(τ) set of formulas of predicate
computation. In that case, the rules of deduction R(
τ) can be written in the form: Bi1∧Bi2∧ ... ∧Bik
⇒ Bi (Ri) where Bik, Bi are formulas from F(τ).
Suppose κ predicate computation of first line, than
it is valid:
R(τ), A(τ) ⎯ B if ⎯ B (1)
κ τ
B is theorem in the theory τ if and only if B is
deductible in computation κ from the set R(τ) ∪
A(τ).
Suppose S is a special predicate computation
(theory of first line) with its own axioms:
A(S) = R(τ) ∪ A(τ) , (rules of deduction in S are
rules of deduction of computation κ) then it is valid
A(S) ⎯ B if ⎯ B , so that (1) can be written:
κ S
⎯ B if ⎯ B (2)
S τ
That means that the deduction of theorem B in
theory τ can be replaced with deduction in special
predicate computation S, that has its own axioms
A(S) = R(τ) ∪ A(τ).
Now we can formulate the following task:
The sequence of formulas has been given B1, B2, ...
, Bn (Bn is B, Bi different from B for i<n) of theory
τ.
Implementation of programs for proving
theorems was in the beginning only in mathematics
area. When it was seen that other problems could
be presented as possible theorems which need to be
proven, application possibilities were found for
areas as program correctness, program generating,
question languages over relation databases,
electronic circuits design.
As for formal presentation where theorem is
being proven, it could be statement calculus, first-
order predicate calculus, as well as higher-order
logic. Theorems in statement calculus are simple
for contemporary provers, but statement calculus is
not expressional enough. Higher-order logic is
extremely expressional, but they have a number of
practical problems. Therefore a first-order predicate
calculus is probably the most used one.
UbiCC Journal – Volume 4 No. 3
Regarding technique of automatic theorems
proving, most investigations have been done in
Resolution is a very
important derivation rule with completeness
property.
Demonstrate that the mentioned sequence is
deduction of formula B in theoryτ.
One way of solving this problem is to verify
that the given sequence corresponds to definition of
deduction in theoryτ. The other way is to use (1),
i.e. (2):
If we demonstrate that
R(τ), A(τ) ⎯ B ∧ B2 ∧ ... ∧ Bn (3)
κ
that is sufficient for conclusion that B1, B2, ... , Bn
is deduction in τ .
And also it is sufficient to demonstrate that R(τ),
A(τ) ⎯ Bi , for i = 1,2,...,n, with this it is
κ
demonstrated (3).
Demonstration for (3) can be deducted with the
resolution invalidation of the set
R(τ)∪A(τ)∪{~B1∨~B2∨ ... ∨~Bn}, or with n
invalidations of sets R(τ)∪A(τ)∪{~Bi}.
Notice that for the conclusion that B1, B2, ..., Bn is
deduction in τ it is not enough to demonstrate R(τ),
A(τ) ⎯ (B1 ∧ B 2 ∧ ... ∧ Bn - 1 ⇒ Bn ) ,i.e. it is
κ
not enough to realize resolution invalidation of the
set R(τ)∪A(τ)∪{B1, B2, ... , Bn-1}∪{~Bn},
because this demonstrate only that Bn is deductible
in τ supposing that in τ is deductible B1∧B2∧...∧
Bn-1 .
Always when B1, B2, ..., Bn is really deduction
in τ, (B1∧B2∧...∧Bn-1 ⇒ Bn) will be correct, but
vice versa is not always valid. It can happen that
(B1∧B2∧...∧Bn-1 ⇒ Bn) is deductible in τ, but
that B1∧B2∧...∧Bn-1 is not deductible in τ, (see
example 1’).
And also, the demonstration for R(τ), A(τ)
⎯ Bn , that can be realized with resolution
κ
invalidation of the set R(τ)∪A(τ)∪{~Bn}, means
that Bn is theorem in τ, i.e. that Bn is deductible in
τ, but this is not enough for the conclusion that B1,
B2, ..., Bn is deduction in τ (except for the case that
623
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

in the invalidation appears invalidation of each Bi, 7.sode, 1.lateral :

see example 1”). =(S(2),3)&
Finally, here it is necessary to underline that not LEVEL=5;resolvent:
correspondence to set R(τ)∪A(τ)∪{~Bn} does not ~r(3,1)~r(4,2)/~r(5,3)~r(X1,2)~=(S(X1),5)&
mean not correspondence to formula ~Bn as it is, 9.side, 1.lateral :
=(S(4),5)&
but only in the presence of R(τ)∪A(τ).
LEVEL= 6; resolvent:
~r(3,1)~r(4,2)&
Example 1. Suppose A(τ) is: {r(1,1), r(1,3)} and R( 4.side, 4.lateral :
τ) contains three rules of deduction: ~r(X1,Y1)~=(S(X1),U1)~=(S(Y1),V1)r(U1,V1)&
α:r(m,n)⇒r(n,m); β: r(m,n)⇒r(m+1,n+1) ; γ LEVEL= 7; resolvent:
:r(m,n)∧r(n,p)⇒r(n,p)
symmetry correspondence with the next one ~r(3,1)/~r(4,2)~r(X1,Y1)~=(S(X1),4)~=(S(Y1),2)&
transitivity 6.side, 1.lateral :
Demonstrate that the sequence J is: r(3,1), r(4,2), =(S(1),2)&
r(5,3), r(5,1) , r is predicate symbol, one correct LEVEL= 8; resolvent:
deduction of formula r(5,1) in theory τ.
~r(3,1)/~r(4,2)~r(X1,1)~=(S(X1),4)&
8.side, 1.lateral :
It is sufficient to demonstrate:
=(S(3),4)&
{α,β,γ}, A(τ) ⎯ J . LEVEL= 9; resolvent:
κ ~r(3,1)&
In the next demonstration x+1 is signed with 2.side, 1.lateral :
S(x) and axioms for ‘the next one’ are added: r(3,1)&
LEVEL= 10; resolvent:
1 &
~r(3,1)~r(4,2)~r(5,3)~r(5,1)& DEMONSTRATION IS PRINTED
9 Example 2
r(1,1)& begin
r(3,1)& p:=x;
~r(X1,Y1)r(Y1,X1)& i:=0;
~r(X1,Y1)~=(S(X1),U1)~=(S(Y1),V1)r(U1,V1)& while i<=n do
~r(X1,Y1)~r(Y1,Z1)r(X1,Z1)& begin
=(S(1),2)& i:=i+1;
=(S(2),3)& p:=p*i;
=(S(3),4)& end;
=(S(4),5)& end.
Demonstration with invalidation: Given program is written:
number of generated resolvents = 934 s(s(d(p,x),d(i,0)),w(i<=n,s(d(i,i+1),d(p,p*i))))
maximum level = 10 Constant b is a mark for predicate i<=n
DEMONSTRATION IS PRINTED constant t1 is i+1, constant t2 is term p*i
level on which empty composition is generated = thus we obtain
10 s(s(d(p,x),d(i,0)),w(b,s(d(i,t1),d(p,t2))))
LEVEL=1; central composition
:~r(3,1)~r(4,2)~r(5,3)~r(5,1)& 1
5.side, 3.lateral : /O(X1,V1)~K(X1,s(h,g),Y1)~K(Y1,w(b,s(d(i,t1),d(
~r(X1,Y1)~r(Y1,Z1)r(X1,Z1)& p,t2))),V1)&
LEVEL= 2; resolvent: 8
~r(3,1)~r(4,2)~r(5,3)/~r(5,1)~r(5,Y1)~r(Y1,1)& ~K(Y1,d(p,x),V1)K(Y1,h,V1)& /reserve for
2.side, 1.lateral : shortening the note
r(3,1)& ~K(Y1,d(i,0),V1)K(Y1,g,V1)& /reserve for
LEVEL= 3; resolvent: shortening the note
~r(3,1)~r(4,2)~r(5,3)& ~K(X1,Y1,U1)~K(U1,Y2,V1)K(X1,s(Y1,Y2),V1)
4.side, 4.lateral : & /sequence rule
~K(k(X1,V2),U0,X1)K(X1,w(V2,U0),k(X1,ng(V2)
~r(X1,Y1)~=(S(X1),U1)~=(S(Y1),V1)r(U1,V1)& ))& /rule for while
LEVEL= 4; resolvent: K(t(X1,Z1,Y1),d(Z1,Y1),X1)& /assigning axiome
~r(3,1)~r(4,2)/~r(5,3)~r(X1,Y1)~=(S(X1),5)~= ~IM(X2,Y1)~K(Y1,U0,V1)K(X2,U0,V1)&
(S(Y1),3)& /consequence rule

UbiCC Journal – Volume 4 No. 3 624

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

~IM(Y1,V1)~K(X1,U0,Y1)K(X1,U0,V1)& /
consequence rule /~IM(k(V1,b),t(t(V1,p,t2),i,t1))/O(X2,k(V1,ng(b)))/
~O(X1,V1)& / negation addition ~K(X2,s(h,g),V1)~K(X2,h,U1)~K(U1,g,V1)&
0 2.lateral, 2.literal :
0 ~K(Y1,d(i,0),V1)K(Y1,g,V1)&
LEVEL= 8; resolvent:
LP system generates next negation
number of resolvents generated = 10 /~IM(k(V0,b),t(t(V0,p,t2),i,t1))/O(X2,k(V0,ng(b)))/
maximal obtained level = 11 ~K(X2,s(h,g),V0)~K(X2,h,Y1)/~K(Y1,g,V0)~K(Y
DEMONSTRATION IS PRINTED 1,d(i,0),V0)&
level where the empty item is generated = 11 5.lateral, 1.literal :
LEVEL=1; central item K(t(X1,Z1,Y1),d(Z1,Y1),X1)&
:/O(X1,V1)~K(X1,s(h,g),Y1)~K(Y1,w(b,s(d(i,t1),d LEVEL= 9; resolvent:
(p,t2))),V1)&
4.lateral, 2.literal : /~IM(k(X1,b),t(t(X1,p,t2),i,t1))/O(X2,k(X1,ng(b)))/
~K(X2,s(h,g),X1)~K(X2,h,t(X1,i,0))&
~K(k(X1,V2),U0,X1)K(X1,w(V2,U0),k(X1,ng(V2) 1 lateral, 2.literal :
))& ~K(Y1,d(p,x),V1)K(Y1,h,V1)&
LEVEL= 2; resolvent: LEVEL= 10; resolvent:

/O(X1,k(X0,ng(b)))~K(X1,s(h,g),X0)/~K(X0,w(b,s /~IM(k(X1,b),t(t(X1,p,t2),i,t1))/O(Y1,k(X1,ng(b)))/
(d(i,t1),d(p,t2))),k(X0,ng(b)))~K(k(X0,b),s(d(i,t1),d ~K(Y1,s(h,g),X1)/~K(Y1,h,t(X1,i,0))~K(Y1,d(p,x),
(p,t2)),X0)& t(X1,i,0))&
3.lateral, 3.literal : 5.lateras, 1.literal :
K(t(X1,Z1,Y1),d(Z1,Y1),X1)&
~K(X1,Y1,U1)~K(U1,Y2,V1)K(X1,s(Y1,Y2),V1) LEVEL= 11; resolvent:
& /O(Y1,k(X1,ng(b)))/~K(Y1,s(h,g),X1)/~K(Y1,h,t(X
LEVEL= 3; resolvent: 1,i,0))~K(Y1,d(p,x),t(X1,i,0))&
5. lateral, 1.literal :
/O(X1,k(V1,ng(b)))~K(X1,s(h,g),V1)/~K(V1,w(b,s K(t(X1,Z1,Y1),d(Z1,Y1),X1)&
(d(i,t1),d(p,t2))),k(V1,ng(b)))/~K(k(V1,b),s(d(i,t1), LEVEL= 11; resolvent:
d(p,t2)),V1)~K(k(V1,b),d(i,t1),U1)~K(U1,d(p,t2),V DEMONSTRATION IS PRINTED
1)& Now we need to prove compliance, i.e. that there is
5.lateral, 1.literal : in effect:
K(t(X1,Z1,Y1),d(Z1,Y1),X1)& ( Xθ ⇒ YθZθTθ ) ∧ (U ⇒ Pθ, ∧ Qθ ⇒ V) that is at
LEVEL= 4; resolvent: LEVEL= 12; resolvent:
/IM(k(X1,b),t(t(X1,i,t1),p,t2))O(t(t(X1,i,0),p,x),k(X
/O(X1,k(X0,ng(b)))~K(X1,s(h,g),X0)/~K(X0,w(b,s 1,ng(b)))&
(d(i,t1),d(p,t2))),k(X0,ng(b)))/~K(k(X0,b),s(d(i,t1), By getting marks to domain level we obtain:
d(p,t2)),X0)~K(k(X0,b),d(i,t1),t(X0,p,t2))& (X1 ∧ (i<=n) ⇒ X1ii+1 pp/i)i (U⇒X1i 0 px) ∧ (X1 ∧
¬
6.lateral, 3.literal : (i<=n) ⇒ V)7
~IM(X2,Y1)~K(Y1,U0,V1)K(X2,U0,V1)& i
LEVEL= 5; resolvent: Putting X1: p = x ⋅ ∏ ( j − 1) we obtain
j =0
/O(X1,k(X0,ng(b)))~K(X1,s(h,g),X0)/~K(X0,w(b,s following correct implications:
(d(i,t1),d(p,t2))),k(X0,ng(b)))/~K(k(X0,b),s(d(i,t1),
d(p,t2)),X0)/~K(k(X0,b),d(i,t1),t(X0,p,t2))~IM(k(X
0,b),Y1)~K(Y1,d(i,t1),t(X0,p,t2))&
5. lateral, 1.literal :
K(t(X1,Z1,Y1),d(Z1,Y1),X1)&
LEVEL= 6; resolvent:

/~IM(k(X0,b),t(t(X0,p,t2),i,t1))/O(X1,k(X0,ng(b)))
~K(X1,s(h,g),X0)&
3.lateral, 3.literal :

~K(X1,Y1,U1)~K(U1,Y2,V1)K(X1,s(Y1,Y2),V1)
&
LEVEL= 7; resolvent:

UbiCC Journal – Volume 4 No. 3 625

Special Issue on ICIT 2009 Conference - Bioinformatics and Image
i program development. Software producers would
p = x ⋅ ∏ ( j − 1) =>
j =0
i quality of product bought and difficulties in
p ⋅ i = x ⋅ i ⋅ ∏ ( j − 1) =>
j =0
i package, which is the best proof that it is a very
p ⋅ i = x ⋅ (i − 1 + 1) ⋅ ∏ ( j − 1) =>
j =0
i +1
p ⋅ i = x ⋅ ∏ ( j − 1)
j =0
For i = 0 we obtain:
i choose such a testing method that will, in given
p = x ⋅ ∏ ( j − 1) =>
j =0
0
p = x ⋅ ∏ ( j − 1) =>
j =0
p=x
By this the compliance is proven, which is enough
to conclude that a given program is (partially)
correct (until terminating).
5. INTERPRETATION RELATED TO
DEMONSTRATION OF PROGRAM
CORRECTNESS
Interpret the sequence J: B1, ... , Bn as program
S. Interpret the elements A(τ) as initial elements for
the composition of program S, and the elements R(
τ) as rules for the composition of program
constructions.
Vice versa, if we consider program S as
sequence J, initial elementary program operators as
elements A(τ) and rules for composition of
program structures as elements R(τ), with this the
problem of verification of the correctness of the
given program is related to demonstration of
correctness of deduction in corresponding formal
theory. It is necessary to represent axioms, rules
and program with predicate formulas.
With all that is mentioned above we defined the
general frame for the composition of concrete
proceedings for demonstration of program
correctness with the deductive method. With the
variety of choices regarding axioms, rules and
predicate registration for the different composition
proceedings are possible.
6. CONCLUSION
Software testing is the important step in
UbiCC Journal – Volume 4 No. 3
like to predict number of errors in software systems
before the application, so they could estimate
maintenance process [18]. Testing often takes 40%
of time needed for development of software
complex process. Aim of testing is to establish
whether software is behaving in the way envisaged
by specification. Therefore, primary goal of
software testing is to find errors. Nevertheless, not
all errors are ever found, but there is a secondary
goal in testing, that is to enable a person who
performs testing (tester) to trust the software system
[19]. From these reasons, it is very important to
functions of software system, find those fatal errors
that bring to highest hazards. In order to realize
this, one of tasks given to programmers is to
develop software that is easy to test ("software is
designed for people, not for machines") [20].
Program testing is often equalized to looking for
any errors [20]. There is no point in testing for
errors that probably do not exist. It is much more
efficient to think thoroughly about kind of errors
that are most probable (or most harmful) and then
to choose testing methods that will be able to find
such errors. Success of a set of test items is equal to
successful execution of detailed test program. One
of big issues in program testing is the error
reproduction (testers find errors and programmers
remove bugs) [21]. It is obvious that there must be
some coordination between testers and
programmers. Error reproduction is the case when
it would be the vest to do a problematic test again
and to know exactly when and where error
occurred. Therefore, there is no ideal test, as well as
there is no ideal product.[22] .Software producers
would like to anticipate the number of errors in
software systems before their application in order to
estimate the quality of acquired program and the
difficulties in the maintenance. This work gives the
summary and describes the process of program
testing, the problems that are to be resolved by
testers and some solutions for the efficacious
elimination of errors[23]. The testing of big and
complex programs is in general the complicated
process that has to be realized as systematically as
possible, in order to provide adequate confidence
and to confirm the quality of given application [24].
The deductions in formal theories represent general
frame for the development of deductive methods
for the verification of program correctness. This
frame gives two basic methods (invalidation of
added predicate formula and usage of rules of
program logic) and their modifications.
The work with formula that is added to the
given program implies the presence of added
axioms and without them, the invalidation cannot
626
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
be realized. The added axioms describe
characteristics of domain predicates and operations
and represent necessary knowledge that is to be
communicated to the deductive system. The
existing results described above imply that kind of
knowledge, but this appears to be notable difficulty
in practice.
ACKNOWELDGEMENTS
The work presented in the paper was developed
within the IT Project “WEB
portals for data analysis and consulting,” No.
13013, supported by the
government of Republic of Serbia, 2008. – 2010.
7. REFERENCES
[1] Marks, David M. “Testing very big systems”
New York:McGraw-Hill, 1992
[2] Manthos A., Vasilis C., Kostis D.
“Systematicaly Testing a Real-Time Operating
System“ IEEE Trans. Software Eng., 1995
[3] Voas J., Miller W. “Software Testability: The
New Verification“ IEEE Software 1995
[4] Perry William E. “Year 2000 Software Testing“
New York: John Wiley& SONS 1999
[5] Whittaker J.A., Whittaker, Agrawal K.
“A case study in software reliability
measurement“ Proceedinga of Quality Week,
paper no.2A2, San Francisko, USA 1995
[6] Zeller A. “Yesterday, my program worked,
Today, it does not. Why?”Passau Germany,
2000
[7] Markoski B., Hotomski P., Malbaski D.,
Bogicevic N. “Testing the integration and the
system“, International ZEMAK symposium,
Ohrid, FR Macedonia, 2004.
[8] McCabe, Thomas J, &Butler, Charles W.
“Design Complexity Measurement and Testing
“Communications of the ACM 32, 1992
[9] Markoski B., Hotomski P., Malbaski D.
“Testing the complex software“, International
ZEMAK symposium, Ohrid, FR Macedonia,
2004.
[10] Chidamber, S. and C. Kemerer, “Towards a
Metrics Suite for Object Oriented Designe”,
Proceedings of OOPSLA, July 2001
[11] J.A. Whittaker, “What is Software Testing?
And Why Is It So Hard?” IEEE Software, vol.
17, no. 1, 2000,
[12] Nilsson N., “Problem-Solving Methods in
Artificial Intelligence “, McGraw-Hill, 1980
[13] Manna Z., “Mathematical Theory of
Computation “, McGraw-Hill, 1978
[14] Floyd. R.W., “Assigning meanings to
programs “, In: Proc. Sym. in Applied
Math.Vol.19, Mathematical Aspects of
Computer Science, Amer. Math. Soc., pp. 19-
32., 1980.
UbiCC Journal – Volume 4 No. 3
[15] Hoare C.A.R. “Proof of a program” Find
Communications of the ACM 14, 39-45.
1971.
[16] Hoare C.A.R, Wirth N., “An axiomatic
definition of the programming language
Pascal “, Acta Informatica 2, pp. 335-355.
1983
[17] Markoski B., Hotomski P., Malbaski D.,
Obradovic D. “Resolution methods in proving
the program correctness “, YUGER, An
international journal dealing with theoretical
and computational aspects of operations
research, systems science and menagement
science, Beograd, Serbia, 2007,
[18] Myers G.J., “The Art of Software Testing, New
York ” , Wiley, 1979.
[19] Chan, F., T. Chen, I. Mak and Y. Yu,
“Proportional sampling strategy: Guidelines
for software test practitioners “, Information
and Software Technology, Vol. 38, No. 12,
pp. 775-782, 1996.
[20] K. Beck, “Test Driven Development: By
Example”, Addison-Wesley, 2003
[21] P. Runeson, C. Andersson, and M. Höst, “Test
Processes in Software Product Evolution—A
Qualitative Survey on the State of Practice”,
J. Software Maintenance and Evolution, vol.
15, no. 1, 2003, pp. 41–59.
[22] G. Rothermel et al., “On Test Suite
Composition and Cost-Effective Regression
Testing”, ACM Trans. Software Eng. and
Methodology, vol. 13, no. 3, 2004, pp. 277–33
[23] N. Tillmann and W. Schulte, “Parameterized
Unit Tests”, Proc. 10th European Software
Eng. Conf., ACM Press, 2005, pp. 253–262.
[24] Nathaniel Charlton “Program verification
with interacting analysis plugins” Formal
Aspects of Computing. London: Aug 2007.
Vol. 19, Iss. 3; p. 375
627
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

Detecting Metamorphic viruses by using Arbitrary Length of Control Flow

Graphs and Nodes Alignment
Essam Al daoud
Zarka Private University, Jordan
essamdz@zpu.edu.jo

Ahid Al-Shbail
Al al-bayt University, Jordan
ahid_shbail@yahoo.com

Adnan M. Al-Smadi
Al al-bayt University, Jordan
smadi98@aabu.edu.jo

ABSTRACT
Detection tools such as virus scanners have performed poorly, particularly when
facing previously unknown virus or novel variants of existing ones. This study
proposes an efficient and novel method based on arbitrary length of control flow
graphs (ALCFG) and similarity of the aligned ALCFG matrix. The metamorphic
viruses are generated by two tools; namely: next generation virus creation kit
(NGVCK0.30) and virus creation lab for Windows 32 (VCL32). The results show
that all the generated metamorphic viruses can be detected by using the suggested
approach, while less than 62% are detected by well-known antivirus software.

Keywords: metamorphic virus, antivirus, control flow graph, similarity

measurement.

1. INTRODUCTION W32/Ghost contains many procedures and generates

Virus writers use better evasion techniques to
transform their virus to avoid detection. For example,
polymorphic and metamorphic are specifically
designed to bypass detection tools. There is strong
evidence that commercial antivirus are susceptible to
common evasion techniques used by virus writers[1].
Metamorphic Virus can reprogram itself. it use code
obfuscation techniques to challenge deeper static
analysis and can also beat dynamic analyzers by
altering its behavior, it does this by translating its
own code into a temporary representation, edit the
temporary representation of itself, and then write
itself back to normal code again. This procedure is
done with the virus itself, and thus also the
metamorphic engine itself undergoes changes.
Metamorphic viruses use several metamorphic
transformations, including Instruction reordering,
data reordering, inlining and outlining, register
renaming, code permutation, code expansion, code
shrinking, Subroutine interleaving, and garbage code
insertion. The altered code is then recompiled to
create a virus executable that looks fundamentally
different from the original. For example, The source
code of the metamorphic virus Win32/Simile is
approximately 14,000 lines of assembly code. The
metaphoric engine itself takes up approximately 90%
of the virus code, which is extremely powerful[2].
huge number of metamorphic viruses, it can generate
at least 10! = 3,628,800 variations[3].
In this paper, we develop a methodology for
detecting metamorphic virus in executables. we have
initially focused our attention on viruses and simple
entry point infection. However, our method is
general and can be applied to any malware and any
obfuscated entry point.
2. RELATED WORKS
Lakhotia, Kapoor, and Kumar believe that
antivirus technologies could counter attack using the
same techniques that metamorphic virus writers use;
identify similar weak spots in metamorphic viruses
[4]. Geometric detection is based on modifications
that a virus has made to the file structure. Peter Szor
calls this method shape heuristics because is far from
exact and prone to false positives [5]. In 2005 Ando,
Quynh, and Takefuji introduced a resolution based
technique for detecting metamorphic viruses. In their
method, scattered and obfuscated code is resolved
and simplified to several parts of malicious code.
Their experiment showed that compared with
emulation, this technique is effective for
metamorphic viruses which apply anti-heuristic
techniques, such as register substitution or

Ubiquitous Computing and Communication Journal

UbiCC Journal – Volume 4 No. 3 628
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
permutation methods[6]. In 2006 Rodelio and others
use code transformation method for undoing the
previous transformations done by the virus. Code
transformation is used to convert mutated
instructions into their simplest form, where the
combinations of instructions are transformed to an
equivalent but simple form [7]. Mohamed and others
use engine-specific scoring procedure that scans a
piece of code to determine the likelihood [8].
Bruschi, Martignoni, and Monga proposed a
detection method control flow graph matching.
Mutations are eliminated through code normalization
and the problem of detecting viral code inside an
executable is reduced to a simpler problem[9]. Wong
and Stamp experimented with Hidden Markov
models to try to detect metamorphic malware. They
concluded that in order to avoid detection,
metamorphic viruses also need a degree of similarity
with normal programs and this is something very
challenging for the virus writer[10].
3. THE PROPOSED METHOD
This section introduces new procedures to extract
partial control flow graph of any binary file. Two
main points are considered during the development
of the suggested algorithms, first point is to reorder
the flow of the code by handling "jmp" and "call"
instructions, and second point is to use one symbol
for all alternatives and equivalent instructions. The
output of Algorithm 1 is stored in the matrix ALCFG
and contains arbitrary number of the nodes.
Moreover the sequence of the nodes is represented
by using symbols to be used in the similarity
measurement.
Algorithm 1: Construction of Arbitrary length of
Control Flow Graph (ALCFG)
Input: Disassembled portable executable file (x), the
number of the file lines (n), the start
location ( j), the required number of the
nodes (m).
Output: ALCFG m×m matrix and node sequence
array NodeSeq contains m nodes
Steps:
1- Call prepare op matrix (the size of op matrix is
n×4)
2- Call prepare the matrices Labels and JumpTo (the
size is c×2 and e×3)
3- Call Construct the matrix ALCFG
Algorithm 2: Prepare op matrix (the size of op
matrix is n×4)
Input: Disassembled portable executable file (x), the
number of the file lines (n), the start
location (j), the required number of the
nodes (m).
Ubiquitous Computing and Communication Journal
UbiCC Journal – Volume 4 No. 3
Output: op matrix of size n×4 (this matrix contains
the jump instructions and the labels)
1- Load the matrix op[n][4] from the file x. Where
the opcode i, is stored at the row i, the column
op[i][1] will be used to store the labels (for
simplicity we will consider each label as an
opcode), the column op[i][2] will be used to
store the instructions (mov ,jmp, add,…). The
column op[i][3] will be used to store the first
operand, the column op[i][4] will be used to
mark the rows that are processed, assume that
default value is 0.
2- Delete the rows that do not contain label or jump
instructions (jump instructions such as call, ret,
jmp, ja, jz, je…). In this step a special action must
be consider if the "ret" instruction is preceded
directly by push instruction, in this case "ret" is
replaced by "jmp" and its operand is replaced by
the value which has pushed.
3- Rename all the conditional jump instructions to
the names in the Table 1.
4- Add to the end of the matrix a row contains
op[n+1][2]="end"
5- Delete the rows that contain inaccessible label
(this means that op[i][3] does not equal to this
label for all i)
6- Delete the rows that contain unreachable operand
(this means that op[i][1] does not equal to this
operand for all i)
Algorithm 3: Prepare the matrices Labels and
JumpTo
Input: op matrix of size n×4
Output: The matrix Labels of size c×2 and the
matrix JumpTo of size e×3
Do the following while count <= m
If op[j][4]=1 then
stack2.pop j
if j = -1 then stack1.pop j
if j= -1 then break
else if op[j][2]="call" then
stack1.push j+1; j=z+1 where
op[z][1]= op[j][3]
else if op[j][2]="ret" then
stack1.pop j
else if op[j][2]= "jmp" then
j=z+1 where op[z][1]= op[j][3]
else if op[j][2]="A" ,"N", .. or "L" then
stack2.push z ,where op[z][1]= op[j][3]
JumpTo [e][1]= op[j][3];
JumpTo [e][2]= m;
JumpTo [e][3]= op[j][2]
m=m+1;e=e+1; j=j+1
else if op[j][1] <> "null" then //label
Labels[c][1]= op[j][1];
Labels[c][2]= m
c=c+1; m=m+1; j=j+1
else if op[j][2]="end"and m<=count then
stack2.pop j
629
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

if j = -1 then break start:

…..
pop esi
sub esi, $-1-start
Algorithm 4:Construct the matrix ALCFG push esi
Input: The matrix Labels of size c×2 and the matrix ….
JumpTo of size e×3 jne tsr_complete
Output:ALCFG represented as m×m matrix and shl edi, 9
….
nodes sequence NodeSeq contains m je tsr_complete
nodes tsr:
1- Fill the upper minor diagonal of matrix ALCFG int 3
by 1 call c000_rw
pusha
2- Fill the array NodeSeq by "K" // labels mov ecx, virsize
3- for each row i in the matrix JumpTo call c000_ro
x=JumpTo[i][2]; tsr_complete:
NodeSeq[x]= JumpTo[i][3] out 80h, al
….
for each row j in the matrix Labels
if JumpTo[i][1]= Labels[j][1] then Figure 1: part from Z0mbie III
y= Labels[j][2] ; ALCFG[x][y]=1
N tsr_complete 0
A tsr_complete 0
Table 1: the instructions and corresponding symbols tsr 0
Instructions Symbol call c000_rw 0
call c000_ro 0
JE, JZ, A H __cycle_1 0
JP, JPE R E __mz 0
tsr_complete 0
JNE,JNZ N restore_program 0
JNP, JPO D A __exit 0
JA, JNBE, JG, JNLE E A restore_program 0
JAE,JNB,JNC, JGE, JNL Q N __exit 0
cf8_io 0
JB, JNAE, JC, JL, JNGE G
- - - -
JBE, JNA, JLE,JNG H - - - -
JO, JS I - - - -
JNO, JNS, JCXZ, JECXZ L
LOOP P
LABEL K Figure 2: the op matrix
GAP M
1 tsr_complete N
All above algorithms can be implemented very 2 tsr_complete A
3 __cycle_1 H
fast and can be optimized. The worst case of 4 __mz E
algorithm 2 is 5n where n is the number of the lines 9 __exit A
in the disassembled file, the worst case of algorithm 10 restore_program A
3 is n and the worst case of algorithm 4 is (m 2) 2 11 __exit N
12 __exit Q
where m ≤ n. Therefore; the total complexity of 14 __exit G
algorithm 1 is O(n)+O(m2). 15 __mz A
17 __exit N
19 __exit I
Definition 1: A skeleton signature of a binary file is
20 __cycle_2_next G
the nodes sequence NodeSeq and the matrix
ALCFG. Figure 3: The Labels Matrix

To illustrate the previous procedures; consider the 5 tsr

input is the virus Z0mbie III, where Figure 1 is part 6 cf8_io
from the source code of Z0mbie III, Figure 2 is the 7 tsr_complete
8 restore_program
op matrix, figure 3 is the Labels matrix and figure 4 13 __cycle_1
is JumpTo matrix of the first 20 nodes of the virus 16 __mz
Z0mbie III. 18 __cycle_2
Figure 4: The JumpTo Matrix

Ubiquitous Computing and Communication Journal

UbiCC Journal – Volume 4 No. 3 630
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

The following is the skeleton signature of Z0mbie III
which is consist from the sequence of the first 10
nodes NodeSeq and the matrix ALCFG:
NAHEKKKKAA
 1 1 

 1 1 


1 
 1
 1
ALCFG 10 ×10 = 
 1
 1
 
 1
 
 1
 1 
4. SIMILARITY MEASURE FUNCTION
To detect the metamorphic viruses that preserve
its control flow graph during the propagation, we can
simply compare ALCFG matrices, but if the control
flow graph is changed during the propagation then a
similarity measure function must be used.
Unfortunately the current similarity measurement
functions such as Euclidean distance, Canberra
distance or even measurements based on neural
network can not be used; the reason is the random
insertion and deletion in the nodes sequence of the
generated control flow graph. In this section we
propose a new similarity measure function to detect
the metamorphic viruses. Consider the following
definitions:
Definition 2: The diagonal sub-block of size m× m
of the matrix ALCFG which has the size n× n is the
matrix A and denoted by A p ALCFG, where the
first row and column start at i+1<n, the last row and
column end at i+m<=n and i is any integer number
less than n.
Definition 3: Let ALCFGp denotes to ALCFG matrix
of size n× n of the program P and ALCFGV denotes
to ALCFG matrix of size m× m of the virus V.
Definition 4: The matrices ALCFGS and ALCFGV
are similar if the following conditions are satisfied:
1- Alignment(NodeSeqS, NodeSeqV)= c ≥T
2- DelMis&Comp(ALCFGS, ALCFGV)=1
We will denote to the similarity measure function by
ϕ such that:

c if satisfied

ϕ ( ALCFG s , ALCFGv ) = 
 0 else

Definition 5: The program P is infected by the virus
V if and only if ϕ ( ALCFGs , ALCFGv ) = c , where
ALCFGs p ALCFGp.
For simplicity we will focus on viruses that use
simple entry point infection, therefore i=0. However

our approach can be applied to any obfuscated entry

point


 Algorithm 5: Check whether the program P is

 infected by the virus V or not.
 Input: The program P, the matrix ALCFGV and a
threshold T, where V is a virus in the
database
Output: yes if infected or no if the program is not
infected
1- Disassemble the program P (In this study the
software IDA Pro 4.8 is used, but this process can
be implemented and embedded in one software)
2- Call Algorithm 1 to find ALCFGp and NodeSeqp
(in this study the first sub block is processed
which is equivalent to the simple entry point.
However to check all the possible entry points we
have to process all m× m sub block in the matrix
ALCFGp)
3- Call Algorithm 6 to find The Percentage c and
the sequence A
4- If c ≥ T then
Call algorithm 7 to Delete the mismatch nodes
and compare the matrices
If algorithm 7 retrun 1 then
Return "Yes"
Else
Return "No"
Else
Return "No"
Algorithm 6: The Alignment of two sequences.
Alignment ( , )
Input: The sequences NodeSeqS and NodeSeqV
Output: The Percentage c and the sequence A, where
c represents the percentage of the match
node to the total number of the nodes and
A contains the index of the mismatched
nodes
1- Apply Needleman-Wunsch-Sellers algorithm
on the sequences NodeSeqS and NodeSeqV
2- Store the index of mismatch nodes in the array
A
3- Find c= number of matched nodes*100/ total
number of nodes
Algorithm 7: Delete the mismatch nodes and
compare. DelMis&Comp(, )
Input: ALCFGS, ALCFGV and the mismatched
sequence A.
Output: 0 or 1
1- If mismatch with gab then delete the row i and
the column i from the matrix ALCFGS for all i in

Ubiquitous Computing and Communication Journal

UbiCC Journal – Volume 4 No. 3 631
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
the mismatched nodes, and delete the last rows
and columns from ALCFGV where the number of
the deleted rows and columns equal to the
number of the gabs
2- If mismatch with symbol then delete the row i
and the column i from the matrices ALCFGS and
ALCFGV for all i in the mismatched nodes.
3- Rename the matrices to ALCFGs d and
ALCFGv d .
4- If ALCFGs d = ALCFGv d then
Return 1
Else
Return 0
The most expensive step in the previous algorithms
is Needleman-Wunsch-Sellers algorithm which can
be implemented in m2 operation, and the total
complexity of all procedures is O(n)+O(m2).
Therefore the suggested method is much faster than
the previous methods; for example the cost of
finding the isomorphic sub graph in [9] is well
known NP-complete problem.
To illustrate the suggested similarity measure
function, assume that we like to the check weather
the program P is infected by the virus Z0mbie III or
not. Assume that the threshold T=70 and m=10 (note
that: to reduce the false positive we must increase the
threshold and the number of the processed nodes),
the first 10 nodes that are extracted from P and the
ALCFG matrix are (the skeleton signature of P):
NAHAEKKKKA
 1 1 
 1 1 
 
 1  instructions. The generated viruses are divided into
 
 1 1 
 1 
ALCFG s =   analyzing and T2 contains 60 viruses for testing.
 1 
 1  Again 100% of the generated viruses are detected by
 
 1  the proposed method, 84% are detected by Norman,
 
 1
 
By using algorithm 6 the nodes of P aligned with the
nodes of Z0mbie III as following:
NAHAEKKKK A -
NAH -EKKKKA A
c= number of matched nodes*100/ total number of
nodes=9*100/10=90 >T.
The mismatch occur with gabs; therefore column 4
and row 4 must be deleted from ALCFGS, column 10
and row 10 must be deleted from ALCFGV. Since
matrices after deletion are identical, we conclude that
Ubiquitous Computing and Communication Journal
UbiCC Journal – Volume 4 No. 3
the program P is infected by a modified version of
Z0mbie III and ϕ ( ALCFGs , ALCFGv ) = 90% .
5. IMPLEMENTATION
The metamorphic viruses are taken from VX
Heavens search engine and generated by two tools;
namely: Next Generation Virus Creation Kit
(NGVCK0.30) and Virus Creation Lab for Windows
32 (VCL32) [11]. Since the output of the kits was
already in the asm format, we used Turbo Assembler
(TASM 5.0) for compiling and linking the files to
generate exe’s, which are later disassembled using
IDA pro 4.9 Freeware Version. Algorithm 4 is
implemented by using MATLAB 7.0. The
NGVCK0.30 has advanced assembly source-
morphing engine, and all variants of the viruses
generated by NGVCK will have the same
functionality, but they have different signatures. In
this study; 100 metamorphic viruses are generated by
using (NGVCK). 40 viruses are used for analyzing
and 60 viruses are used for testing, let us call the first
group A1 and the second group T1. After applying
the suggested procedures on A1 we note that all the
viruses in A1 have just seven different skeleton
signatures when T=100 and m=20 and have four
different skeletons when T=80 and m=20 and have
three different skeletons when T=70 and m=20. T1
group is tested by using 7 antivirus software; the
results are obtained by using the on-line service [12].
100% of the generated viruses are recognized by the
proposed method and by McAfee, but none of the
viruses are detected by using the rest software.
Another 100 viruses are generated by using VCL32,
where all of them are obfuscated manually by
inserting dead code, transposition the code,
reassigning the registers and substituting the
two groups, A2 and T2, A2 contains 40 viruses for
23% are detected by McAfee and 0% are detected by
the rest software. Figure 5 describes the average
detection percentage of the metamorphic viruses in
T1 and T2.
6. CONCLUSION
The antivirus software trying to detect the viruses by
using variant static and dynamic methods. However;
all the existing methods are not adequate. To develop
new reliable antivirus software some problems must
be fixed. This paper suggested new procedures to
detect the metamorphic viruses by using arbitrary
length of control flow graphs and nodes alignment.
The suspected files are disassembled, the opcode
encoded, the control flow analyzed, and the
632
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

similarity of the matrices is measured by using a new [8] M. R. Chouchane and A. Lakhotia: Using engine
similarity measurement. The implementation of the signature to detect metamorphic malware, In
suggested approach show that all the generated WORM '06: Proceedings of the 4th ACM
metamorphic viruses can be detected while less than workshop on Recurring malcode, New York,
62% are detected by other well known antivirus NY, USA, pp. 73-78, (2006).
software. [9] D. Bruschi, L. Martignoni, and M.Monga:
Detecting self-mutating malware using control
flow graph matching, In DIMVA, pp. 129-143,
120 (2006).
[10] W. Wong and M. Stamp: Hunting for
100 metamorphic engines, Journal in Computer
Virology, vol 2 (3), pp. 211-229, (2006).
[11] http://vx.netlux.org/ last access March (2009).
80
[12] http://www.virustotal.com/ last access March
(2009).
60

40

20

0
G
AV

an
ee
ft

ed
AV
y
nt e
so

rm

os
am
er s

cA
o

ma

No

op
ic r

Cl
sp

M
Sy

Pr
M

Ka

Figure 5: The average percentage of the detected

viruses from group T1 and T2.

REFERENCES
[1] M. Christodorescu, J. Kinder, S. Jha, S.
Katzenbeisser, and H. Veith: Malware
Normalization, Technical Report # 1539 at the
Department of Computer Sciences, University
of Wisconsin, Madison, (2005).
[2] F. Perriot: Striking Similarities: Win32/Simile
and Metamorphic Virus Code, Symantec
Corporation (2003).
[3] E. Konstantinou: Metamorphic Virus: Analysis
and Detection Technical Report, RHUL-MA-
2008-02 Department of Mathematics Royal
Holloway, University of London, (2008).
[4] A. Lakhotia, A. Kapoor, and E. U. Kumar: Are
metamorphic computer viruses really invisible?,
part 1. Virus Bulletin, pp 5-7, (2004).
[5] P. Szor: The Art of Computer Virus Research and
Defense, Addison Wesley Professional, 1
edition, pp. 10-33 (2005).
[6] R. Ando, N. A. Quynh, and Y. Takefuji:
Resolution based metamorphic computer virus
detection using redundancy control strategy, In
WSEAS Conference, Tenerife, Canary Islands,
Spain, Dec. pp. 16-18. (2005).
[7] R. G. Finones and R. T. Fernande: Solving the
metamorphic puzzle, Virus Bulletin, pp. 14-19,
(2006).

Ubiquitous Computing and Communication Journal

UbiCC Journal – Volume 4 No. 3 633
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

RELIABILITY OPTIMIZATION USING ADAPTED ANT COLONY

ALGORITHM UNDER CRITICALITY AND COST CONSTRAINTS

Belal Ayyoub Asim El-Sheikh

Al-Balqa‟a Applied University- FET - Arab Academy for Banking and
Computer Engineering Dep, Jordan Financial Sciences (AABFS)
belal_ayyoub@hotmail.com a.elsheikh@aabfs.org

ABSTRACT

Reliability designers often try to achieve a high reliability level of systems. The
problem of system reliability optimization where complex system is considered.
The system reliability maximization subject to component‟s criticality and cost
constraints is introduced as reliability optimization problem (ROP). A procedure,
which determines the maximal reliability of non series–non parallel system
topologies is proposed. In this procedure, system components are chosen to be
maximized according to it‟s criticalities. To evaluate the systems reliability, an
adapting approach is used by the ant colony algorithm (ACA) to determine the
optimal system reliability. The algorithm has been thoroughly tested on bench mark
problems from literature. Our numerical experiences show that our approach is
promising especially for complex systems. The proposed model proves to be robust
with respect to its parameters.

Key Words: System reliability, Complex system, Ant colony, Component‟s criticality.

1 INTRODUCTION Boolean truth Tables, etc.) can be used to

System reliability can be defined as the
probability that a system will perform its intended
function for a specified period of time under stated
conditions [1]. Many modern systems, both
hardware and software, are characterized by a high
degree of complexity. To enhance the reliability of
such systems, it is vital to define techniques and
models aimed at optimizing the design of the system
itself. This paper presents a new metaheuristic-
based algorithm aimed at tackling the general
system reliability problem, where one wants to
identify the system configuration that maximizes the
overall system reliability, while taking into account
a set of resource constraints. Estimating system
reliability is an important and challenging problem
for system engineers. [2]. It is also challenging since
current estimation techniques require a high level of
background in system reliability analysis, and thus
familiarity with the system. Traditionally, engineers
estimate reliability by understanding how the
different components in a system interact to
guarantee system success. Typically, based on this
understanding, a graphical model (usually in the
form of a fault tree, a reliability block diagram or a
network graph) is used to represent how component
interaction affects system functioning. Once the
graphical model is obtained, different analysis
methods [3–5] (minimal cut sets, minimal path sets,
quantitatively represent system reliability. Finally,
the reliability characteristics of the components in
the system are introduced into the mathematical
representation in order to obtain a system-level
reliability estimate. This traditional perspective aims
to provide accurate predictions about the system
reliability using historical or test data. This
approach is valid whenever the system success or
failure behavior is well understood. In their paper,
Yinong Chen, Zhongshi He, Yufang Tian [6],they
classified system reliability in to topological and
flow reliability. They considered generally that the
system consists of a set of computing nodes and a
set of components between nodes. They assume that
components are reliable while nodes may fail with
certain probability, but in this paper we will
consider components subject to failure in a
topological reliability. Ideally, one would like to
generate system design algorithms that take as input
the characteristics of system components as well as
system criteria, and produce as output an optimal
system design, this is known as system synthesis[7],
and it is very difficult to achieve. Instead, we
consider a system that is already designed then try
to improve this design by maximizing the
components reliability which will maximize the over
all system reliability. In the most theoretical
reliability problems the two basic methods of
improving the reliability of systems are improving

UbiCC Journal – Volume 4 No. 3 634

Special Issue on ICIT 2009 Conference - Bioinformatics and Image
the reliability of each component or adding
redundant components [8]. Of course, the second
method is more expensive than the first. Our paper
considers the first method. The aim of this paper is
to obtain the optimal system reliability design with
the following constrains. :
1: Basic linear-cost-reliability relation used for
each component [7].
2: Criticality of components [9]. The designer
should take this in to account before building a
reliable system and according to criticality of
component increasing reliabilities will go toward the
most critical component. Components‟ criticality
can be derived from its failure effects to system
reliability failure. Which the position of a
component will play an important role for its
criticality which we called it the index of criticality.
2 SYSTEM RELIABILITY PROBLEM
2.1 Literature view
Many methods have been reported to
improve system reliability. Tillman, Hwang, and
Kuo [10] provide survey of optimal system
reliability. They divided optimal system reliability
models into series, parallel, series-parallel, parallel-
series, standby, and complex classes. They also
categorized optimization methods into integer
programming, dynamic programming, linear
programming, geometric programming, generalized
Lagrangian functions, and heuristic approaches. The
authors concluded that many algorithms have been
proposed but only a few have been demonstrated to
be effective when applied to large-scale nonlinear
programming problems. Also, none has proven to be
generally superior. Fyffe, Hines, and Lee [11]
provide a dynamic programming algorithm for
solving the system reliability allocation problem. As
the number of constraints in a given reliability
problem increases, the computation required for
solving the problem increases exponentially. In
order to overcome these computational difficulties,
the authors introduce the Lagrange multiplier to
reduce the dimensionality of the problem. To
illustrate their computational procedure, the authors
use a hypothetical system reliability allocation
problem, which consists of fourteen functional units
connected in series. While their formulation
provides a selection of components, the search
space is restricted to consider only solutions where
the same component type is used in parallel.
Nakagawa and Miyazaki [12] proposed a more
efficient algorithm. In their algorithm, the authors
use surrogate constraints obtained by combining
multiple constraints into one constraint. In order to
demonstrate the efficiency of their algorithm, they
also solve 33 variations of the Fyffe problem. Of the
33 problems, their algorithm produces optimal
solutions for 30 of them. Misra and Sharma [13]
presented a simple and efficient technique for
UbiCC Journal – Volume 4 No. 3
solving integer-programming problems such as the
system reliability design problem. The algorithm is
based on function evaluations and a search limited
to the boundary of resources. In the nonlinear
programming approach, Hwang, Tillman and Kuo
[14] use the generalized Lagrangian function
method and the generalized reduced gradient
method to solve nonlinear optimization problems
for reliability of a complex system. They first
maximize complex-system reliability with a tangent
cost-function and then minimize the cost with a
minimum system reliability. The same authors also
present a mixed integer programming approach to
solve the reliability problem [15]. They maximize
the system reliability as a function of component
reliability level and the number of components at
each stage. Using a genetic algorithm (GA)
approach, Coit and Smith [16], [17], [18] provide a
competitive and robust algorithm to solve the
system reliability problem. The authors use a
penalty guided algorithm which searches over
feasible and infeasible regions to identify a final,
feasible optimal, or near optimal, solution. The
penalty function is adaptive and responds to the
search history. The GA performs very well on two
types of problems: redundancy allocation as
originally proposed by Fyffe, et al., and randomly
generated problems with more complex
configurations. For a fixed design configuration and
known incremental decreases in component failure
rates and their associated costs, Painton and
Campbell [19] also used a GA based algorithm to
find a maximum reliability solution to satisfy
specific cost constraints. They formulate a flexible
algorithm to optimize the 5th percentile of the mean
time-between-failure distribution. In this paper ant
colony optimization will be modified and adapted,
which will consider the measure of criticality will
gives a guidance to the ants for its nest and ranking
of critical components will be taken into
consideration to choose the most reliable
components which then will be improved till reach
the optimal system‟s components reliability value.
2.2 Ant colony optimization approach
Ant colony optimization (ACO) algorithm [20,
21], which imitate foraging behavior of real life
ants, is a cooperative population-based search
algorithm. While traveling, Ants deposit an amount
of pheromone (a chemical substance). When other
ants find pheromone trails, they decide to follow the
trail with more pheromone, and while following a
specific trail, their own pheromone reinforces the
followed trail. Therefore, the continuous deposit of
pheromone on a trail shall maximize the probability
of selecting that trail by next ants. Moreover, ants
shall use short paths to food source shall return to
nest sooner and therefore, quickly mark their paths
twice, before other ants return. As more ants
complete shorter paths, pheromone accumulates
635
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

faster on shorter paths and longer paths are less be additive in term of cost at constitute
reinforced. Pheromone evaporation is a process of components. See Fig. (1).
decreasing the intensities of pheromone trails over Rs
time. This process is used to avoid locally
convergence (old pheromone strong influence is
avoided to prevent premature solution stagnation), 1
to explore more search space and to decrease the
probability of using longer paths. Because ACO has Pi min
been proposed to solve many optimization problems
[22],[23], our proposed idea is also to adapt this Cost
algorithm to optimize system reliability and Ci Ct
specially complex system
Figure 1: cost-reliability curve
3 METHODOLOGY As show in Fig 1. and by equaling the slopes of two
triangles we can derive equation number (1) as
3.1 Problem definition following:
3.1 .1 Notation
In this section, we define all parameters used in p 1 - p(i)min p 2 - p(i)min
Cc   Ct   Ct  ...n . (1)
our model. 1 - p(i)min 1 - p(i)min
Rs : Reliability of system
Pi : Reliability of components i.
qi : probability of failure of components (i). 3: In [9] calculation of ICRi and ISTi derivation
Qn : Probability of failure to system equation s (2) and (3) for each components from its
n : Total number of components. structural measure, which given by,
ICRi : Index of criticality measure.
ICRp : index of criticality for path to destination
ISTi : Index of structure measure. (2)
Ct : Total cost of components. Where,
Ci : Cost of component
Cc : Cost for improvement
P(i)min: Minimum accepted reliability value (3)
ACO
:start node for ant,
: next node chosen. 4-Every ICRi must be lower than initial value ai.
τi :initial pheromone trail intensity This value is a minimum accepted level of criticality
measure to every component.
τi(old) :pheromone trail intensity of combination
before update of 5-After the complex system presented
mathematically, a set of paths will be available from
τi(new) :pheromone trail intensity of combination
after update specified source to destination. those paths will be
:problem-specific heuristic of combination ranked each one according to its components
criticalities.
η ij : relative importance of the pheromone trail
intensity
3.2 Formulation of the problem:
: relative importance of the problem-
The objective function in general, has the form :
specific heuristic for global solution
:index for component choices from set AC
Maximize, Rs= f (P1,P2,P3,....Pn).
trail persistence for local solution
subject to the following constrains,
:number of best solutions chosen for offline 1. ICRi : i =1,2,…n
pheromone update index 2. To ensure that the total cost of components not
3.1.2 Assumption more than proposed cost value the following
In this section, we present the assumptions equation number (4) can be used:
under which formulation of our model is presented.
1: There are many different methods used to derive
the expression of total reliability of complex system, :Pi(min) > 0 (4)
which are derived in a certain system topology, we
state our system expressions according to the
methods of papers [3-5]. Note that this set of constrains permits only
2: We used a cost-reliability curve [7] to derive an positive components cost.
equation to express each cost components according
to its reliability and then the total system cost will

UbiCC Journal – Volume 4 No. 3 636

 Special Issue on ICIT 2009 Conference - Bioinformatics and Image

4 MODEL CONSTRUCTION

The algorithm uses an ACO technique with the (6)

criticality approach to ensure global converges from
any starting point. The algorithm is iterative. At
The update equation will become as follows:
each iteration, the set of ants are identified using
some indicator matrices. Below are the main steps
of our proposed model . As we see in the Fig. 2 (7)
which illustrating a set of steps illustrated below: 5. A new reliabilities will be generated.
1. Ant colony parameters are initialized 6. Till reach best solution and all ant moved to
2. The criticality of components will be achieve maximum reliability of the system with
calculated according to derived reliability equation, minimum cost.
then will be ranked according to its values
3. Using equation number(5) Ant equation: 5 EXPERIMINTAL RESULTS

In the following examples, we use a bench

(5) mark systems configurations like a Bridge, and
Delta .
5.1 Bridge problem:

The probability to choose the next node will be 2 3

estimated after a random number generated. and
until the destination node. The selected nodes will S 5 D
be chosen .According to the criticality components
through this path. 1 4

Input system reliability Figure 3: Bridge system

equation
To find the polynomial for a complex system we
Randomly initialize Pi and minimum values
must know that it always given at a certain time to
and generate random number choose n Ants be transmitted from source (s) to destination (D),
see Fig. 3.
The objective function to be maximized has the
Evaluate ICRi for components & rank form: Rs=
rankcomponents
1- (q1+q4.q5.p1+q3.q4.p1.p5+q2.q4.p1.p5.p3)
Calculate

Subject to:
3

Generate new Pi If random No. < Ant Move

1.  Ci * (pi)  45
i 1

2. The ICRi constraint.

Do same until the destination then Select path
ICRi calculated : i=1,2,…5..
Update pheromone :
- We use the values in the Fig. 3 as initial values for
components‟ reliabilities to improve the system:

P(1)min=0.9, P (2)min=0.9,
P(3)min=0.8, P (4)min=0.7, p(5)min=0.8.
ants reached
NO destination?
3. We choose the cost-reliability curve to
Yes
permit distribution of cost depending on ranking of
Get optimized values
components according to there criticality. The
model was built in such a way that reduce the fail of
Figure 2: Flow diagram adapted ant system the most critical components, this is done by
increasing the reliability of the most critical
4. Eq. (6): update the pheromone according to the
components, which tend to maximizes the over all
criticality measure. Which can be calculate
reliability what is our goal. We summarized our
product of components criticalities‟ value
results in the following Table (1) and Table

UbiCC Journal – Volume 4 No. 3 637

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

(2).With initial values of ant colony algorithm as in

Table ( 3). 5.2 Delta Problem:

Table 1: Reliabilities of the Bridge system. S T

1
Reliab- New ICRi rank
ities values
p1 0.9998 1

3
p2 0.9 3
p3 0.8 4
p4 0.9998 2 Figure 4: Delta system
p5 0.8 5
Rs 0.9999 Using the same procedures as in bridge problem
we obtain the following optimization problem for
Table 2: Costs of the Bridge system . delta system given in Fig 4.
cost Value in units
Max .Rs= P1+ P1.P2 - P1.P2.P3
C1 9.9988
C2 8.8888 Subject to
C3 7.7777 1. ICRi calculated for i=1, 2,3.
C4 9.9978 3

C5 7.7777 2.  Ci * (Pi)  4.5

i 1
Ct 44.441
p(1)min=0.7 i=1,2,3.
Table 3: ACO initial values The following two Tables (4) and (5) summarized
the results.
2
3 Table 4: Reliabilities of the Delta s ystem.
0.2 Computed ICRi Rank
1 value
Q 10 P1 0.9999 1
Ants 10 P2 0.7 2
P3 0.7 3
5.1.1Comments on results
Rs 0.9999
As cleared in Tables 2 and 3 results indicate
that according the criticality of components, the
improvement will be occurred as the more critical Table 5: Costs of the Delta system.
component the more chance to be improved which Cost values
will highly effect to the system reliability C1 0.9998
improvement with minimal cost too, this is better C2 0.4
than to increase reliability components randomly.
C3 0.4
Now it is clear also the best path from S to D is to
follow component 1 and component 4 . if we have Ct 1.799
more available cost it will increase the other
component reliability according to it‟s criticality Beside comments noted in bridge system, delta
ranking. Finally if all components have the same system have two paths from S to T as shown in the
initial reliability values the path through Fig 4. The results shows that it is preferred to
components 1 and 4 have the same chance for path increase the component one rather than others this
through component 2 1nd 3, and according for two reasons, it have most critical value and
algorithm which depend on the topological pheromone value biased toward the path with lower
reliability it will goes to improve the higher critical number of components (Path1=P1) according to
component according to it‟s position in the system. the equation :

UbiCC Journal – Volume 4 No. 3 638

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

As we see from results in Tables 6 and 7

components 6 and 7 have the most reliability values
according to it‟s criticality and the path chosen
through components 6 and 7, and to achieve
minimal cost the system take only 4.22 which
5.4. Mesh Problem:
achieve our objectives
2
1 3
5.5 Important Comments
S T To study the effect of modifying of ant
4

5
parameters such as initial pheromone in a delta case
and biased to component 2 the results will become
6 7 as shown in Table 8. The reliably for components
was P1=0.2, P2=0.3 and P3=0.3 and values of
Figure 5: Mesh system =10 , =2 and =10
This system have more components and large and
The objective Function for the mesh system is:
Table 8: Effects of Ant colony parameters
Max. Rs=(p6*p7)+(p1*p2*p3* Cost values
(1-p6))+(p1*p2*p3*p6*(1-p7))+(p1*p4*p7* C1 0.7777
(1-p2)*(1-p6))+(p1*p4*p7*p2*(1-p6)*(1-
C2 0.9997
p3))+(p3*p5*p6*(1-p7)*(1p1))+(p3*p5*p6*p1*(1-
p7)*(1-p2))(p1*p2*p5*p7*(1-p3)*(1-p4)*(1-p6))- C3 0.999
(p2*p3*p4*p6*(1-p1)*(1-p5)*(1- Ct 14.777
p7))+(p1*p3*p4*p5*(1-p2)*(1-p6)*(1-p7)); Computed ICRi Rank
value
Subject to, P1 0.3 1
1. ICRi calculated for i=1,2,..n...
7 P2 0.9999 2
 Ci * (Pi)  6.6 P3 0..9999 3
i 1
Rs 0.9999
P(i)min=0.5 i=1,2,3..
It is clear that the solution biased to the components
2 and 3 path rather than component one, because of
Table 6: Reliabilities of the Mesh system
there initial pheromone values.
Reliabiliti New ICRi rank
es values
P1 0.5 5 6 CONCLUSION
P2 0.5 4
P3 0.5 3 We propose a new effective algorithm for
P4 0.5 7 general reliability optimization problem. Using ant
colony. The ant colony algorithm is a promising
P5 0.5 6
heuristic method for solving complex combinatorial
P6 0.9999 1 problems.
P7 0.9999 2 To solve complex system design problem:
Rs 0.9997 1. We must formulate a system, that is correctly
Table 7: Costs of the Mesh system representing the real system with all paths from
cost Value in units source to destination by choose an efficient
reliability estimation method.
C1 0.4444
2. To the best of maximization of total reliability
C2 0.4444 and minimization of the total cost of a system take
C3 0.4444 in to consideration the components according to its
C4 0.4444 criticality, then arrange the most critical components
gradually.
C5 0.4444 3. Index of criticality achieve maximum system
C6 0.9998 reliability with minimum cost according to
reliability of system topology
C7 0.9997
4. resolve model without index of criticality
Ct 4.22 maximum reliability and minimum cost but this
method ignore the topology of the system.

UbiCC Journal – Volume 4 No. 3 639

 Special Issue on ICIT 2009 Conference - Bioinformatics and Image
5. The ant colony algorithm improved by the
previous experience which was given by the index
of criticality which gives to ant an experience to
deposit of pheromone on a trail which will
maximize the probability of selecting that trail by
next ants. Moreover, ants shall use more reliable
paths. Our numerical experiences show that our
approach is promising especially for complex
systems.
7 REFERENCES
[1] A. Lisnianski,. H. Ben-Haim, and D.
Elmakis: “Multistate System Reliability
optimization: an Application”, Levitin,
Gregory book , USA, pp.1-20. ISBN
9812383069. (2004)
[2] S. Krishnamurthy, AP. Mathur.: On the
estimation of reliability of a software
system using reliabilities of its components
.In: Proceedings of the ninth international
symposium on software reliability
engineering(ISSRE„97).Albuquerque;.p.146.
(1997)
[3] T. Coyle, RG. Arno, PS.: Hale.
Application of the minimal cut set
reliability analysis methodology to the gold
book standard network. In the commercial
and power systems technical conference;.
p. 82–93. industrial (2002)
[4] K. Fant, Brandt S. : Null convention logic,
a complete and consistent logic for
asynchronous digital circuit synthesis. In:
the international conference on application
specific systems, architectures, and
processors (ASAP ‟96); p. 261–73. (1996).
[5] C. Gopal H, Nader A.: A new approach to
system reliability. IEEE Trans
Reliab;50(1):75–84. (2001).
[6] Y. Chen, Z. hongshi:" : Bounds on the
Reliability of Systems With Unreliable
Nodes & Components". IEEE, Trans. on
reliability, vol.53, No. 2, June.(2004).
[7] B. A. Ayyoub.:” An application of reliability
engineering in computer networks
communication” AAST and MT Thesis,
p.p17Sep.(1999).
[8] S. Magdy, R.d Schinzinger: "On Measures
of computer systems Reliability and Critical
Components", IEEE, Trans. on Reliability
(1988).
[9] B. A. Ayyoub. M. Baith Mohamed,
UbiCC Journal – Volume 4 No. 3
ElAlem: " An Application of Reliability
Engineering in Complex Computer System
and Its Solution Using Trust Region
Method", WSES , software and hardware
Engineering for 21st century book,
pp261,(1999).
[10] ATillman,C.Hwang,,K.Way : “Optimization
Techniques for System Reliability with
Redundancy,A Review”, IEEE Transactions
on Reliability, vol. R-26, no. 3, , pp. 148-
155. August (1977).
[11] E. David. Fyffe, W. William. K. L Hines,
Nam: “System Reliability Allocation And
a Computational Algorithm”, IEEE
Transactions on Reliability, vol. R-17, no. 2,
, pp. 64-69. June (1968).
[12] Y. Nakagawa, S. Miyazaki: “Surrogate
Constraints Algorithm for Reliability
Optimization Problems with Two
Constraints”, IEEE Transactions on
Reliability, vol. R-30, no. 2, , pp. 175-180.
June (1981).
[13] K. Behari Misra, U. Sharma: “An Efficient
Algorithm to Solve Integer-Programming
Problems Arising in System-Reliability
Design ”,IEEE Transactions on Reliability,
vol. 40, no. 1, , pp. 81 91. April (1991).
[14] C. Lai Hwang, A. Frank Tillman, W. Kuo, :
“Reliability Optimization by Generalized
Lagrangian - Function and Reduced-
Gradient Methods”, IEEE Transactions on
Reliability, vol. R-28, no. 4, pp. 316-319.
October (1979).
[15] A. Frank Tillman, C.Hwang, W Kuo, :
“Determining Component Reliability and
Redundancy for Optimum System
Reliability”, IEEE Transactions on
Reliability, vol. R-26, no. 3, pp. 162- 165.
August (1977).
[16] D. Coit, Alice E.Smith, “Reliability
Optimization of Series-Parallel Systems
Using a Genetic Algorithm”, IEEE
Transactions on Reliability, vol. 45, no. 2, ,
pp. 254-260 June,(1996 ).
[17] W. David. Coit, Alice E. Smith: “Penalty
Guided Genetic Search for Reliability
Design Optimization”, Computers and
Industrial Engineering, vol. 30, no. 4, pp.
95-904. (1996).
[18] W. David Coit, E. Alice Smith, M. David
640
 Special Issue on ICIT 2009 Conference - Bioinformatics and Image

Tate,: “Adaptive Penalty Methods for

Genetic Optimization of Constrained
Combinatorial Problems”, INFORMS
Journal on Computing, vol. 8, no. 2, Spring,
pp. 173-182. (1996).

[19] L. Painton, C. James: “Genetic Algorithms

in Optimization of System Reliability”,
IEEE Transactions on Reliability, vol. 44,
no. 2, , pp. 172-178. June (1995)

[20] N. Demirel,., Toksar, M.: Optimization of

the quadratic assignment problem using an
ant colony algorithm, Applied Mathematics
and Computation, Vol. 183, optimization
,Applied Mathematics and Computation,
Vol. 191, pp. 42--56 (2007).

[21] Y. Feng, L. Yu,G.Zhang,: Ant colony pattern

search algorithms for unconstrained and
bound constrained optimization ,Applied
Mathematics and Computation, Vol. 191,
pp. 42--56 (2007).

[22] M. Dorigo, L. M. Gambardella: “Ant

Colony System: A Cooperative Learning
Approach to the Travelling Salesman
Problem”, IEEE Transactions on
Evolutionary Computation, vol. 1, no. 1, ,
pp. 53-66. April (1997).

[23] B. Bullnheimer, F. Richard, H. Christine

Strauss, “Applying the Ant System to the
Vehicle Routing Problem”, 2nd Meta-
heuristics International Conference (MIC-
97), Sophia-Ant polis, France, pp. 21-24.
July, (1997).

UbiCC Journal – Volume 4 No. 3 641

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

A COMPREHENSIVE QUALITY EVALUATION

SYSTEM FOR PACS

Dinu Dragan, Dragan Ivetic

Departmant for Computing and Automatics, Republic of Serbia
dinud@uns.as.rs, ivetic@uns.ac.rs

ABSTRACT
An imposing number of lossy compression techniques used in medicine, represents
a challenge for the developers of a Picture Archiving and Communication System
(PACS). How to choose an appropriate lossy medical image compression
technique for PACS? The question is not anymore whether to compress medical
images in lossless or lossy way, but rather which type of lossy compression to use.
The number of quality evaluations and criteria used for evaluation of a lossy
compression technique is enormous. The mainstream quality evaluations and
criteria can be broadly divided in two categories: objective and subjective. They
evaluate the presentation (display) quality of a lossy compressed medical image.
Also, there are few quality evaluations which measure technical characteristics of a
lossy compression technique. In our opinion, technical evaluations represent an
independent and invaluable category of quality evaluations. The conclusion is that
quality evaluations from each category measure only one quality aspect of a
medical image compression technique. Therefore, it is necessary to apply a
representative(s) of each group to acquire the complete evaluation of lossy medical
image compression technique for a PACS. Furthermore, a correlation function
between the quality evaluation categories would simplify the overall evaluation of
compression techniques. This would enable the use of medical images of highest
quality while engaging the optimal processing, storage, and presentation resources.
The paper represents a preliminary work, an introduction to future research and
work aiming at developing a comprehensive quality evaluation system.

Keywords: medical image quality metrics, medical image compression, PACS

1 INTRODUCTION The second group of compression techniques

Picture Archiving and Communication System
(PACS) represents an integral part of modern
hospitals. It enables communication, storage,
processing, and presentation of digital medical
images and corresponding data [1]. Digital medical
images tend to occupy enormous amount of storage
space [2, 3]. The complete annual volume of medical
images in a modern hospital easily reaches hundreds
of petabytes and is still on the rise [4]. The increased
demand for digital medical images introduced still
image compression for medical imaging [5], which
relaxes storage and network requirements of a PACS,
and reduces the overall cost of the system [3].
In general, all compressed medical images can
be placed in two groups: lossless and lossy. The first
group is more appealing to physicians, because
decompression restores the image completely,
without data loss. It achieves modest results and
maximum compression ratio of 3:1 [6, 7, 8]. Several
studies [9, 10] showed that this is not suitable for
PACS, and that at least 10:1 compression ratio has to
be achieved.
achieves greater compression ratios, but with data
distortion in restored image [6, 7, 8]. Lossy
compression provoked serious doubts and opposition
from medical staff. The opposition rose from the fact
that the loss of data can influence medical image
interpretation and can lead to serious errors in
treatment of a patient. Therefore, the main research
area for lossy compression of medical images is
finding of the greatest compression ratio that still
maintains diagnostically important information. The
degree of lossy compression of medical images
which maintains no visual distortion under normal
medical viewing conditions is called “visually
lossless” compression [10]. Several studies
[8, 11, 12] and standards [13] proved clinical
acceptability to use lossy compression of medical
images as long as the modality of the image, the
nature of the imaged pathology, and image anatomy
are taken into account during lossy compression. The
medical organization involved has to approve and
adopt a lossy compression of medical images applied
in PACS. Therefore, it is necessary to provide a

UbiCC Journal – Volume 4 No. 3 642

Special Issue on ICIT 2009 Conference - Bioinformatics and Image
quality evaluation of different compression
techniques from PACS point of view.
During our work on a PACS for a lung hospital,
we tried to adopt image compression for medical
images which achieves highest compression ratio
with minimal distortion within decompressed image.
Also, we needed image compression suitable for
telemedicine purposes. We consulted the technical
studies in search for quality evaluation of image
compression technique. The sheer amount of studies
is overwhelming [14, 15]. There is no unique quality
evaluation which is suitable for various compression
techniques and different applications of image
compression [16, 17]. In most cases the studies are
focused only on presentation (display) quality of the
lossy compressed medical image. Technical features
of compression technique are usually ignored.
This paper represents a preliminary research. Its
purpose is to identify all the elements needed to
evaluate the quality of a compression technique for
PACS. We identified three categories of quality
evaluations and criteria: presentation-objective,
presentation-subjective, and technical-objective.
Overview of technical studies led us to conclusion
that quality evaluations from each category measure
only one quality aspect of an image compression
technique. To perform the complete evaluation of
medical image compression technique for PACS, it
is necessary to apply a representative of each
category. A correlation function between the
representatives of each category would simplify the
overall evaluation of compression techniques. A 3D
evaluation space introduced by the paper is a 3D
space defined by this correlation function and quality
evaluations used. Our goal is to develop an
evaluation tool based on the 3D evaluation space
which is expected for 2011. All the elements of the
quality evaluation system are identified in the paper.
The organization of the paper is as follows:
section 2 gives the short overview of the lossy
compression techniques used in medical domain;
section 3 describes the quality evaluations used to
measure the quality of compression techniques; 3D
evaluation space is discussed in section 4; section 5
concludes the paper.
2 LOSSY COMPRESSION OF MEDICAL
IMAGES
Over the past decades an imposing number of
lossy compression techniques have been tested and
used in medical domain. Industry approved standards
have been used as often as the proprietary
compressions. On the part of the image affected, they
can be categorized in two groups:
1. medical image regions of interest (ROI) are
compressed losslessly while the rest of the image
background is compressed lossy,
UbiCC Journal – Volume 4 No. 3
2. the entire medical image is compressed lossy
targeting the “visually lossless” threshold.
The first group offers selective lossy compression of
medical images. Parts of the image containing
diagnostically crucial information (ROI) are
compressed in a lossless way, whereas the rest of the
image containing unimportant data is compressed
lossy. This approach enables considerable higher
compression ratio than ordinary lossy compression
[18, 19]. Larger regions of the medical image contain
unimportant data which can be compressed at higher
rates [19]. Downfall of this approach is
computational complexity (an element of technical-
objective evaluation). Each ROI has to be marked
before compression. Even for images of the same
modality, ROIs are rarely in the same place. ROIs
are identified either manually by qualified medical
specialist or automated based on a region-detection
algorithm [20]. The goal is to find a perfect
combination of automated ROI detection algorithms
and selective compression technique.
Over the years various solutions for ROI
compression of medical images emerged which
differ in image modalities used, ROI definitions,
coding shames and compression goals [20]. Some of
them are: a ROI-based compression technique with
two multi-resolution coding schemes reported by
Strom [19], a block based JPEG ROI compression
and a importance schema coding based on wavelets
reported by Bruckmann [18], a motion compensated
ROI coding for colon CT images reported by
Bokturk [21], a region based discrete wavelet
transform reported by Penedo [22], a JPEG2000 ROI
coding reported by Anastassopoulos [23].
The second group of lossy compression
techniques applies lossy compression over entire
medical image. Considerable efforts have been made
in finding and applying the visual lossless threshold.
Over the years various solutions emerged which
differ in goals imposed on a compression technique
(for particular medical modality or for a group of
modalities), and in compression techniques used
(industry standards or proprietary compression
techniques).
Some of the solutions presented over the years
are: a compression using predictive pruned tree-
structured vector quantization reported by Cosman
[17], a wavelet coder based on Set Partitioning in
Hierarchical Trees (SPIHT) reported by Lu [24], a
wavelet coder exploiting Human Visual System
reported by Kai [25], a JPEG coder and wavelet-
based trelliscoded quantization (WTCQ) reported by
Slone [10], a JPEG2000 coder reported by Bilgin
[26].
Although the substantial effort has been made to
develop a selective lossy compression of medical
images, the industry standards that apply lossy
compression on the entire medical image are
commonly used in PACS.
643
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
3 QUALITY EVALUATIONS
The significant effort has been made to solve the
problem of measuring digital image quality with
limited amount of success [13, 14]. Various studies
tried to develop new metrics or to adopt existing
ones for medical imaging [5, 6, 7, 8, 17]. The quality
evaluations used can be broadly categorized as
[5, 17]:
• objective quality evaluations – based on a
mathematical or a statistical model, which is easy
to compute and rate,
• subjective quality evaluations – based on a
subjective observer evaluation of restored image,
or questionnaires with numerical ratings.
These categories can be further sub-categorized, but
this falls out of the scope of the paper [5, 17].
The quality evaluations proposed measure
presentation (display) quality of the lossy
compressed medical image. Therefore, they can be
categorized as presentation-objective and
presentation-subjective quality evaluation. Although,
these quality evaluations have been devised for
image quality measurement, they can be also used
for evaluation of lossy compression techniques. The
quality of the reconstructed image should not be the
only criteria for adoption of a compression technique
for PACS. The quality evaluation of medical image
compressions for PACS is inseparable from technical
aspects of the system. The lossy compression can
uphold remarkable presentational quality (objective
and subjective) of medical images but with high
technical demands. In some cases these technical
demands are not achievable and in most cases they
are too expensive. In many countries this will impose
too high price for PACS. Evaluations measuring
image compression quality from technical point of
view can be categorized as technical-objective
quality evaluations.
3.1 Presentation-objective evaluations
Presentation-objective evaluations represent the
most desirable way to measure image quality. They
are based on a mathematical model, and are usually
easy to compute. Their main advantage is objectivity
[27]. The numerical distortion evaluations like mean
squared error (MSE), Eq. (1), signal-to-noise-ratio
(SNR), Eq. (2), or peak-signal-to-noise-ratio (PSNR),
Eq. (3), are commonly used [6].
m n
∑∑ [ f (i, j ) − f ′(i, j )]2 / m ⋅ n (1)
i =1 j =1
n m
σ x2 1 m n ∑∑ f (i, j )
∑∑ ( f (i, j) −
i =1 j =1
; σ x2 = ) (2)
MSE m⋅n i =1 j =1 m⋅n
UbiCC Journal – Volume 4 No. 3
(2 b ) 2
(3)
MSE
These measures fail to measure local
degradations and do not provide precise descriptions
of image degradations [5, 27]. Still, many studies use
this quality evaluations to rate their implementations
of lossy medical image compression techniques.
Quality of the lossy compressions studied in [9,
24, 25, 26, 28] was measured by these numerical
distortion evaluations. For example, Chen [9] used
PSNR to evaluate propriety DCT based SPIHT
compression, original SPIHT and JPEG2000. The
DCT based compression achieved highest PSNR
values for the tested medical images, which indicated
that it is more suitable for medical imaging then the
other two compression techniques.
Beside scalar numerical evaluations, graphical
evaluations such as Hosaka plots and Eskicioglu
charts, and evaluations based on HVS model have
been used [14, 15, 29]. Their applicability in medical
domain has been reported in [6, 27].
Also, a hybrid presentation-objective metrics
have been studied for medical domain.
Przelaskowski [27] proposed a vector quality
measure reflecting diagnostic accuracy, Eq. (4).
6
HVM = ∑α iVi (4)
i =1
The values Vi represents one presentation-
objective measure. The vector measure was designed
to include the formation of a diagnostic quality
pattern based on the subjective ratings of local image
features. This quality measure represents a way of
combining presentation-objective and presentation-
subjective evaluations. Evaluation of lossy
JPEG2000 compressed medical images found that
compression ration of 20:1 is diagnostically
acceptable.
3.2 Presentation-subjective evaluations
Presentation-subjective evaluations have been
used to evaluate lossy compressed medical images
more often than presentation-objective [30].
Presentation-subjective evaluations are based on
observer’s subjective perception of reconstructed
image quality [5]. The subjective quality of a
reconstructed medical image can be rated in many
ways [5]. In some studies, observer is presented with
several reconstructed versions of the same image.
The observer has to guess the image compression
level and to order the sample images in order from
the least compressed to the most compressed [5, 31].
If the difference between original image and
reconstructed image at some level of compression is
not distinguishable, then that level of compression is
644
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
diagnostically acceptable [32]. Other studies used
qualified observers to interpret reconstructed medical
images compressed at various levels. The
compression levels on which results were the same
as for the original image have been rated as
acceptable [5]. Also, some studies used qualified
technicians to define a “just noticeable” difference
used to select the point at which compression level is
not diagnostically usable. The observers have been
presented with series of images, each compressed at
higher level. They simple had to define the point at
which changes became obvious. The studies were
based on presumption that one can perceive
“changes” in the image long before an image is
degraded enough to lose its diagnostic value [5].
When subjectively evaluating medical images, it
is not sufficient to say that image looks good. It
should be proved that image did not loose the
essential information and that it has at least the same
diagnostic values as the original medical image [6].
Therefore, beside pure subjective evaluations, semi-
subjective evaluations of a reconstructed medical
image which measure diagnostic accuracy have been
used. Observers often rated the presented images on
a scale of 1 to 5 [10, 17]. Collected data have been
further statistically analyzed highlighting averages
and other trends in collected data. Quality of
reconstructed medical images is most often measured
by semi-subjective evaluation based on Receiver
Operating Characteristic (ROC) analysis, which has
its origins in theory of signal detection [6, 7, 27, 33].
A filtered version of the signal plus Gaussian noise is
sampled and compared to a threshold. If it exceeds
the threshold then the signal is declared to be there.
As the threshold varies, so does the probability to
erroneously declare the signal present or absent. The
ROC analyses are based on ROC curves (see Fig. 1),
which are a simple complete empirical description of
this decision threshold effect, indicating all possible
combinations of the relative frequencies of the
various kinds of correct and incorrect decisions [6].
The plot is a summary of the trade off between true
positive rate (sensitivity) and false positive rate (the
complement of specificity). The area under the curve
can be used to summarize overall quality or the
efficiency of the detection process [6, 7].
The ROC curves are not applied directly to
medical imaging. The decision threshold is based on
diagnostic accuracy and physician’s judgment.
Reconstructed medical images, which either
possessed or not an abnormality, were presented to
qualified specialist. Observers had to provide a
binary decision if abnormality is present or not,
along with a quantitative value for their degree of
certainty (a number from 1 to 5). A subjective
confidence rating of diagnoses is then used as if it
were a threshold to adjust for detection accuracy [6].
A resulting diagnostic accuracy is compared with
original image and used to define an acceptable
UbiCC Journal – Volume 4 No. 3
Figure 1. Example of ROC curve
compression level. The results for the different
compressions or compression levels could be used
for quality evaluation of compression techniques.
The success of ROC analysis depends on the
number of test images and observers included in the
study. Therefore, the ROC analyses tend to be
expensive and time consuming. For example, a
typical ROC study would require over 300 images to
obtain a reasonable statistical confidence level, five
or more radiologists to view these images, and a full-
time statistician to coordinate and analyze the data
[6].
Various results were obtained for image quality
by presentation-subjective evaluations. Smith [31]
reported that lossy JPEG compression of chest
radiographs can be set at levels as high as 30:1.
Perlmutter [7] reported that lossy wavelet
compression of digital mammograms can achieve
compression ratio of 80:1 with no influence on
diagnostic accuracy. Przelaskowski [33] reported
even better results for JPEG2000 compression of
digital mammograms of 140:1 compression ratio.
The study [33] was based ROC analysis.
3.3 Technical-objective evaluations
PACS as a part of a modern hospital becomes a
highly interactive environment that is forming a
ubiquitous computing environment for medical work
[34, 35]. It is not limited to only one medical facility
or to a group of closely spaced facilities. PACS often
spreads over vast areas including not only the most
prominent and richest of medical facilities, but also
the facilities in rural and less developed areas [3].
The best, and expensive devices, are not available for
this facilities. Also, it is unreal to expect a 100 Mbit
connection (minimum for efficient PACS
communication [3]) to all sides of such a sparse
system.
As a part of the mobile health, devices with less
storage, processing, and display capabilities are also
a common part of a PACS. These devices can
645
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
process only the limited number of medical images
and images of limited size [2]. Also, these devices
usually use wireless networks which have
capabilities far beneath connected ones [2].
Therefore, to view a medical image on these devices
it is necessary to have images scaled for the display
size of the mobile device. This could have a negative
impact on PACS storage space [36], but it is
minimized when the scaled medical images are
acquired from the same image codestream as the
original sized image i.e. when streaming of medical
images is used [37, 38]. Image streaming is a process
of gradual buildup of an image by resolution or by
pixel accuracy [28]. It enables extraction of a lower-
resolution image from the codestream.
The architecture of modern PACS is described
by Fig. 2. Beside high class hospitals, the system
contains less equipped hospital in rural areas and
medical mobile devices.
These are all reasons for adopting lossy
compression of medical images for a PACS, but they
are also restrictions which one developing a PACS
system should consider. They represent technical-
objective criteria for evaluating a medical image
compression technique for a PACS. The parameters
of the criteria are overall cost of the system
equipment, storage and network requirements, the
cost for implementation of the compression
technique, compression/decompression speed,
streaming possibility of the compression technique,
image modalities suitable for the compression, and
compression ratio achieved under certain quality
assumption. Technical studies comparing different
compression techniques evaluated several things
[39, 40, 41, 42]:
• Compression speed [39, 40, 41, 42]. The studies
measured the time elapsed while the sample
image was compressed to target compression
ratio, and the time elapsed during decompression.
Figure 2. Architecture of a Modern PACS
UbiCC Journal – Volume 4 No. 3
This time has impact on overall performance of
the system because it can cause data transmission
delay. Better PACS performance is achieved if
decompression time is minimized, because
decompression occurs more often than
compression. Therefore, the retrieval oriented
compression techniques are common for medical
imaging.
• Memory and processor power used [40]. The
study measured the amount of memory and
processor power used during compression and
decompression process. The values measured
inform about the overall complexity of
compression technique which influence overall
cost of the system. High requirements influence
higher cost.
• Compression ratio [40]. The influence of
compression technique on storage requirements is
expressed as achievable compression ratio. It is
measured in respect to image presentation quality,
like numerical distortion measures, section 3.A.
Storage requirements influence the overall cost of
PACS.
• Functionalities of a compression technique [41].
Most applications require other features beside
quality and coding efficiency of the compression
technique. The technical-objective quality
evaluations in consulted studies did not evaluate
functionalities of a compression technique
numerically. Rather they use some method of
description. Santa-Cruz [41] provided a
functionality matrix that indicated the supported
features in each compression technique and an
appreciation of how well they are fulfilled, Table
1. They compared JPEG2000, JPEG-LS, JPEG,
MEPG-4 VTC, and PNG compression techniques.
A set of features (functionality) is included in
Table 1. A “+” mark indicates whether the
functionality is supported. The more “+” marks,
646
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
Table 1: Functionality matrix – various functionalities of different compression techniques are compared [41].
JPEG2000
lossless compression performance +++
lossy compression performance +++++
progressive bitstreams +++++
Region of Interest (ROI) coding +++
arbitrary shaped object - -
random access ++
low complexity ++
error resilience +++
non-iterative rate control +++
generality +++
the more efficiently or better is the functionality
supported by compression technique.
• Error resilience [40, 41]. It is important to
measure the error resilience of compressed
images sent over network transmission channels.
This is tested by transmitting the compressed data
over simulated noise channel.
The quality evaluation of medical image
streaming has not been studied in the consulted
literature. Streaming of medical images is important
issue for PACS trying to achieve mobile health (and
ubiquitous healthcare, also) and it should be
considered during quality evaluation. Because it is
supported by limited number of compression
techniques, quality evaluation should indicate
whether the streaming is supported or not. If
compression techniques support image streaming,
the quality of extracted low-resolution images should
be evaluated.
An important issue considering technical aspects
of medical image compression techniques is weather
to use industry wide standards or to develop a
proprietary compression technique [43]. The second
approach could lead to more efficient compression
techniques, but in long term, it would show more
costly. It could compromise PACS communication
with equipment and networks not supporting the
proprietary compression technique [43]. The long
term archives of medical images could be
compromised if the system transgresses to another
compression technique. The use of industry approved
standards can reduce the cost and risk of using
compression.
4 3D QUALITY EVALUATION SPACE
Quality evaluations of lossy compression
techniques differ in many ways. They differ in way
whether they consider the application for which the
compressed image has been used. Some quality
evaluations measure only the performance of the
compression technique while other measure only the
presentation quality of the restored image. Overall,
there is no quality evaluation which measures all the
UbiCC Journal – Volume 4 No. 3
JPEG-LS JPEG MPEG-4 VTC PNG
++++ + - +++
+ +++ ++++ -
- ++ +++ +
- - + -
- ++ -
- - - -
+++++ +++++ + +++
++ ++ +++ +
- - + -
+++ ++ ++ +++
elements of a medical image compression technique.
When evaluating medical image compression
techniques it is important to measure the quality of
the restored images. Presentation evaluations
(objective and subjective) measure the presentation
(perceptual) quality of a restored image. The values
obtained are used to compare the quality of the
compressed images and to observe which
compression technique achieves higher compression
ratio under the same quality assumption. It is easier
to compute the presentation-objective measure which
is usually presented as a scalar or a vector. These
values are comparable and it is easy to obtain which
compression technique is better – the one heaving
bigger value. They fail to measure precise (local)
characteristics of the restored image i.e. they do not
consider the medical application of the compression
technique. On the contrary, the presentation-
subjective evaluations consider the medical
application of the compression technique, but they
are harder to obtain and cost more than presentation-
objective evaluations. The presentation-subjective
evaluations are harder to interpret and compare, and
they are dependable of observer’s knowledge,
experience and perception. The advantage of
presentation-subjective evaluations is that they are
recommended by official medical organizations
which consider compression of medical images (like
CAR).
The presentation quality evaluations fail to
measure technical aspects of a compression
technique. Beside restored image quality, it is
necessary to obtain technical information about
compression technique, like: efficiency (compression
speed, achievable ratio, transmission possibilities),
error resilience, features (image streaming), and
implementation cost and maintenance. The technical-
objective quality evaluations measure technical
elements of a compression technique. There are
several important technical-objective evaluations
measuring different features of a compression
technique. The issue is how to correlate them to one
value.
647
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
Figure 3. The 3D quality evaluation space for compression of medical images
To obtain the complete quality evaluation of a
medical image compression technique, it is necessary
to use all three previously described quality
evaluations: presentation-objective, presentation-
subjective, and technical-objective. Only then will
the observers adopting a medical image compression
for PACS have a complete insight of a given
compression technique. This will present the medical
staff with highest quality medical images while
engaging the optimal processing, storage, and
presentational resources.
The complete quality evaluation could be
improved if there is a correlation function between
the quality evaluations used, such as the one
described by Eq. (5).
ev = f (a ⋅ po, b ⋅ ps, c ⋅ to ) (5)
Variables po, ps, and, to represent values
obtained by applying presentation-objective,
presentation-subjective, and technical-objective
quality evaluations. Factors a, b, and c are weighting
factors ranging from 0 to 1 used to define the
influence of a particular quality evaluation. Value of
0 cancels the influence of a particular evaluation.
Ideally, the result of the correlation function
should be a scalar which should define the quality of
a compression technique in a simple and a
comparable way. A higher value indicates a better
quality. Unfortunately, it is more realistic to expect
that the result of the correlation function would be a
vector which defines the quality of a compression
technique in a space defined by presentation-
objective, presentation-subjective, and technical-
objective evaluations, Fig. 3. Higher vector intensity
indicates a better quality.
One possible combination for 3D quality
evaluation space would be the use of PSNR [6] (as it
is the one most often used), Eq. (3), or Przelaskowski
[27] vector measure (because combines it several
objective measures), Eq. (4), for presentation-
UbiCC Journal – Volume 4 No. 3
objective evaluation, ROC analysis [6] (as it is
subjective measure used most often) for
presentation-subjective evaluation, and functionality
matrix [41] (being the most comprehensive technical
evaluation) for technical-objective evaluation. The
proposed combination is still under review.
5 CONCLUSION
This paper represents a preliminary research. We
identified three categories of quality evaluations and
criteria: presentation-objective, presentation-
subjective and technical-objective. To obtain the
complete quality evaluation of a medical image
compression technique, it is necessary to use all three
categories of quality evaluations.
The development of a comprehensive evaluation
of all the aspects of a compression technique would
ease the task of adopting a medical image
compression for a PACS. Our future research will
include devising a technical-objective correlation
function which will uniformly present the results of
technical-objective quality evaluations. The major
focus of our future research will be devising a
correlation function between all the groups of quality
evaluations. We strive to achieve quality evaluation
space like the one described by the Fig. 3 which
would represent an environment for simple and
comprehensive evaluation of medical image
compression techniques for PACS.
In the case of PACS for the lung hospital we did
not have time to wait for development of 3D quality
evaluation space. Therefore, we adopted the
compression technique that in our opinion (which
was drawn from numerous technical studies) offered
the most - JPEG2000 compression [36, 37, 38, 39, 40,
41, 42]. It would be interesting to see if this decision
correlates with the results in the 3D quality
evaluation space for compression of medical images.
648
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
ACKNOWLEDGEMENT
This research has been conducted under IT
project: “WEB portals for data analysis and
consulting”, No.13013, financed by the Government
of Republic of Serbia.
6 REFERENCES
[1] H.K. Huang: Enterprise PACS and image
distribution,” Computerized Medical Imaging
and Graphics, Vol. 27, No. 2-3 , 2003, pp. 241-
253 (2003).
[2] A.N. Skodras: The JPEG2000 Image
Compression Standard in Mobile Health, M-
Health: Emerging Mobile Health Systems,
SpringerLink, pp. 313-327 (2006).
[3] M.K. Choong, R. Logeswaran, and M. Bister:
Cost-effective handling of digital medical
images in the telemedicine environment,
International Journal of Medical Informatics,
Vol. 76, No. 9, pp. 646-654 (2007).
[4] A. N. Belbachir, and P. M. Goebel: Medical
Image Compression: Study of the Influence of
Noise on the JPEG 2000 Compression
Performance, The 18th International Conference
on Pattern Recognition, Vol. 3, No. 3, pp. 893-
896 (2006).
[5] B.J. Erickson: Irreversible compression of
medical images, Journal of Digital Imaging, Vol.
15, No. 1, pp. 5-14 (2002).
[6] D. Smutek: Quality measurement of lossy
compression in medical imaging, Prague
Medical Report, Vol. 106, No. 1, pp. 5-26 (2005).
[7] S.M. Perlmutter, P.C. Cosman, R.M. Gray, R.A.
Olshen, D. Ikeda, C.N. Adams, B.J. Betts, M.B.
Williams, K.O. Perlmutter, J. Li, A. Aiyer, L.
Fajardo, R. Birdwell, B.L. Daniel: Image quality
in lossy compressed digital mammograms,
Signal Process, Vol. 59, No.2, pp. 189-210
(1997).
[8] E. Seeram: Irreversible compression in digital
radiology. A literature review, Radiography, Vol.
12, No. 1, pp. 45-59 (2006).
[9] Y.Y. Chen: Medical image compression using
DCT-based subband decomposition and
modified SPIHT data organization, International
Journal of Medical Informatics, Vol. 76, No. 10,
pp. 717-725 (2007).
[10] R.M. Slone, D.H. Foos, B.R. Whiting, E. Muka,
D.A. Rubin, T.K. Pilgram, K.S. Kohm, S.S.
Young, P. Ho, D.D. Hendrickson: Assessment of
visually lossless irreversible image compression:
comparison of three methods by using an image-
comparison workstation, Radiology, Vol. 215,
No. 2, pp. 543-553 (2000).
[11]D.A. Koff and H. Shulman: An overview of
digital compression of medical images: can we
use lossy image compression in radiology?,
UbiCC Journal – Volume 4 No. 3
Canadian Association of Radiology Journal, Vol.
57, No. 4, pp. 211-217 (2006).
[12]P.R.G. Bak: Will the use of irreversible
compression become a standard of practice?,
SCAR News Winter 2006 [Online], Vol. 18, pp.
1–11. Available at:
www.siimweb.org/assets/6D50192B-239D-
413E-96CE-04933B9C17F0.pdf .
[13]Group of authors (2008, Jun.). CAR Standards
for Irreversible Compression in Digital
Diagnostic Imaging within Radiology. The
Canadian Association of Radiologists. Ottawa,
Ontario, Canada [Online]. Available at:
www.car.ca/Files/%5CLossy_Compression.pdf
[14]M.P. Eckert and A.P. Bradley: Perceptual quality
metrics applied to still image compression,
Signal Process, Vol. 70, No. 3, pp. 177-200
(1998).
[15]A.M. Eskicioglu: Quality measurement for
monochrome compressed images in the past 25
years, Proceedings of the Acoustics, Speech, and
Signal Processing – ICASSP, pp. 1907-1910
(2000).
[16]Z. Wang, A.C. Bovik, and L. Lu: Why is image
quality assessment. so difficult?, Proceedings of
the IEEE International Conference on Acoustics,
Speech, and Signal Processing – ICASSP’02, pp.
3313-3316 (2005).
[17]P.C. Cosman, R.M. Gray, and R.A. Olshen:
Evaluating quality of compressed medical
images: SNR, subjective rating, and diagnostic
accuracy, Proceedings of the IEEE, pp. 919-932
(1994).
[18]A. Bruckmann and A. Uhl: Selective medical
image compression techniques for telemedical
and archiving applications, Compututers in
Biology and Medicine, Vol. 30, No. 3, pp. 153-
169 (2000).
[19]J. Strom and P. Cosman: Medical image
compression with lossless regions of interest,
Signal Processing, Vol. 59, No. 2, pp. 155-171
(1997).
[20]X. Bai, J.S. Jin, and D. Feng: Segmentation-
based multilayer diagnosis lossless medical
image compression, Proceedings of the Pan-
Sydney Area Workshop on Visual information
Processing, pp. 9-14 (2004).
[21]S. Bokturk, C. Tomasi, B. Girod, and C.
Beaulieu: Medical image compression based on
region of interest with application to colon CT
images, Proceedings of the 23rd Annual
International Conference of the IEEE on medical
and Biomedical Engineering, pp. 2453- 2456
(2001).
[22]M. Penedo, W.A. Pearlman, P.G. Tahoces, M.
Souto, and J.J. Vidal: Region-based wavelet
coding methods for digital mammography, IEEE
Transactions on Medical Imaging, Vol. 22, No.
10, pp. 1288-1296 (2003).
649
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
[23]G. K. Anastassopoulos and A. Skodras:
JPEG2000 ROI coding in medical imaging
applications, Proceedings of the 2nd IASTED
International Conference on Visualisation,
Imaging and Image Processing – VIIP2002, pp.
783-788 (2002).
[24]Z. Lu, D.Y. Kim, and W.A. Pearlman: Wavelet
compression of ECG signals by the set
partitioning in hierarchical trees (SPIHT)
algorithm, IEEE Transactions on Biomedical
Engeeniring, Vol.47, No. 7, pp. 849-856 (2007).
[25]X. Kai, Y. Jie, Z.Y. Min, and L.X. Liang: HVS-
based medical image compression, Europian
Journal of Radiology, Vol. 55, No. 1, pp. 139-
145 (2005).
[26]A. Bilgin, M.W. Marcellin, and M.I. Altbach:
Wavelet Compression of ECG Signals by
JPEG2000, in Proceedings of the Conference on
Data Compression – DCC, pp. 527 (2004).
[27]A. Przelaskowski: Vector quality measure of
lossy compressed medical images, Computers in
Biology and Medicine, Vol. 34, No.3, pp. 193-
207 (2004).
[28]M. Eyadat and I. Muhi: Compression Standards
Roles in Image Processing: Case Study,
Proceedings of the International Conference on
Information Technology: Coding and Computing
– ITCC’05, pp. 135-140 (2005).
[29]H.R. Sheikh and A.C. Bovik: Image Information
and Visual Quality, IEEE Transactions on Image
Processing, Vol. 15, No. 2, pp. 430-444 (2006).
[30]M. Aanestad, B. Edwin, and R. Marvik: Medical
image quality as a socio-technical phenomenon,
Methods of Information in Medicine, Vol. 42,
No. 4, pp. 302-306 (2003).
[31]I. Smith, A. Roszkowski, R. Slaughter, and D.
Sterling: Acceptable levels of digital image
compression in chest radiology, Australasian
Radiology, Vol. 44, No. 1, pp. 32-35 (2000).
[32]T.J. Kim, K.H. Lee, B. Kim, K.J. Kim, E.J. Chun,
V. Bajpai, Y.H. Kim, S. Han, and K.W. Lee:
Regional variance of visually lossless threshold
in compressed chest CT images: Lung versus
mediastinum and chest wall, Europian Journal of
Radiology, Vol. 63, No. 3, pp. 483-488, (2009).
[33]A. Przelaskowski: Compression of
mammograms for medical practice, Proceedings
of the 2004 ACM Symposium on Applied
Computing – SAC '04, pp. 249-253 (2004).
[34]J.E. Bardram: Hospitals of the future –
ubiquitous computing support for medical work
in hospitals, The 2nd international workshop on
UbiCC Journal – Volume 4 No. 3
ubiquitous computing for pervasive healthcare
applications – UbiHealth 2003, Seattle,
Washington, USA, (2003).
[35]C. Atkinson, B. Kaplan, K. Larson, H.M.G.
Martins, J. Lundell, and M. Harris: Ubiquitous
Computing for Health and Medicine, Designing
Ubiquitous Information Environments: Socio-
technical Issues and Challenges, London:
Kluwer Academic Publishers, pp. 355-358
(2005).
[36]D. Dragan and D. Ivetic: An Approach to
DICOM Extension for Medical Image Streaming,
Proceedings of 19th DAAAM International
Symposium 2008, "Intelligent Manufacturing &
Automation", Trnava, Slovakia, pp. 215 (2008).
[37]J. Mirkovic, D. Ivetic, and D. Dragan:
Presentation of Medical Images Extracted From
DICOM Objects on Mobile Devices, The 9th
International Symposium of Interdisciplinary
Regional Research “ISIRR 2007” Hungary –
Serbia – Romania, Novi Sad, Serbia (2007).
[38]D. Dragan and D. Ivetic: Chapter 3:
DICOM/JPEG2000 Client/Server
Implementation, "Environmental, Health, and
Humanity Issues in Down Danubian Region,
Multidisciplinary Approaches", edited by
Dragutin Mihailović & Mirjana Vojinović
Miloradov, ISBN: 978-981-283-439-3, World
Scientific Publishing Co. Pte. Ltd., pp. 25-34
(2009).
[39]L. Chen, C. Lian, K. Chen, and H. Chen:
Analysis and Architecture Design of JPEG2000,
Proceedings of the ICME pp 277-280 (2001).
[40]H. Man, A. Docef, and F. Kossentini:
Performance Analysis of the JPEG2000 Image
Coding Standard, Multimedia Tools and
Applications, Vol. 26, pp. 27-57 (2005).
[41]D. Santa-Cruz, T. Ebrahimi, J. Askelof, M.
Larsson, and C.A. Christopoulos: JPEG 2000
still image coding versus other standards,
Proceedings of the SPIE’s 45th annual meeting,
Applications of Digital Image Processing XXIII,
pp. 446-454 (2000).
[42]M.D. Adams, H. Man, F. Kossentini, and T.
Ebrahimi: JPEG 2000: The Next Generation Still
Image Compression Standard, Contribution to
ISO/IEC JTC 1/SC 29/WG 1 N 1734, (2000).
[43]D.A. Clunie: Lossless Compression of Grayscale
Medical Images - Effectiveness of Traditional
and State of the Art Approaches, Proceedings of
the SPIE 2000, pp. 74-84, (2000).
650
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

A MULTI-LEVEL METHOD FOR CRITICALITY EVALUATION TO

PROVIDE FAULT TOLERANCE IN MULTI-AGENT SYSTEMS

Mounira BOUZAHZAH, Ramdane MAAMRI

Lire Laboratory, Mentouri University, Constantine, Algeria
mbouzahzah@yahoo.fr

ABSTRACT
The possibility of failure is a fundamental characteristic of distributed applications.
The research community in fault tolerance has developed several solutions mainly
based on the concept of replication. In this paper, we propose a fault tolerant
hybrid approach in multi-agent systems. We have based our strategy on two main
concepts: replication and teamwork. Through this work, we have to calculate the
criticality of each agent, and then we divide the system into two groups that use
two different replication strategies (active, passive). In order to determine the agent
criticality, we introduce a multi-level method for criticality evaluation using agent
plans and dependence relations between agents.

Keywords: agent local criticality, agent external criticality, hybrid approach, the
decision agent, the action criticality.
.

1 INTRODUCTION active replication is defined as the existence of

Multi-agent systems offer a decentralized and
cooperative vision of the problems solving, so, they
are particularly well adapted to dynamic distributed
problems, but they are prone to the same failures that
can occur in any distributed software system. A
system faults are classified into to main classes:
• Software faults: those are caused by burgs
in the agent program or in the supporting
environment.
• Hardware faults: these faults are related to
material failures such as: machine crash,
communication breakdown…
Several researches are addressed to solve the
problem of fault tolerance in multi-agent systems
using different strategies. The most important ones
are based on the concept of replication. There are
different strategies to apply replication, the static
strategy which decides and applies replication at
design time like in [1], [2] and [3]. The dynamic
strategy applies replication during the processing
time. This strategy introduces the notion of agent
criticality. It is used by [4] and [5]. According to the
relation between the agent and its replicas there are
two different types of replication. The passive
replication that is defined as the existence of one
active replica that processes all input messages and
transmits periodically its current state to the other
replicas in order to maintain coherence and to
constitute a recovery point in case of failure [6]. The
several replicas that process concurrently all input
messages [7].
This article introduces an approach for fault
resistance in dynamic multi-agent systems. Our
approach is based on the criticality calculation using
agent's plan to determine the agent local criticality.
The interdependence relations are used to calculate
the agent external criticality. According to their
criticalities agents will be oriented towards two
different groups: the critical group managed by an
agent called the supervisor, this group uses the active
replication strategy. The other group uses the passive
replication strategy and it is managed by an agent
called the controller.
The whole system is controlled by the decision agent
that initializes agents to criticality evaluation and
decides which agents are the most critical.
Our approach is general because, first, it is hybrid, it
uses the passive and the active replication strategies
at the same time; and it uses two levels of criticality
evaluation (the local level and the external level).
Through this approach we calculate the agent
criticality dynamically.
The rest of this paper is organized as follows:
section2 covers the related works in the field of fault
tolerance. Section3 gives a description to the
proposed approach based on dynamic replication.
Section4 describes the general architecture of the
system, and finally, Section5 that gives an insight
into our future directions and concludes the paper.

UbiCC Journal – Volume 4 No. 3 651

Special Issue on ICIT 2009 Conference - Bioinformatics and Image
2 RELATED WORKS
Here we review some important works dealing
with fault tolerance in multi-agent systems.
Hagg [2] proposes a strategy for fault tolerance using
sentinels. The sentinel agents listen to all broadcast
communications, interact with other agents, and use
timers to detect agent crashes and communicate link
failure. So, sentinels are guardian agents which
protect the multi-agent system from failing in
undesirable states. They have the authority to
monitor the communications in order to react to fault.
The main problem within this approach is that
sentinels also are subject of faults.
Kumar and al [1] introduce a strategy based on
Adaptive Agent Architecture. This strategy uses the
teamwork to cover a multi-agent system from broker
failures. This approach does not deal completely
with agent failures since only some agents (the
brokers) or part of them can be replicated.
A strategy based on transparent replication is
proposed by [3]. All messages going to and from a
replicated group are funneled through the replicate
group message proxy. This work uses the passive
replication strategy.
These several approaches apply the replication
mechanism according to the static strategy which
allows replication at design time. But recent
applications and mainly those which use the multi-
agent systems are very dynamic the fact that makes it
too difficult to determine the critical agents at the
design time. There are other proposed works that
other use the dynamic replication strategy such as:
Guesssoum and al [4] introduce an automatic and
dynamic replication mechanism. They determine the
criticality of an agent using various data such as:
time processing, the role taken by an agent in the
system… This mechanism is specified for adaptive
multi-agent systems. They focus their work the
platform DIMA [8].
Almeida A. and al [9] propose a method to calculate
the criticality of an agent in a cooperative system.
They use agent plan as the basic concept in order to
determine critical agent. This work uses the
framework DARX [10].
These two works use the dynamic replication that
allows replication at the processing time. This
strategy requires the criticality calculation. The agent
criticality is defined as the impact of a local failure
of an agent on the whole system [11]. The dynamic
strategy is more important than the static one when
dealing within dynamic applications, but it must use
a mechanism able to determine when it is necessary
to duplicate agents.
UbiCC Journal – Volume 4 No. 3
3 THE HYBRID APPROACH
Agents are subject of failure that can cause the
whole system failure. We propose an approach to
introduce fault tolerance in dynamic multi-agent
systems by the use of two main concepts which are:
replication and teamwork. Under our approach the
two replication strategies are used (active and
passive). Since we deal with dynamic multi-agent
systems, we will use the dynamic replication, which
means that agents are not duplicated at the same time
and within the same manner. The question that
arises, therefore, is which are the agents to be
replicated?
4 THE CRITICALITY EVALUATION
The agent criticality denoted CX is defined as the
impact of a local failure of the agent X on the
dysfunction of the whole system. An agent that
causes a total failure of the system will have a strong
criticality.
The criticality evaluation in our approach is realized
at two main levels:
• The local level: here we determine the agent
criticality using its plan of actions.
• The external level: In order to achieve its
current goal the agent does not only use its own data
but it relies on other agents. So, we try to evaluate
the agent external criticality using the relations
between agents.
4.1 Agent Local Criticality
In order to calculate the agent local criticality,
we defined an agent according to the model proposed
by [12]. Each agent is composed of the following
elements:
• Goals: the goals an agent wants to achieve.
• Actions: the actions the agent is able to
perform.
• Resources: the resources an agent has
control on.
• Plans: the plan represents the sequence of
actions that the agent has to execute in order to
achieve a certain goal.
The title should be typed in capital letters, using
Times New Roman type face with 14 points in size,
bold. It should be centered on the first page
beginning on the 6th line.
4.1.1 Agent Plan
We conceder that each agent knows the actions
sequence that he has to execute in order to achieve
its current goal. Therefore, we propose the use of a
graph to represent the sequence of actions called
agent's plan. These plans are established for short
terms because the environment considered is
dynamic. The graph that we use in this work is
652
Special Issue on ICIT 2009 Conference - Bioinformatics and Image
inspired from that proposed by [9]. The agent plan is
represented by a graph where the nodes represent
actions and edges represent relations between
actions. These relations are the logical functions
AND and OR. A node n which is connected to k
other nodes (n1, n2... nk) using AND edges
represents an action that will be achieved only if all
its following actions are executed. However, a node
n connected to its k followers using OR edges
represents an action that is achieved if only one
following action is executed. The work proposed in
[5] uses a different description concerning the agent
plan and it proposes the existence of internal and
external actions. However, we are interested to
actions which are executed by the agent (local
actions), Thus, according to our description an agent
X will be represented as follows (Figure 1):
Agent X
A considered action when they are connected using
AND
AND
AND
B1 B2 Bk
OR
OR
C1 C2 Cn
Figure 1. Agent X plan.
4.1.2 Action Criticality
In this paper we propose the use of two types of
action’s criticality: the action initial criticality given
by the designer, and the action dynamic criticality
calculated according to the agent plan.
Thus, the criticality of an action A denoted CA is
calculated as follows:
CA = initial criticality + dynamic criticality
CA= CIA + CDA
4.1.3 Action Initial Criticality
We admit that a critical agent is the one which
executes critical actions. And we propose the
following criteria to define the initial criticality of an
action:
• An action which can be done by several
agents can be regarded as being not too critical, but if
one other action is done by few agents it will be
regarded as a critical one.
UbiCC Journal – Volume 4 No. 3
• The number of necessary resources that
are required for the execution of an action can be
also a factor to determine the initial criticality of an
action. When an action requires many resources to be
executed, it introduces a strong criticality.
• Hardware data influence, also, the action
initial criticality.
• Finally, according to the application field,
the designer can determine semantic information that
can define the initial criticality of an action.
Thus, at the design time each action A has a value
called the initial criticality denoted CIA.
4.1.4 Action Dynamic Criticality
The dynamic criticality of an action denoted CD
is defined as the value attributed to an action
according to its position in the agent plan. There is
one factor that can influence the action criticality
which is the set of its following actions.
We use the function MULTIPLICATION to
represent the following actions influence on the
AND edges. Since we have indicated that when an
action A connected to its followers (B1, B2,…, Bk) by
AND edges, the achievement of A implies that all its
following actions are achieved. If we represent the
actions with a group of sets we will have the
following result:
A= (B1 B2 ... Bk ).
CA = CIA + (CB1 * CB2 *...* CBK)
One other function SUM is used to represent the case
where one action is connected to its followers by OR
edges. If we consider action B2 (figure 1) connected
to its followers (C1, C2, …, Cn) by OR edges, in
term of sets we will have:
B2 = (C1 ∪ C2 ∪...∪ Cn )
Thus, B2 criticality is calculated as follows:
CB2 = CIB2 + (CC1 + CC2 +...+ CCn)
An action which has no follower is called a terminal
action. The dynamic criticality of a terminal action
equals to 0. This means that the criticality of a
terminal action equals to its initial criticality.
4.1.5 Agent Local Criticality Calculation
In order to determine the agent local criticality,
we admit that each agent knows at an instant t the
actions sequence which it has to execute to achieve
its current goal. The local criticality of agent CL agent
is calculated as follows:
CL agent = Sum ( Caction1 +....+ Caction n).
This criticality calculation is made directly by the
agent.
653
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

Example: The relation between agents is defined in our model

Let's calculate the agent local criticality following using the following set:
the agent plan (Figure2):
Set = {T, P, N}

T: represents the relation type, it can be cooperative

Agent X or adoptive.
P: is the relation weight, here it represents the sum of
A the initial criticalities of the actions that are executed
using this relation:
AND P = Sum CI of the actions executed using the relation

N: the number that represent the agents having the

B C same current goal.
The external criticality in this case is calculated as
OR follows:
Cex agent = p/N
D E
In adoptive case N = 1.

Figure2. Agent X plan 4.3 Agent Criticality

The agent criticality denoted Cagent is considered
Table1. The actions initial criticalities. as agent propriety, it is calculated by the agent
directly using the following relation:
CIA CIB CIC CID CIE
2 1 3 5 10 Cagent = CL agent + Cex agent

CA = CIA + (CB * CC) 4.4 Determine the Most Critical Agents

CB = CIB Each agent must pass the calculated criticality at
= 1 B is a terminal action. the instant t to an other agent called the decision
CC = CIC + (CD + CE) agent. This later uses these values to determine the
CD = CID = 5 D is a terminal action most critical agents. According to usual arithmetic,
CE = CIE = 10 the median value of N numbers gives an index to
CC = 18 divide a unit into two parts. The decision agent uses
CA = 20 the following algorithm in order to determine the two
groups of agents.
The local criticality of agent X:
C LX = (CA + CB + CC + CD + CE ) Algorithm: decision
= 54. Begin
Sumcriticalities 0
4.2 Agent External Criticality For each agent I do
According to the agent definition shown in the Read Cagent i /* Cagent i the criticality of the
previous section the agent possesses a set of plans. agent I*/
Each plan is formed of a sequence of actions that the /* the sum of agents criticalities calculation*/
agent has to execute in order to achieve its current Sumcriticalities Sumcriticalities + Cagent i
goal. These actions do not necessarily belong to the For each agent I do
agent set of actions; therefore, an agent may depend If (Cagent i >= Sumcriticalities / number of the agents)
on other agents to carry on a certain plans. Then
There are six different dependence situations GT =1
identified by [12]. Through this work we are Else
interested to two main dependence relations which GT=2
are: /* GT is an agent property, if GT=1 then the agent
• The cooperative relation when an agent is affected to the critical group, else it is in the other
infers that he and other agents are depending on each group*/
other to realize the same current goal. End.
• The adoptive relation the situation when
an agent infers that he and other agents are Finally, agents are oriented towards two different
depending on each other to realize different current groups.
goals.

UbiCC Journal – Volume 4 No. 3 654

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

4.5 Criticality Re-Evaluation

The criticality calculated in the previous sections
DA
is determined at the instant t; it must be updated
throughout the execution since our system is
dynamic. We propose a solution based on two
strategies: SUP CONT
• Time strategy: the decision agent has a
clock that gives alarms to re-evaluate agents'
criticalities at each fixed time interval t.
• Event strategy: There are many events that
act on the system and caused criticality revision such
as: an agent failure, a machine failure. NCG
CG SA
4.6 Determine the Agents Groups
The concept of teamwork is used by different
approaches such as [1] and [2]. Concerning this Figure3. The system's architecture.
approach, criticality calculation leads to the creation
of two agents' groups. This stage makes it possible to DA: The Decision Agent.
determine a strategy for fault tolerance. SUP: The Supervisor.
• The critical agents' group: uses the active CONT: The Controller.
replication. Each critical agent will have only one SA: The system's Agents.
active replica called the follower. This later is an CG: Critical Group.
agent that has the same plan and executes the same NCG: Non Critical Group.
action processed by the critical agent but after the The system consists of the dynamic multi-agent
reception of a permission message sent from the system and the three added agents: the decision agent
supervisor. The supervisor is an agent that that controls the whole system, the supervisor which
guarantees the management of the critical group. manages the critical group and the manager of the no
critical group called the controller.
• The no critical agents' group: this group
uses the passive replication strategy. Each no critical
5.1 The Decision Agent
agent will have only one passive replica. It is the no
This agent offers two fundamental services. First
critical agent that executes all the actions and
it determines critical agents the fact that allows the
transmits its current state. If the active agent is lost
division of the whole system into two main groups.
its replica is activated by an other agent called the
And it initializes the agents to the process of
controller which is the group's manager.
criticality re-evaluation following the dynamicity of
The criticality revision is done by the decision
the system.
agent according to two factors: time-driven factor
We use the concept of the sequence diagram [13] in
and event-driven factor .When an agent is considered
order to represent the decision agent's role as follows
as critical at a given time t. It establishes a contract
(Figure 4).
with the supervisor agent. So, the agent will have an
active replica. If at the instant t + t, the re-
evaluation of the criticality considered the same DA SA SUP CONT
agent as no critical its contract will be deleted. And 1
one other contract will be established within the
2
controller.
3
5 SYSTEM ARCHITECTURE 4

In order to guarantee fault tolerance in dynamic

multi-agent systems, we have added three agents that
allow error detection and data recovering. The
general architecture of the system is given by the
following diagram (figure3):
Figure 4. The sequence diagram for the decision
agent.

DA: The Decision Agent.

SA: The System's Agent.
SUP: The Supervisor.

UbiCC Journal – Volume 4 No. 3 655

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

CONT: The Controller. agent replication using the passive strategy. This
1: The Criticality Evaluation. agent verifies and detects failure among its group's
2: Pass the Criticality C. agents using the same technique employed by the
3: Decision. supervisor. Since the detection of failure, the passive
4: GT= 1. replica will be active and an other passive replica
5: Establish contract with the Supervisor. will be added. The controller's sequence diagram is
6: GT= 2. represented as follows (Figure 6):
7: Establish contract with the Controller.
The Controller Non critical agent Passive replica
5.2 The Supervisor 1
This agent allows the active replication. During
execution time, the critical agent transmits 2
periodically its current state to the supervisor, this 3
latter gives permission messages in order to validate
the replica's execution.
The supervisor allows also failure detection. This 4
service makes it possible to detect if an agent is still
alive and that it does not function in a synchronous 5
6
environment [14]. The supervisor achieves this 8 7
service within the use of a clock that initializes the 9
control messages sent to the critical agents. Each
activated (critical replica) has a failure – timer which
gives the max time used by the agent to answer. If
Figure 6. The sequence diagram for the controller.
the agent does not give an answer a failure is
detected.
1: Establish contract.
Since the failure detection, the supervisor creates a
2: Passive replication process.
replica and the follower takes up the failed agent.
3: Current state's message.
The supervisor's services are represented by the
4: Controlling message.
following diagram (Figure 5).
5: Yes.
6: Answer.
The supervisor Critical agent Active replica
7: No.
1 8: T > Max Time.
2 9: replica activated + Agent recovering.

3 6 CONCLUSION
4
This article proposes a rich approach for fault
5
resistance in dynamic multi-agent systems based on
6 replication and teamwork. We use the two strategies
7
9 8 (active and passive) within the existence of one
strong replica at one time; this fact allows the
10
decreasing of charges. In order to guarantee failure
detection and system controlling three other agents
are added.
Figure5. The sequence diagram for the supervisor In further work, we are interesting to propose a more
formal model for criticality calculation and to
1: Establish contract. validate our approach trough implementation.
2: Active replication process.
3: Current state's message. 7 REFERENCES
4: Permission message.
5: Controlling message. [1] S.Kumar, P. R Cohen., H.J. Levesque:The
6: Yes. adaptive agent architecture: achieving fault-
7: Answer. tolerance using persistent broker teams. , The
8: No. Fourth International Conference on Multi-Agent
9: T > Max Time. Systems (ICMAS 2000), Boston, MA, USA,
10: Agent recovering. July 7-12, 2000.
5.3 The Controller [2] S. Hagg : A sentinel Approach to Fault Handling
Is the no critical agent group's manager it allows

UbiCC Journal – Volume 4 No. 3 656

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

in Multi-Agent Systems . , Proceedings of the Impossibility of distributed consensus with one

second Australian Workshop on Distributed AI, faulty process. , JACM, 1985.
Cairns, Australia, August 27, 1996.

[3] A. Fedoruk, R. Deters: Improving fault –

tolerance by replicating agents. , Proceedings
AAMAS-02, Bologna, Italy, P. 144-148.

[4] Z.Guessoum , J-P.Briot, N.Faci, O. Marin : Un

mécanisme de réplication adaptative pour des
SMA tolérants aux pannes. , JFSMA, 2004.

[5] A. Almeida, S. Aknine, et al : Méthode de

réplication basée sur les plans pour la
tolérance aux pannes des systèmes multi-
agents. , JFSMA, 2005.

[6] M. Wiesmann, F. Pedone, A. Schiper, et al:

Database replication techniques : a three
parameter classification". Proceedings of 19th
IEEE Symposium on Reliable Distributed
Systems (SRDS2000),Nüenberg ,Germany,
October 2000 . IEEE Computer Society.

[7] O. Marin : Tolerance aux Fautes. , Laboratoire

d'Informatique de Paris6, Université PIERRE &
MARIE CURIE.

[8] N. Faci, Z. Guessoum, O. Marin: DIMAX: A

Fault Tolerant Multi - Agent Platform. ,
SELMAS' 06.

[9] A. Almeida, and al: Plan-Based Replication for

Fault Tolerant Multi-Agent Systems. , IEEE
2006.

[10] O. Marin, P. Sens,"DARX: A Framework For

Tolerant Support Of Agent Software. ,
Proceedings of the 14th International Symposium
on Software Reability Engineering, IEEE,2003.

[11] A. Almeida, S. Aknine, et al: A Predective

Method for Providing Fault Tolerance in Multi-
Agent Systems. , Proceedings of the IEEE /
WIC/ACM International Conference of
Intelligent AgentTechnologie (IAT'06).

[12] J. S. Sichman, R. Conte, et al: A Social

Reasoning Mechanism Based On Dependence
Networks. , ECAI 94, 11th European Conference
On Artificial Intelligence, 1994.

[13] M. Jaton : Modélisation Objet avec UML. ,

cours,chapitre13.
http://www.iict.ch/Tcom/Cours/OOP/Livre/Livre
OOPTDM.html.

[14] M. Fischer, N. Lynch, M. Patterson:

UbiCC Journal – Volume 4 No. 3 657

 Special Issue on ICIT 2009 Conference - Bioinformatics and Image

A MODIFIED PARTITION FUSION TECHNIQUE OF MULTIFOCUS

IMAGES FOR IMPROVED IMAGE QUALTITY
Dheeraj Agrawal1, Dr.Al-Dahoud Ali2, Dr.J.Singhai3
1,3
Department of Electronics and Communication Engineering, MANIT, Bhopal. (M.P.), INDIA
2
Faculty of Science and Information Technology, Al-Zaytoolah university of Amman, Jorden.
1
dheerajagrawal@manit.ac.in,2aldahoud@alzaytoonah.edu.jo,3 j_singhai@rediffmail.com

ABSTRACT
This paper presents a modified Partition fusion technique for multifocus images for improved
image quality. In the conventional partition fusion technique image sub blocks are selected for
fused image based on their clearity measures. The clearity measure of an image sub block can be
determined by second order derivative of the sub image. The performance of these clearity
measures is insufficient in noisy environment. In the modified technique, before dividing the
image into sub images, it is filtered through linear phase 2-D FIR low pass digital filter to
overcome the effect of noise. The modified technique uses choose max selection rule to select the
clearer image block from the differently focused source images. Performance of the modified
technique is tested by calculating the value of RMSE. It is found that EOL gives lowest RMSE
with unequal block sizes while SF gives lowest RMSE with equal block sizes when used as
clearity measure in modified partition fusion technique.

Keywords: EOL, RMSE, MI, FIR.

1. INTRODUCTION of its robustness to noise. This method does not

The images are the real description of objects. When
these images are taken from camera there are some
limitations of a camera system. One of which is the
limitation of depth of focus. Due to this an image
cannot be captured in a way that all of its objects are
well focused. Only the objects of the image with in
the depth of field of camera are focused and the
remaining will be blurred. To get an image well
focused everywhere we need to fuse the images taken
from the same view point with different focus
settings. The term image fusion is used for practical
methods of merging images from various sensors to
provide a composite image which could be used to
better identify natural and manmade objects. In the
recent research works the researchers have used
various techniques for multi-resolution image fusion
and multi focus image fusion. . Li’ et al.,(2001-2002)
introduced a method based on the selection of clearer
image blocks from source images[8,9].In this
method, image is first partitioned into blocks then
focus measure is used as activity level measurement.
Based on activity level, best image block is selected
by choosing image block having maximum value of
activity for fused image. The advantage of this
method is that it can avoid the problem of shift-
variant, caused by DWT. Also according to the
analysis of the image blocks selection method, the
implementation is computationally simple and can be
used in a real-time. The limitation of this method is
perform quit well for noisy images. To overcome this
limitation preprocessing of the image has been done
with the help of a low pass filter.
The measure of clarity plays an important role in this
kind of fusion method. A better measure results in a
superior fusion performance. However, little work
has been done on the image clarity measures in the
field of multi-focus image fusion. The image clarity
measures, namely focus measures, are deeply studied
in the field of autofocusing. The paper also
considered the fact that the background information
lie in low frequency component of the image; so
while using different focusing parameters the method
proposed will be able to extract the features of
background information when the image is passed by
a low pass filter. This paper is organized as follows.
A brief description of focus measures is given in
Section 2. Proposed modified technique for obtaining
low RMSE fused image is discussed in Sections 3
and Sections 4 presents results of the proposed
method in comparison with existing methods.
2. FOCUS MEASURES
A value which can be used to measure the depth of
field from the acquired images can be used as focus
measure. Depth of field is maximum for the best
focused image and generally decreases as the defocus
increases.

UbiCC Journal – Volume 4 No. 3 658

 Special Issue on ICIT 2009 Conference - Bioinformatics and Image

A typical focus measure satisfies following 1 M N

requirements: RF = ∑∑
M × N x=1 y =2
(f ( x, y) − f ( x, y − 1))2 and
1. Independent of image content;
2. monotonic with respect to blur; 1 M N
3. The focus measure must be unimodal, that is, it
must have one and only one maximum value;
CF = ∑∑
M × N x=2 y=1
(f ( x, y) − f ( x −1, y))2
4. Large variation in value with respect to the degree 5. Visibility (VI): This focus measure is inspired
of blurring; from human visual system, and is defined as
5. Minimal computation complexity; M N f ( m ,n ) - µ
6. robust to noise. V I= ∑ ∑
The conventional focus measures used to measure the m =1 n=1 µ α +1
clearity of the images are variance, EOG, EOL, and Where µ is the mean intensity value of the image, and
SF. These focus measures are expressed as following α is a visual constant ranging from 0.6 to 0.7.
for an M x N image with f(x, y) be the gray level
intensity of pixel (x, y). 3. MODIFIED TECHNIQUE FOR LOW RMSE
Most of the focus measures are based on the
1. Variance: The simplest focus measure is the idea of emphasizing high frequency contents of the
variance of image gray levels. The expression for the image and measure their quantity. This comes from
M × N image f(x, y) is: an idea that blurring suppresses high frequencies
regardless of particular Point Spread Function. [13]
1 M N Considering the performance of various focus
variance =
M×N
∑ ∑ (f ( x , y ) − µ )
x =1 y =1
2
, measures, EOL found to be the best among all [8].
Laplacian of an image is determined by second order
Where µ is the mean value and is given as derivative of the image. The performance of the
M N second order derivative decreases if noise is present
1
µ =
M×N
∑ ∑ f ( x, y )
x =1 y =1
in the source images as show in Fig-1

2. Energy of image gradient (EOG): This focus Fig-1 (A), (E)

measure is computed as:
M −1 N −1
EOG= ∑ ∑ (f
x =1 y =1
x
2
+ f y2 )
Where
Fig-1 (B), (F)
f x = f ( x + 1, y ) − f ( x , y )
f y = f ( x , y + 1) − f ( x , y )
3. Energy of Laplacian of the image (EOL): It
is used for analyzing high spatial frequencies
associated with image border sharpness is the
Fig-1 (C), (G)
Laplacian operator.
M −1 N −1
EOL= ∑ ∑
x=2 y=2
( f xx + f yy ) 2

Where
fxx +fyy =−f(x −1, y −1) −4f(x −1, y) −f(x −1, y +1) −4f(x, y −1) Fig-1 (D), (H)
+20f(x, y) −4f(x, y +1) −f(x +1, y −1) −4f(x +1, y) −f(x +1, y +1)
4. Spatial frequency (SF): Strictly speaking
frequency is not a focus measure. It is a modified
version of the Energy of image gradient (EOG).
Spatial frequency is defined as:
SF = RF2 + CF2 Fig-1 the performance of second order derivative in
Where RF and CF are row and column frequencies presence of various degree of noise.
respectivly:

UbiCC Journal – Volume 4 No. 3 659

 Special Issue on ICIT 2009 Conference - Bioinformatics and Image

Fig-1(A) shows ramp edges profile of an image
separating black region and white region. The entire
transition from black to white represents a single
edge. In fig-1(A) image is free of noise and its grey
level profile is sharp and smooth.Fig-1(B-D) are
corrupted by additive Gaussian noise with zero mean
and standard deviation of 0.1, 1.0 and 10.0 intensity
levels respectively and their respective grey level
profile shows noise added on the ramp by ripple
effects. The images in the second column are the
second derivatives of the images on the left. Fig-1(E)
shows two impulses representing presence of edge in
the image.Fig-1(F-H) shows that as the noise
increases in the image the detection of impulses
becomes difficult making it nearly impossible to
detect the edge in the image. This shows that the
focus measure using the second order derivative also
fails to decide about the best focused image in noisy
environment. Thus for selection of best focused
image removal of noise is essential before applying
fusion technique to obtain best focused image.
The proposed focusing technique uses the linear-
phase 2-D FIR low pass digital filter to remove the
noise from the differently focused images. Filter uses
Parks-McClellan algorithm [19], [20].The Parks-
McClellan algorithm uses filter with Equiripple or
least squares approach over sub-bands of the
frequency range and Chebyshev approximation
theory to design filters with an optimal fit between
the desired and actual frequency responses. The
filters are optimal in the sense that the maximum
error between the desired frequency response and the
actual frequency response is minimized. Filters
designed this way exhibit an equiripple behavior in
their frequency responses and are sometimes called
equiripple filters. Filters exhibit discontinuities at the
head and tail of its impulse response due to this
equiripple nature.These filters are used in existing
fusion algorithm before partitioning the image as
shown in fig-3. The source images are passed through
2D FIR low pass filter of order 4 and having
characteristic as shown in fig-2. For these low pass
filtered images conventional focus measure such as
Variance, Energy of Gradient, Energy of Laplacian,
Spatial frequency are computed.
Fig.2.Perspective plot of linear phase 2-D FIR
Lowpass digital filter
Setup for proposed algorithms
A schematic diagram for proposed image fusion
method is shown in Fig-3.The paper proposes
modification for obtaining best focus measure in
noisy environment by use of filter at step -2 in the
existing algorithms used by Li et. al [8].
The fusion method consists of the following steps:
Step 1. Decompose the differently focused source
images into blocks. Denote the ith image block of
source images by Ai and Bi respectively.
Step 2. Filter the images through a 2D FIR low pass
filter for removal of noise.
Step 3. Compute the focus measure of each block,
and denote the results of Ai and Bi by MiA and ,MiB
respectively.
Step 4. Compare the focus measure of two
corresponding blocks Ai , and Bi and construct the ith
block Di of the composite image as
⎧A Mi > Mi
A B
Di = ⎨ i
Mi > Mi
B A
⎩ Bi
Step 5. Compute root mean square error (RMSE) for
the composite image with a reference image

A FIR LPF
Activity
level
Combining
measure
Portitioned by choose
images max

B FIR LPF Fused image

Fig.3: Schematic diagram for evaluating proposed focusing technique in Multi-focus image fusion

UbiCC Journal – Volume 4 No. 3 660

 Special Issue on ICIT 2009 Conference - Bioinformatics and Image

the clarity of image block. However using a block

4. RESULTS:
The experiment is performed on toy image of size
512×512. The multifocus images used for fusion are
left focused, right focused and middle focused as
shown in Fig 4, 5 and 6 respectively. These
multifocus images are filtered through linear phase
2D FIR low pass digital filter to reduce low
frequency noise then filtered images are fused using
Li’s algorithm for various focus measures. The
performance of existing and modified algorithm is
compared qualitatively by calculating RMSE of fused
images.
RMSE is defined as:
M N
∑ ∑ {R ( x, y ) − F( x, y )}
2
size too large is undesirable because larger block of
sub image may contain two or more objects at
different distances from the camera, and
consequently will lead to a less clear image.
The experimental results in table-1 and table-2 show
that the performance of proposed method for all the
focus measures improves with reduced RMSE with
nearly one forth of RMSE of existing algorithm.
Visual analysis is shown form fig-4 to Fig-14. Fig-7
is the reference image taken all parts focused. Fig 8
to Fig 11 shows the fused images while considering
different focus measures with existing partition
fusion method. Fig 12 to Fig 14 shows the fused
images while considering different focus measures

x =1 y =1
RMSE=
M×N
Where R and F are reference image and composite
image respectively, with size M × N pixels.
Table-1 shows the RMSE of fused images using
different focus measures and for equal block size of
images. Table-2 shows the RMSE of fused images
for unequal block size of source images. Table-1
shows that fused images using SF as focus measures
gives lowest RMSE values and Table-2 shows that
for unequal block size of images EOL perform better
then other clearity measures when used in modified
partition fusion technique. The analysis of Table-1
shows that RMSE of fused image decreases with
increase in the block size of sub image only with SF.
Analysis of Table-2 shows that RMSE of fused
image decreases with increase in the block size of sub
image for all clearity measures because the larger
image block gives more information for measuring
with proposed modified algorithm of partition fusion
with 2-D low pass filter.
5. CONCLUSION:
In this paper modified method of image fusion was
used considering various focus measure capabilities
of distinguishing clear image blocks form blurred
image blocks. Experimental results show that
preprocessed, 2-D FIR low pass filtered image in
modified method provide better performance in terms
of low RMSE than the previous methods of
information fusion. Also from the results it is
concluded that performance of the image fusion
method depends on block size taken during the
partitioning of source images. The experiment shows
that EOL gives low RMSE with unequal block sizes
while SF gives low RMSE with equal block sizes.
This is an issue that will be investigated in future on
adoption methods for choosing the image block size.

Table-1
Evaluation of different focus measures with equal block sizes on basis of RMSE
Block Focus measure
size
Partition fusion method Modified Partition fusion Method
Variance EOG EOL SF of
variance EOG EOL SF VI of LPF of LPF of LPF LPF
images images images images
4×4 4.5814 3.9383 3.6437 3.9383 4.1383 0.9514 0.9514 0.9301 0.9514
8×8 4.3658 4.0264 3.1466 3.9292 4.2110 0.9606 0.9373 0.8686 0.9373
16×16 4.7037 4.7720 3.4659 4.0517 3.9574 1.1872 1.1820 1.1561 0.8827
32×32 4.4221 4.6485 3.0888 3.8506 3.6183 1.2043 1.1382 1.1531 0.8949
64×64 4.6588 4.6000 3.8727 3.8927 3.4368 1.2194 1.2248 1.2013 0.8744
Numbers in bold and italic indicate the lowest RMSE obtained over different block sizes

UbiCC Journal – Volume 4 No. 3 661

 Special Issue on ICIT 2009 Conference - Bioinformatics and Image

Table-2
Evaluation of different focus measures with unequal block sizes on basis of RMSE

Block Focus measure

size
Partition fusion method Modified Partition fusion Method
Variance EOG EOL SF of
variance EOG EOL SF VI of LPF of LPF of LPF LPF
images images images images
4×8 4.5447 4.0106 3.3199 4.0118 4.2340 0.9626 0.9626 0.9073 0.9626
8×16 4.4089 4.0035 3.1806 4.0407 4.1160 0.9346 0.9284 0.8843 0.9255
16×32 4.6329 4.0159 3.8220 3.9351 3.9399 0.9119 0.8923 0.8776 0.8889
32×64 4.2559 3.9797 3.5020 3.8944 3.5630 0.9066 0.8893 0.8715 0.8874
Numbers in bold and italic indicate the lowest RMSE obtained over different block sizes

REFERENCES: [10] Ligthart,G.,Groen, F.,1982.A Comparison of

[1] Burt, P.J., Andelson, E.H., 1983.The Laplacian different Autofocus Algorithms.
pyramid as a compact image code.IEEE Trans. In:Proc.Int.Conf.on Pattern Recognition.pp.597-
Commun.31, 532-540. 600.
[2] Burt,P.J.,Kolezynski,R.J., 1993.Enhanced [11] Miao,Q.,Wang,B.,2005.A Novel Adaptive
image capture through fusion.In:Proc.4th Int. Multi-focus Image Fusion Algorithm Based on
Conf. on Computer PCNN and sharpness.In:Proc.of
Vision,Berlin,Germany,pp.173-182. SPIE,VOL.5778.pp.704-712.
[3] Eltouckhy,H.A., Kavusi,S.,2003.A [12] Nayar,S.K.,Nakagawa,Y.,1994.Shape from
Computationally Efficient Algorithm for Multi- focus.IEEE Trans.Pattern Anal.
Focus Image Reconstruction.In:Proc. Of SPIE Mach.Intell.16(8),824-831.
Electronic Imaging.pp.332-341. [13] Subbarao,M.,Choi,T.,Nikzad,A.,1992.Focusing
[4] Eskicioglu,A.M.,Fisher,P.S., 1995.Image Techniques.In:Proc.SPIE. Int.Soc. Opt. Eng.,
quality measures and their performance.IEEE 163-174.
Trans. Commun. 43(12), 2959-2965. [14] Toet,A.,Van Ruyven,L.J.,Valeton ,
[5] Hill,P.,Canagarajah,N.,Bull,D.,2002.Image J.M.,1989.Merging thermal and visual images
Fusion using Complex Wavelets.In:Complex by a contrast pyramid.Opt.Eng.28(7),789-792.
Proc. 13th British Machime Vision [15] Unser,M.,1995.Texture classification and
Conf.University of Cardiff,UK,pp.487-496. segmentation using wavelet frames.IEEE
[6] Krotokov,E.,., Trans.Image Process.4(11),1549-1560.
1987.Focusing.Int.J.Comput.vis.1,223-237 [16] Yeo, T.,Ong, S.,Jayasooriah,S.R.,
[7] Li,H.Manjunath,B.S., 1993.Autofocussing for tissue
Mitra,S.K.,1995.Multisensor image fusion using microscopy.Image Vision Comput.11,629-639.
wavelet transform.Graph. Models Image [17] Wei Huang, Zhongliang Jing .,2006.Evaluation
Process.57 (3), 235-245. of Focus Measures in Multi-focus image
[8] Li,S.,Kwok,J.T.,Wang,Y.,2001.Combination of fusion.Pattern Recognit.pp.lett .28(2007).493-
images with diverse focuses using the spatial 50
frequency.Inf.Fusion 2,169-176.
[9] Li,S.,Kwok,J.T.,Wang,Y.,2002.Multi focus
image fusion using Artificial Neural
Networks.Pattern Recognit.Lett.23,985-997.

UbiCC Journal – Volume 4 No. 3 662

 Special Issue on ICIT 2009 Conference - Bioinformatics and Image

Fig. 4 left focused image Fig. 5 right focused image Fig. 6 middle focused image

Fig.7.All focused image Fig .8. Fused images Formed from Fig.9.Fused images formed
(reference image) variance From EOG(16×32)

Fig.10. Fused images formed Fig.11.Fused images formed Fig.12.Fused images

from EOL(32x64) from SF (32x 64) formed from LPF and SF (32×32)

Fig.13.Fused images formed Fig.14.Fused images formed

from LPF and EOG(16×32) from LPF and EOL(64×64)

UbiCC Journal – Volume 4 No. 3 663

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

INTEGRATING BIOMEDICAL ONTOLOGIES – OBR-SCOLIO

ONTOLOGY

Vanja Luković
Information technology

Danijela Milošević
Information technology

Goran Devedžić
Information technology

ABSTRACT
This paper analyses a broad scope of research papers dealing with the process of integrating biomedical
ontology with the FMA reference ontology. Namely, we want to investigate the capability of this process
appliance in development of the OBR-Scolio application ontology for the pathology domain of spine, rather
the scoliosis domen. Such ontology is one of the many objectives in the realization of the project named:
“Ontological modeling in bioengineering” in the domain of orthopedics and physical medicine.

Keywords: Biomedical ontology, Formal ontology, Reference ontology, Application ontology

1 INTRODUCTION ontology, forming this way FMA-RadLex

Biomedical ontologies are being developed in
ever growing numbers, but there is too little
attention paid for ontology alignment and
integration, in other that results already obtained by
the one terminology based application ontology can
be utilized in other similar application ontologies.
No scientific advance can be obtained with the
horizontally integration between two application
ontologies, although vertical integration between
ontologies in all categories is needed [1]. In this
way formal, top level ontologies should provide the
validated framework for reference ontologies,
which represent the domains of reality studied by
the basic biomedical sciences. The latter should
then in turn provide the scientifically tested
framework for a variety of terminology-based
ontologies developed for specific application
purposes.
In this paper according to [1], we denote how
the process of vertical integration of the FMA
(Foundational Model of Anatomy) reference
ontology [5] with the BFO (Basic Formal
Ontology) top-level ontology [3] can support the
process of horizontal integration of the two
reference ontologies: PRO (Physiology Reference
Ontology) [8] and PathRO (Pathology Reference
Ontology), forming accordingly the new reference
ontology OBR (Ontology of Biomedical Reality),
which is therefore federation of the three
independent reference ontologies which range over
the domains of anatomy, physiology and pathology.
Moreover according to [9], we denote the
process of vertical integration of the RadLex
radiology terminology with the FMA reference
application radiology ontology. This described
process is then utilized for forming the OBR-Scolio
application ontology for the pathology domain of
spine (scoliosis domen) from the OBR reference
ontology.
2 BFO ONTOLOGY
BFO [3] is a formal, top-level ontology which is
based on tested principles for ontology
construction. BFO consists of the SPAN and SNAP
ontologies. The SPAN ontology relates to
occurrents, processing entities (events, actions,
procedures, happenings) which unfold over an
interval of time. The complementary SNAP
ontology relates to continuants, the participants in
such processes, which are entities that endure over
the time, during the period of their existence.
Anatomy is a science that studies biological
continuants, while physiology studies biological
occurrents. Pathology, on the other hand, is
concerned with structural alterations of biological
continuants and with perturbations of biological
occurrents which together are manifested as
diseases. Moreover, BFO draws distinctions also
between instances and universals and specifies
relations which link them.
3. FMA ONTOLOGY
The FMA [5] is reference ontology for anatomy,
which according independent evaluations satisfies
fundamental requirements for ontological
representation of human anatomy [6, 7].

UbiCC Journal - Volume 4 No. 3 664

Special Issue on ICIT 2009 Conference - Bioinformatics and Image
Hence, the domain of the FMA is anatomy of
the idealized human body. FMA uses the hierarchy
of classes of anatomical entities (anatomical
universals) which exist in reality through their
instances. The root of the FMA’s anatomy
taxonomy (AT) is Anatomical entity and its
dominant class is Anatomical structure. Anatomical
structure is defined as a material entity which has
its own inherent 3D shape and which has been
generated by the coordinated expression of the
organism’s own structural genes. This class
includes material objects that range in size and
complexity from biological macromolecules to
whole organisms. The dominant role of Anatomical
structure is reflected by the fact that non-material
physical anatomical entities (spaces, surfaces, lines
and points) and body are conceptualized in the
FMA, in terms of their relationship to anatomical
structures.
4. OBR ONTOLOGY
Figure 1. Ontology of Biomedical Reality OBR
The root of OBR is the universal Biological
entity (Fig. 1). A distinction is then drawn between
the classes: Biological continuant and Biological
occurrent, the definitions of which are inherited
from BFO [3]. The class Biological continuant is
subdivided into classes: Оrganismal continuant,
which includes entities that range over single
organisms and their parts and Extra-organismal
continuant, which includes entities that range over
aggregates of organisms. Accordingly, the class
Biological occurrent is subdivided into classes:
Оrganismal occurent and Extra-organismal
occurent, which include processes associated with
single organisms and their parts i.e. processes
associated with aggregates of organisms.
The class Organismal continuant is subdivided
into classes: Independent organismal continuant
and Dependent organismal continuant.
Extrapolating from the FMA’s principles,
Independent organismal continuants have mass and
are material, whereas Dependent organismal
continuant are immaterial and do not have mass.
OBR ontology distinguishes anatomical
(normal) from pathological (abnormal) material
entities. Accordingly, the class Independent
organismal continuant is subdivided into classes:
Material anatomical entity and Material
UbiCC Journal - Volume 4 No. 3
pathological entity. The class Material anatomical
entity is subdivided into classes: Anatomical
structure and Portion of canonical body substance,
on the basis of the possession or non-possession of
inherent 3D shape. Within the class anatomical
structure OBR ontology make a distinction between
canonical anatomical structures, which exist in the
idealized organism, and variant anatomical
structures, which result from an altered expression
pattern of normal structural genes, without health
related consequences for the organism. The class
Material pathological entity is subdivided into
classes: Pathological structure and Portion of
pathological body substance, on the basis of the
possession or non-possession of inherent 3D shape,
too. Pathological structures are result from an
altered expression pattern of normal structural
genes, with negative health consequences for the
organism.
The class Dependent organismal continuant is
subdivided into classes: Immaterial anatomical
continuant, Immaterial pathological continuant and
Physiological continuant. Although the existence of
immaterial anatomical and pathological spaces and
surfaces and anatomical lines and points depends
on corresponding independent continuant entities,
they are dependent continuants. Besides them
classes: Function, Physiological state and
Physiological role and classes: Malfunction,
Pathological state and Pathological role also
belongs to Dependent organismal continuant,
because their entities do not exist without
corresponding independent continuant entities.
Functions are certain sorts of potentials of
independent anatomical continuants for
engagement and participation in one or more
processes through which the potential becomes
realized. Тhe function is a continuant, since it
endures through time and exists even during those
times when it is not being realized.
Whether or not a function becomes realized
depends on the physiological or pathological state
of the associated independent anatomical
continuant. Thereat, physiological and pathological
state is a certain enduring constellation of values of
an independent continuant’s aggregate physical
properties. These physical properties are
represented in the Ontology of Physical Attributes
(OPA), which provides the values for the physical
properties of organismal continuants. Namely, the
states of these continuants can be specified in terms
of specific ranges of attribute values.
The independent continuants that participate in
a physiological or pathological process may play
different roles in the process (e.g. as agent, co-
factor, catalyst, etc.). Such a process may transform
one state into another (for example a physiological
into another physiological, or into a pathological
state).
The class Organismal occurent is subdivided
into classes: Physiological process and
Pathological process. Physiological process
courses transformations of one physiological state
665
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

into another physiological state, whereas 6. DERIVATION THE FMA-RADLEX

pathological process courses transformation of a APPLICATION ONTOLOGY
physiological state into a pathological state or one
pathological state into another pathological state.
The relative balance of these processes results
either in the maintenance of health or in the
pathogenesis of material pathological entities, and
thus in the establishment and progression of
diseases. Transformation of a pathological state into
a physiological, manifest as healing or recovery
from a disease, comes about through physiological
processes that successfully compete with and
ultimately replace pathological processes, namely
function is restored. Processes are extended not
only in time but also in space by virtue of the nature
of their participants.

5. RADLEX TERMINOLOGY

The Radiological Society of North America

(RSNA) developed a publicly available
terminology, RadLex [12], to provide a uniform
standard for all radiology-related information.
RadLex terminology is organized into a hierarchy
(Fig. 2) and subsumes over 7400 terms organized in
9 main categories or types with RadLex term as the
root. However RadLex terminology does not yet
have a principled ontological framework [14] for
these three reasons:
1) being term-oriented, RadLex currently
ignores the entities to which its terms
project;
2) the lack of a taxonomy grounded in
biomedical reality;
3) the ambiguity and mixing of relations
(such as is_a, part_of, contained_in)
represented by the links between the nodes
of the term hierarchy (Fig. 2).
In the next section, according [9] is described
how a portion of reference ontology, such as the
FMA, can be adopted to lend application ontology
in which all challenges mentioned above are
resolved.
Figure 2. RadLex hierarchy in Protégé
Figure 3: FMA-RadLex (right) derived from the
FMA (left)
Terms relating to anatomy are represented in the
RadLex terminology category Anatomic location,
which corresponds to the category Anatomical
entity, used by other disciplines of biomedicine.
This is not radiology image entity, yet the entity
that exists in the reality. Anatomic location is
therefore renamed as the FMA root term
Anatomical entity (Fig. 3). For the image findings
representing radiology images entities the separate
ontology should be created.
Application ontology from the FMA can be
derived either by:
1. Obtaining an entire copy of the FMA and
pruning the ontology down to the required
specifications - de novo construction.
2. Mapping the existing terminology project to
the FMA, carving out the ontology around
the mappings and finally incorporating the
derivatives into the existing terminology
project.
The latter method was applied in constructing
the anatomy application ontology for RadLex [9].
Hence, high level RadLex terms are first mapped to
the corresponding FMA terms, and then their
corresponding FMA super-types are imported into
the RadLex taxonomy. After that, other terms at
different levels of the RadLex tree are mapped to
the corresponding FMA terms, and then their
corresponding FMA super-types are imported into
the RadLex taxonomy super-types. In RadLex
anatomy taxonomy the highest level parents of the
imported super-types of the FMA are incorporating,
as well: Anatomical structure which subsumes 3-D
objects that have inherent shape, e.g. body, organ
system, and organ, and Immaterial anatomical
entity which encompasses types that have no mass
property, such as: anatomical space, anatomical
surface, anatomical line and anatomical point.
Hence, this conclusion can be divided: the
operation of construction the same ontology via the

UbiCC Journal - Volume 4 No. 3 666

Special Issue on ICIT 2009 Conference - Bioinformatics and Image

de novo approach, would involve a series of The hierarchical tree of the OBR ontology
deletion and addition of links (Figure 3, left) from class Pathological structure and also its subclasses:
the FMA reference ontology. For example, the is_a Subdivision of pathological organ system,
link of the class Anatomical structure is deleted Subdivision of pathological skeletal system and
from Material anatomical entity and then added Subdivision of pathological axial skeletal system,
directly to Anatomical entity. Both Physical from which all subclasses which are not relevant
anatomical entity and Material anatomical entity for the pathological domen of spine are deleted, are
are then deleted from the FMA taxonomy. Beside illustrated in Fig. 5, Fig. 6, Fig. 7 and Fig. 8.
that, FMA types representing microscopic entities
which are not relevant to radiology such as Cell,
Cardinal cell part, Biological macromolecule,
Cardinal tissue part, are also deleted from
Anatomical structure. These operations can be
carried out in all levels of the hierarchical tree.

7. DERIVATION OF THE OBR-SCOLIO

APPLICATION ONTOLOGY

In constructing the OBR-Scolio application

ontology for the pathology domain of spine
(scoliosis domen) from the OBR reference ontology
the de novo method was applied (Figure 4). All the
classes which are not relevant to the pathology
domain of spine, such as: Matherial anatomical
entity, Immaterial anatomical continuant,
Physiological continuant, Extra–Organismal
continuant, Biological occurent and also theirs
relevant subclasses are deleted from the Figure 5: Subclasses of the Pathological
hierarchical tree of the OBR reference ontology. structure class

Figure 6: Subclasses of the Subdivision of

pathological organ system class

Figure 4: OBR-Scolio application ontology

derived from the OBR reference ontology

UbiCC Journal - Volume 4 No. 3 667

Special Issue on ICIT 2009 Conference - Bioinformatics and Image
Figure 7: Subclasses of the Subdivision of
pathological skeletal system class
Figure 8: Subclasses of the Subdivision of
pathological axial skeletal system class
Ultimately, in the Fig. 9 all subclasses of the
Pathological vertebral column class are illustrated.
Figure 9: Subclasses of the Subdivision of
pathological vertebral column class
ACKNOWLEDGEMENT
By vertical integration of the FMA reference
ontology with the BFO top-level ontology the
process of horizontal integration of the two
reference ontologies: PRO and PathRO is
supported, forming accordingly the new reference
ontology OBR, which range over the domains of
anatomy, physiology and pathology. This ontology
can be successfully applied in development of the
UbiCC Journal - Volume 4 No. 3
OBR-Scolio application ontology in the pathology
domain of spine (scoliosis), which is one of the
many objectives in the realization of the project
named: “Ontological modeling in bioengineering 1 ”
in the domain of orthopedics and physical
medicine.
8. REFERENCES
[1] Cornelius Rosse, MD, DSc, Anand Kumar,
MD,PhD, Jose LV Mejino Jr, MD, Daniel L Cook,
MD, PhD, Landon T Detwiler, Barry Smith, PhD:
A Strategy for Improving and Integrating
Biomedical Ontologies, AMIA Annu Symp Proc.
(2005), pp. 639-643.
[2] http://www.loucnr.it/DOLCE.html
[3] Grenon P, Smith B, Goldberg L: Biodynamic
ontology: applying BFO in the biomedical domain,
In DM Pisanelli (ed.), Ontologies in Medicine,
Amsterdam: IOS Press, 2004,20-38.
[4] Open Biomedical Ontologies:
http://obo.sourceforge.net/
[5] Rosse C, Mejino JLV Jr.: A reference ontology
for biomedical informatics: the Foundational Model
of Anatomy, J Biomed Inform, 2003 Dec;
36(6):478-500.
[6] Smith B, Köhler J, Kumar A.: On the
application of formal principles to life science data:
A case study in the Gene Ontology, DILS 2004:
Data Integration in the Life Sciences. 2004; 124-
139.
[7] Zhang S, Bodenreider O.: Law and order:
Assessing and enforcing compliance with
ontological modelling principles, Computers in
Biology and Medicine 2005: in press.
[8] Cook DL, Mejino JLV, Rosse C.: Evolution of a
Foundational Model of Physiology: Symbolic
representation for functional bioinformatics,
Medinfo 2004; 2004:336-340.
[9] Jose L.V. Mejino Jr, Daniel L. Rubin, and
James F. Brinkley: FMA-RadLex: An Application
Ontology of Radiological Anatomy derived from
the Foundational Model of Anatomy Reference
Ontology, Proceedings AMIA Symposium 2008:
Page 465-469.
[10] Rosse C, Mejino JLV 2007: The Foundational
Model of Anatomy Ontology, in: Burger A,
Davidson D, Baldock R. (eds.), Anatomy
1
“Ontological modeling in bioengineering”, Project
funded by national Ministry of science, Faculty of
mechanical engineering, University of Kragujevac,
Serbia (2008-2010)
668
Special Issue on ICIT 2009 Conference - Bioinformatics and Image

Ontologies for Bioinformatics: Principles and [13] Rubin DL 2007: Creating and curating a
Practice, pp 59-117, New York: Springer. terminology for Radiology: Ontology Modeling and
Analysis, J Digit Imaging.
[11] Online аvailable at:
http://www.rsna.org/radlex. [14] Marwede D, Fielding M and Kahn T. 2007
RadiO: A Prototype Application Ontology for
[12] Langlotz CP: RadLex 2006. a new method for Radiology Reporting Tasks, Proc AMIA 2007,
indexing online educational materials, Chicago. IL, pp 513-517.
Radiographics 26:1595–1597.
[15] FMA Online аvailable at:
http://fma.biostr.washington.edu.

UbiCC Journal - Volume 4 No. 3 669