EDITORIAL BOARD
EDITOR-IN-CHIEF
Professor IRENA ROTERMAN-KONIECZNA
Medical College  Jagiellonian University, Krakow, st. Lazarza 16
HONORARY ADVISOR
Professor RYSZARD TADEUSIEWICZ
AGH  University of Science and Technology
Professor JAN TRĄBKA
Medical College  Jagiellonian University
MANAGING EDITORS
BIOCYBERNETICS  Professor PIOTR AUGUSTYNIAK
AGH  University of Science and Technology, Krakow, al. Mickiewicza 30
BIOLOGICAL DISCIPLINES  Professor LESZEK KONIECZNY
Medical College  Jagiellonian University, Krakow, Kopernika 7
MEDICINE  Professor KALINA KAWECKA-JASZCZ
Medical College  Jagiellonian University, Krakow, Pradnicka 80
PHARMACOLOGY  Professor STEFAN CHA
OPICKI
Medical College  Jagiellonian University, Krakow, Grzegórzecka 16
PHYSICS  Professor STANISA
AW MICEK
Faculty of Physics  Jagiellonian University, Krakow, Reymonta 4
MEDICAL INFORMATICS AND COMPUTER SCIENCE  Professor MAREK OGIELA
AGH  University of Science and Technology, Krakow, al. Mickiewicza 30
TELEMEDICINE  Professor ROBERT RUDOWSKI
Medical Academy, Warsaw, Banacha 1a
 Dr SYBILLA STANISA
AWSKA-KLOC
LAW (and contacts with business)  Dr SYBILLA STANISA
AWSKA-KLOC
Law Faculty  Jagiellonian University, Krakow, Kanoniczna 14, Institute of Intellectual Property Law
Law Faculty  Jagiellonian University, Krakow, Kanonicza 4
ASSOCIATE EDITORS
Medical College  Jagiellonian University, Krakow, Kopernika 7e
EDITOR-IN-CHARGE  PIOTR WALECKI
E-LEARNING (project-related)  ANDRZEJ KONONOWICZ
E-LEARNING (general)  WIESA
AW PYRCZAK
DISCUSSION FORUMS  WOJCIECH LASOŃ
ENCRYPTION  KRZYSZTOF SARAPATA
TECHNICAL SUPPORT
Medical College  Jagiellonian University, Krakow, st. Lazarza 16
ZDZISA
AW WIR
NIOWSKI  in charge
WOJCIECH ZIAJKA
ANNA ZAREMBA-R
MIETAŃSKA
PoIish Ministry of Science and Higher Education journaI rating: 4.000
3.000
Sustaining institution: Ministry of Science and Higher Education
Edition: 300 copies
� COPYRIGHT BY INDIVIDUAL AUTHORS AND MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
ISSN 1895-9091 (print version)
ISSN 1896-530X (electronic version)
http://www.bams.cm-uj.krakow.pl
Contents
OPENING PAPER
5 Use of Personal Data and Patients Image in Scientific Publications
Paweł Markiewicz
BIOINFORMATICS
9 GenFoSS  java application for presentation of coding stretches of Never Born Proteins
Monika Piwowar, Ewa Matczyńska, Piotr Ochlawski, Mateusz Kosibowicz, Damian Kość, Michał Swatowski,
Piotr Więcek, Tomasz Szepieniec
MODELING AND SIMULATION
19 Simulation of ratio of old to young people in countries like Poland
Dietrich Stauffer and Krzysztof Kułakowski
BIOMEDICAL ENGINEERING
25 Magnetocardiography  measurement conduction, data formats and diagnostic applications
Kamila Baron-Pałucka
PATTERN RECOGNITION
35 An algorithm for detecting lesions in CBF and CBV perfusion maps
Tomasz Hachaj
43 The registration and atlas construction of noisy brain computer tomography images based on free
form deformation technique
Tomasz Hachaj
51 Image processing application for enhancement of medical diagnostic features
Joanna Jaworek, Eliasz Kańtoch
57 Digital Skeletonization as a Probe of Correlation between Sutural Bones and Diffusion Limited
Agreggation Clusters
Janusz Skrzat, Jerzy Walocha
TELEMEDICINE
63 Learning-by-e-teaching: Experience from involving students in preparation of e-learning materials
Andrzej A. Kononowicz, Aleksandra J. Stachoń, Anna Romanowska-Pawliczek, Piotr Obtułowicz, Wiesław
Pyrczak
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 5-8
USE OF PERSONAL DATA AND PATIENTS IMAGE IN SCIENTIFIC
PUBLICATIONS
PAWEA
MARKIEWICZ
Student of the Law Faculty of the Jagiellonian University, TBSP
(Association of Law Students Library of the Jagiellonian University), Bracka 12/302, 31-005 Krakow
Doctors who present in scientifi c and medical pub- possibility to identify a given patient by the reader.
lications clinical cases in which particular patients Thus the descriptions, photos, etc. published must
are indicated may face the problem of illegal dis- make it possible to disclose the person s identity. It
closure of patients personal data and consequently is suffi cient that the person is recognized by a cer-
the problem of civil and criminal liability. The pur- tain group of people, e.g. by his or her colleagues.
pose of this article is to present legal regulations The possibility of identifi cation may obviously result
limiting freedom of the authors of medical publica- from a direct disclosure of the name and surname
tions in respect of reference to examples from phy- of the person whose case has been described, yet
sician s practice in the eye of the regulations of the naturally such situations are very rare. Much more
Civil Code1, Act on the Protection of Personal Da- common is the possibility of identifi cation of a per-
ta2 and Act on Copyright and Neighboring Rights3. son through the photos attached to such a publica-
Such information as: case history, photos showing tion, where apart from pathological changes some
pathological changes, information about the age, other characteristic features of that person can be
etc. may lead to violation of personal rights of the seen: for example a tattoo, a rare body deformation,
patients they relate to, or violation of rights of their physical appearance etc. The person can also be
relatives and friends. In the person s lifetime authors recognized through the description of the situation
have to obtain his or her consent for such publica- which lead to the injury mentioned in the text, espe-
tion. The issue of consent takes a different angle af- cially if it was a rare incident and additional informa-
ter patient s death and further remarks are devoted tion indicating the place of the incident took place.
to this problem. The possibility of linking a theoretically anonymous
patient presented in the publication with a specific
person may also result from the unique nature of
Protection of personal data of the the disease or from the fact that this person is well
known. However, the remarks concerning the ele-
deceased patients
ments of the publication which can lead to identifica-
Act on the Protection of Personal Data constitutes tion of a patient should be considered in the eye of
fundamental legal act governing the use of other Article 6.3 of the above mentioned Act, which states
persons personal data. Pursuant to the wording of that:  A piece of information shall not be regarded as
art. 6.1. of the Act,  ... personal data shall mean any identifying where the identifi cation requires an un-
information relating to an identifi ed or identifiable reasonable amount of time, cost and manpower. 4
natural person. Thus, in the eye of this act, in order Thus, identifi cation of a patient on the basis of data
to denote a medical publication presenting a partic- from a publication must be possible with the use of
ular case as inadmissible, there must be objective the means which are generally accessible and com-
mensurate.
Here one should answer the question whether
1
Act of 23rd April 1964: the Civil Code (Journal of Laws
the Act under consideration is to be applied also
No. 16, item 93 as amended).
2
in case of deceased people. The Act on Personal
Act of 29th August 1997 on the Protection of Personal
Data (i.e. Journal of Laws of 2002, No. 101, item 926 as Data Protection contains two types of provisions:
amended).
3 4
Act of 4th February 1994 on Copyright and Neighbor- Act of 29th August 1997 on the Protection of Personal
ing Rights (i.e. Journal of Laws of 2006, No. 90, item 631, Data (i.e. Journal of Laws of 2002, No. 101, item 929 as
as amended). amended).
OPENING PAPER
Use of Personal Data and Patients Image in Scientific Publications
6
those related to public law aspect and those related the persons mentioned in the publication, is the im-
strictly to private law aspect. Private law aspect co- age protection introduced in article 81 of the Act on
vers all regulations addressed to natural and legal Copyright and Neighboring Rights. Image means
persons, whereas public law aspect marks a certain  a visual presentation of a person i.e. typical physi-
standard concerning personal data processing and cal features of a person, which give the idea about
is addressed to database administrators. the person s appearance 6. For example a photo of
It is undisputable that at the moment of a patient s a patient printed in a publication can fall into this
death provisions of private law character cease to defi nition. It must be noted, however, that a photo,
apply, for example the provisions requiring consent, which makes it possible to identify a person, yet
as such consent obviously cannot be granted. An- does not illustrate the likeness (i.e. does not give
other question should be asked however: whether the idea of one s looks  pictures of particular parts
after a patient s death protection resulting from pub- of someone s body, pictures of tissues etc.) cannot
lic law provision still applies, i.e. the duties subject be considered to constitute a person s image.
to public authority control, regardless of the activity In such a case the subject of protection is the
of the person directly involved. exclusiveness of image management  i.e. the right
In case of a patient s death, regulations impos- to decide about its publication. This right evidently
ing duties on third parties or requiring certain behav- does not refer to the deceased patient. Emotional
ior standards from them theoretically still apply. For sphere connected with publication of the image may
example article 26 paragraph 1 concerns database be protected pursuant to Civil Code regulations on
administrators, requiring from them all their possible the protection of personal interest (article 23 and 24
diligence in data processing in order to protect inter- of the Civil Code). Law infringement requires nega-
ests of the people whose data is being processed. tive psychological experience of the patient caused
However, possibility to apply this act to deceased by the publication of the image. Protection resulting
people should be assessed negatively. The letter of from articles 23 and 24 of the Civil Code does not
the Inspector General for the Protection of the Per- allow for creation of legal fi ction, i.e. existence of
sonal Data (this official is generally known in Poland personal interest of a subject who is no longer alive.
under the acronym GIODO) of 12th February 2002 Article 24 of the Civil Code, stating that the meas-
(symbol GI  DP  024 /145/01/427), where he not- ures of personal rights protection may be used by
ed that the act does not concern deceased people,  a person whose personal right is threatened... , evi-
confi rms that such an opinion is right. It is evident dently denies the possibility to apply this protection
that the Act on the Protection of Personal Data in in relation to the deceased. The foregoing opinion is
any case does not affect the limitations of the per- absolutely undisputable in the eye of civil law.
mitted publication of data in the situation of patient s
death.5 It must be also added that the Act will be cer-
tainly applied in relation to data of deceased people Protection of the rights of third parties
when such data somehow also relates to the third
parties, for example to the relatives. A situation may From the hereto presented institutions protecting
occur when, for instance, the publication contains the rights of persons being  the subject of publi-
the description of the course of a hereditary disease cation one may draw a conclusion, that on the ba-
of the deceased patient. Then, it would be possi- sis of Polish law there is a rule under which when
ble to consider such publication as law infringement a publication describes cases of the deceased peo-
pursuant to the Act on the Protection of Personal ple there are no limitations in respect of disclosure
Data but only in relation to the issue (offspring) and of information relating to these people and their
not to the deceased patient. image. However, this conclusion would be false.
To sum up: from the point of view of the Act on The scope of permissible data disclosure concern-
the Protection of Personal Data placing in a scien- ing a deceased patient is specifi ed not by protec-
tifi c publication information such as: the deceased tion of a deceased patient s rights but by protection
patient s case history, his/her age, blood group, of third parties rights (relatives of the deceased).
medicines taken, addictions etc. in no case infring- Pursuant to the legal structure adopted in articles
es duties of data administrator. 23 and 24 of the Civil Code, at the moment of the
patient s death his/her rights ensuring protection of
non-pecuniary interest expire. One does not find an
Image protection of the deceased analogical institution to the one related to copyright,
based on  transfering author s moral rights from
patients
the deceased to a determined group of relatives,
with the possibility to use them as own rights. As the
Another legal institution determining the limits of wording of article 23, sentence 1 ( personal interest
permissible publication with regard to the rights of
6
J. Błeszyński, Glosa do wyroku SN z 27.02.2003 r.,
5
Compare J. Barta, P. Fajgielski, R. Markiewicz, Och- [Comment to the Decision of the Supreme Court of Feb-
rona Danych Osobowych  Komentarz, Kraków 2004, ruary 27, 2003] IV CKN 1819/00, OSP 2004, No. 6, item.
p. 376. 75, p. 320.
OPENING PAPER
Use of Personal Data and Patients Image in Scientific Publications 7
of a man in particular... ) indisputably settles an ex- spect of the cult of memory of the deceased per-
emplary nature of the list of personal interest subject son. At the same time entering in this manner into
to protection, both the doctrine and the jurisdiction the protected sphere will be illegal. In this context
distinguish personal right of the deceased person s one should refer to the Doctor and Dentist Profes-
relatives to protection of their feelings connected sion Act7, which specifi es what is legal and illegal
with the memory and respect of the deceased. Per- in the behavior of these professionals. Paragraph
sonal interest of those people is defined as  the cult 1 of article 40 of this Act states that:  The physician
of memory of the deceased person . Therefore the is obliged to keep confi dential any information re-
problem of publication of the information about the lating to the patient obtained in connection with his
deceased person is looked at through the prism of practice . Paragraph 2 of the article 42 enumerates
the feelings of the deceased person s relatives and situations when despite disclosure of information no
not through the prism of the feelings of the subject infringement of medical confi dentiality occurs. For
who is no longer alive. example the cases where disclosure of informa-
The analysis of court decisions leads to the tion is  necessary for practical teaching of medical
conclusion that the subject of protection is men- profession (item 6) or  it is necessary for medical
tal peace connected with the experience of going research (item 7). When a publication contains
through the loss of a close person, thus the subject information concerning a deceased patient, and
of protection is the emotional sphere related to the such information infringe personal rights of his/her
cult of memory of the deceased person. Keeping relatives, the illegality of this infringement shall be
this in mind I shall try to present logical reasoning excluded if a prerequisite of necessity of the publi-
which should be adopted in the process of exami- cation occurs, in the eye of scientifi c purposes and
nation whether personal interest in question has the necessity of practical teaching of the profession.
been violated. The fi rst step is to establish if the Of course the word  necessary used in the act is
elements included in the publication enter into the not clear and determination of existence of such
sphere protected by the law. It is therefore neces- prerequisite will be possible only in a specific case,
sary to answer the question whether publication of based on the knowledge of experts in a given field
the deceased patient s case history leads to viola- of medicine. It must be noted that the prerequisite
tion of the aforementioned mental peace of the rela- of  necessity also refers to the manner of publica-
tives. It is obviously very difficult to determine it in an tion of the information  not only there has to be the
abstract manner, without reference to any specific necessity to publish the information, justifi ed by the
case. It seems, however, that in the majority of cas- aspects provided in article 42, items 6 and 7 of the
es the answer to this question will be positive, no Doctor and Dentist Profession Act8, but also the ap-
matter whether the publication contains drastic de- propriate form and content of the publication should
scriptions, the history of a disease commonly con- consider possibly slightest disturbance for the peo-
sidered as embarrassing or it is only a dry relation ple whose personal interest are infringed. To sum
which does not arouse any emotions in the reader. up, article 42, items 6 and 7 of the aforementioned
It results from the fact that it is suffi cient for viola- act specify circumstances excluding illegality of the
tion of the cult of memory of a deceased person to actions taken by the author of publication in which
infringe the aforementioned mental peace and that a particular case is being presented. Publication of
will occur at the moment of publication of the infor- data necessary for scientifi c purposes and for ac-
mation evoking the memories of the deceased. In quiring practical knowledge does not always have
order to talk about the infringement of a personal to be combined with publication of data allowing
right it is necessary that the discussed infringement for the identification of the patient.
of mental peace leads to negative mental experi- There is also another occasion excluding illegal-
ence in a subject of the right to the cult of memory ity of physician s publication of information about
of the deceased person, i.e. in a relative. a patient which was acquired in the course of medi-
Determination of such negative experience can- cal practice. It is the situation when a patient or his
not be made subjectively. It is not enough to rely only statutory representative expresses his/her consent
on a declaration of a person claiming protection of to the use of such information. Such consent may
his/her rights as in case of hypersensitive persons also be expressed by the deceased patient s rela-
the assigned protection would go absolutely too far. tive after the patient s death but its effectiveness will
It is therefore necessary to determine the relation be different. It will exclude illegality of physician s
between the infringement and the mental experi- actions only in relation to persons who expressed
ence, which would be commensurate to an average
person, i.e. it would be objective.
7
Ustawa z dnia 5 grudnia 1996 o zawodach lekarza
Article 23 of the Civil Code determines that pro-
i lekarza dentysty [Doctor and Dentist Profession Act of
tection of personal interest is assigned in case of
December 5, 1996], i.e. Journal of Laws 2008, No. 136,
illegal infringement of these rights. Thus infringe- item. 857.
8
Ustawa z dnia 5 grudnia 1996 o zawodach lekarza
ment of a personal interest takes place in the event
i lekarza dentysty [Doctor and Dentist Profession Act of
of entering into the protected sphere and causing
December 5, 1996], Journal of Laws 2005, No. 226, item.
negative feelings of the subject of the right in re-
1943.
OPENING PAPER
Use of Personal Data and Patients Image in Scientific Publications
8
such a consent. In consequence, so as the publi- Conclusions
cation would not be considered illegal the author
should obtain consent of all the people authorized To sum up the foregoing we may talk about imper-
to claim protection of their rights in connection with missibility of a publication presenting clinical cases
the protection of the cult of memory of the deceased concerning deceased people only in the context of
person. Determining who actually belongs of the infringement of rights of the persons related to the
group of those people will be possible only within deceased. In order to state that the personal inter-
the context of a specifi c case. The discussed right est has been violated and such violation justifies the
is the personal interest assigned to the relatives of use of measures provided in article 24 of the Civil
the deceased. Such determination of the subject of Code (such as compensation, submission of an ap-
this right does not settle exactly who can enjoy its propriate statement, forbidding infringement) it is
protection. It seems, however, that such right can necessary to establish that such activity, i.e. publi-
be assigned not only to the relatives or members of cation of any of the aforementioned elements such
the family of the deceased person but also to peo- as: case history, photos of disease symptoms etc.,
ple emotionally attached to the deceased. The task lead to entering a sphere which is legally protected
of the court will be to determine in relation to whom by this right and therefore it has disturbed mental
it is possible the existence of a relation justifying peace of a relative. At the same time such an in-
examination whether the aforementioned personal fringement must cause negative mental experience.
interest has not been violated. The last prerequisite that needs to be established,
Obviously all refl ections presented in this paper in order to consider the publication as impermissi-
refer to the situation when in a scientific research ble, is the illegality of the author s activity, assessed
medical publication a case of a deceased patient is on the basis of the provisions of the Doctor and
presented and the publication infringes the rights of Dentist Profession Act9 obliging a physician to keep
the patient s relatives in relation to protection of the confi dential any information relating to the patient
worship of a deceased person, assuming that the which was obtained in connection with physician s
relatives can recognize a person to whom they medical practice.
are emotionally attached.
9
Ustawa z dnia 5 grudnia 1996 o zawodach lekarza
i lekarza dentysty [Doctor and Dentist Profession Act of
December 5, 1996], Journal of Laws 2005, No. 226, item.
1943.
OPENING PAPER
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 9-17
GENFOSS  JAVA APPLICATION FOR PRESENTATION OF CODING
STRETCHES OF NEVER BORN PROTEINS
MONIKA PIWOWAR1, EWA MATCZYC
SKA1, PIOTR OCHLAWSKI2, MATEUSZ KOSIBOWICZ2,
DAMIAN KOŚĆ2, MICHAA
SWATOWSKI2, PIOTR WI
CEK2, TOMASZ SZEPIENIEC3
1
Department of Bioinformatics and Telemedicine, Jagiellonian University, Collegium Medicum,
Sw. Anny 12, 31-008 Krakow, Poland
2
Astronomy and Applied Informatics, Jagiellonian University, Faculty of Physics, Reymonta 4,
Krakow, Poland
3
Academic Computer Center CYFRONET, Nawojki 11, 30-950 Krakow, Poland
Abstract: Application of genomes analysis results with respect to project of  never born protein (NBP) is presented in this
paper. GenFoSS application concerns visualisation of genomic sequence stretches that could encode proteins that are not
observed in the Nature. The sequences of these proteins were generated randomly aimed to recognize the proteins of potential
pharmacological activity. Identification of sequences similar to these of NBP can be viewed by GenFoSS application. GenFoSS
gives information about sequence frequencies, percent identity, percent of similarity of alignments and E-value for aligned
sequences. Comparison between two chromosomes in common organism and between different genomes as well can be
done by these applications. There are also possibilities to check codon frequencies of nucleotide sequences that were found
as a stretch of NBP.
Key words: never born proteins, genomes analysis, nucleotide sequences distribution
Introduction
The structure prediction of proteins of potential pharmacologi- cent of identity in aligned sequences by using BLAST program
cal application was one of the aims that EUChinaGrid project with proper matrices (Fig. 1).
was focused on [3]. Materials to reach it were sequences of Identified stretches allow comparison of aligned sequenc-
70 amino acids in polypeptide (107 of them are classified as es distribution in both compared chromosomes in one genome
 never born protein ) which were used to produce their three di- as well as between genomes of different level of evolutionary
mensional structures [4]. Never born proteins were defined as tree. The conclusions from such analysis may recognize the
proteins that were not identified in known protein (and genetic) sequences composition (high or low probability to find random
databases [7,9]. Finding of traces of NBP among completely sequences in genomes can suggest less or more  planed
sequenced as well as in progress DNA sequences was of spe- sequence composition). GenFoSS application allow check
cial interest [1]. It is expected to identify stretches of genomic codon frequencies of nucleotide sequences that were found as
sequences of potential biological function not identified (so far) stretch of NBP.
in the Nature.
The genFoSS tool was generated to visualize the results of
genomic analysis. Material and Methods
Identification of sequences similar to NBP in particular
genomes gives information about dispersion sequences than
could encode proteins that are not existing in nature currently Proteins data:
but could have existed in nature as proteins formerly (in evolu-
tion time) and could have had a particular function in the past. Never born proteins were obtained from Roma Tre University
Searching of traces of NBP was done on different level of ac- proteomic group [7]. Research material (besides DNA) was
cordance. There were accessed percent of similarity and per- proteins written in Fasta format that were randomly generat-
Bioinformatics
genFoSS  java application for presentation of coding stretches of Never Born Proteins
10
Fig. 1. genFoSS input. Data to genFoSS application come form BLAST results. DNA AA: DNA Amino Acids
(DNA translated to amino acid sequences), RS: Random sequences
ed. Such generated database represented  NBP . Number of applied to translate human genetic information to amino acids
polypeptides that was classified as proteins not existing in na- sequences [6].
ture was 104. Selected NBPs that were found in genetic mate- Next step was scanning translated genome sequence by
rial are separated for further description and characterization. BLAST  The Basic Local Alignment Search Tool for compar-
ing gene and protein sequences against others [5] (fig. 2).
Software that was used:
Results
BAST  Basic Local Alignment Search Tool [5]
GenFoSS is a 3-tier database application that used Java serv-
lets and the Java Database Connection (JDBC). User interacts
Technology used for creating application: with the application by choosing the concrete values from the
query form. Once the form is submitted then the Java servlet
MySQL  data for genFoSS were collected in database cre- uses JDBC to find the information in a database. Afterwards,
ated in MySQL technology. a Java object is created by the servlet and returned to the user
Java  genFoSS is a tool created to research some of ge- application using object serialization.
netic parameters in known and already stored in databases The application is partitioned into three tiers: user inter-
genomes. The technology we used is simply Java Web Start, face layer(1), the business rules layer(2) and the data storage
because of its portability and stability. This is why the only thing layer(3) (fig. 4):
you have to had installed before running it is JRE  Java Run-
time Environment. If you don t have it installed just go in your
web browser to address: http://www.java.com/en/download/
manual.jsp and download suitable version. All the information
about installation of JRE can also be found there.
Genetic material
DNA sequences were taken for analysis from National Center
of Biotechnology Information [8]. Most of them were repre-
sented by nucleotide sequences that have not been noticed
information about stretches encoding known proteins. Fig. 3: Tiers of genFoSS application user interface layer(1), the
business rules layer(2) and the data store layer(3):
Searching method
" fi rst tier is a stand alone application. In the fi rst phase of
The first step in finding similarities was translating genome the application, Swing components were used for user-
sequences to amino acids sequences. Tree reading frames input and displaying the database query results.
were taken into consideration. The standard genetic code was " second tier of the application is implemented with a Web
server capable of executing Java servlets. The Java serv-
Bioinformatics
genFoSS  java application for presentation of coding stretches of Never Born Proteins 11
Fig. 2. Diagram of job execution.
let harnesses the power of JDBC to access the database sults are sent back to genFoSS, which presents them in a sim-
to retrieve information as needed. ple and accessible way.
" third tier is composed of our database server. The data-
base server stores the information that is used by the ap-
plication. Thanks to the JDBC API, the servlet can access Stand alone application  servlet communication
the database in a portable fashion by using the SQL call-
level interface. Since servlets support the HTTP interface, there can be
communication with the servlet over HTTP socket connec-
tions. The application simply has to open a connection to the
Database structure specified servlet URL. Once this connection is made, then the
application can get an output stream or input stream on the
Databases for storing data used by genFoSS application were servlet. The application can send data to the servlet by send-
created in a relational database management system MySQL ing a POST method. To send a POST method to a servlet, the
because of its usefulness in storing and processing large application can use the java.net.URLConnection class. It must
amounts of data (fig. 3). BLAST output is parsed and stored in also inform the URL connection that data will send over the
this database. genFoSS connects with the database through output stream. The POST method is powerful because it s pos-
Java Servlet (described below). All needed data is quickly sible to send any form of data (plain text, binary, serialized ob-
gathered and calculated thanks to SQL queries. Obtained re- ject, etc) using it. All that must be done is set the content type
Bioinformatics
genFoSS  java application for presentation of coding stretches of Never Born Proteins
12
Fig. 4. MySQL database (GENOMIC database) schema.
in the HTTP request header  in this case it will be the serial- Objects can also be serialized to any output stream. This
ized object  application/x-java-serialized-object . even includes an output stream based on a socket connection.
The code fragment: It can serialize an object over a socket output stream. In order
for a Java object to be serializable, its class must implement
// connect to the servlet
the java.io.Serializable interface. The java.io.Serializable inter-
URL testServlet = new URL( http://www.sample.com/servlet/Serv-
face is simply a tag for the Java Virtual Machine. It can be cre-
let );
ated a custom class as follows:
URLConnection servletConnection = testServlet.openConnection();
class Example implements java.io.Serializable
// inform the connection that it will send output and accept input
{
servletConnection.setDoInput(true);
// normal declaration of data members,
servletConnection.setDoOutput(true);
// constructors and methods
}
// There can not be used a cached version of URL connection.
servletConnection.setUseCaches (false);
There were four classes created: Experiments, Organisms,
servletConnection.setDefaultUseCaches (false);
Chromosomes and Codons. Each of them implements the
java.io.Serializable interface and has two constructors  one
// Specify the content type that it will send binary data
for creating object that stores information from query form and
servletConnection.setRequestProperty( Content-Type ,  one for creating object that will be passed to client application
vorite mime type> );
with data from database.
// get input and output streams on servlet
Options selection in genFoSS
&
When genFoSS starts, welcome screen disappears (Fig. 5).
// send data to the servlet
& A second screen (by Query Form) makes possible to choose
the experiment of interest.
Passing an object between the application and the servlet By defining it you can select the type of organism speci-
is done by Java object serialization. Java 1.1 introduced object fied by that experiment. To select the organism simply click on
serialization, which allows an object to be flattened and saved the picture at context menu. It shows all available organisms
as a binary file. The values of the data members are saved in the selected experiment. The next step is selecting required
so in fact, the state of the object is persisted or serialized. At parameter of organism s genome (%identity, similarity, E-value
a later time, the object can be loaded or deserialized from the or Length) and its range (From and To) with its interval (how
binary file with the values of its data members intact. Object many  boxes creates specified range). When everything is se-
serialization is fascinating in that it frees the developer from lected  press Submit button (fig. 6).
low-level details of saving and restoring the object.
Bioinformatics
genFoSS  java application for presentation of coding stretches of Never Born Proteins 13
Fig. 5. Welcome and Query Form screen of genFoSS
After selecting parameters under consideration the re- New buttons on context menu allow specification of se-
sults are available. There are two buttons at the top of context lected chromosomes. Buttons in  Chromosomes are named
menu: Histograms and Comparison. It allows few aspects of according to chromosomes names. To show/hide histogram
results available. the appropriate button should be pressed.  + Button shows
all chromosomes,  - button hides them all.  S button shows/
hides summary histogram. Interested details of single chart ap-
 Histograms pear after clicking on it (fig. 8).
Clicking the required bar (number of them is defined by in-
Histograms show distribution of parameters for complete set of terval size) allows detailed view. It shows how many cases are
organism s chromosomes. There is also a summary distribu- included inside this sub-range.
tion of theme (fig. 7).
Fig. 6. Filled form with Homo Sapiens organism selected
Bioinformatics
genFoSS  java application for presentation of coding stretches of Never Born Proteins
14
Fig. 7. Histograms of E-Value for all selected Homo Sapiens chromosomes
Fig. 8. Detailed E-value (in range 0-10 and with interval 30) histogram of Homo Sapiens chromosome 1
named buttons in chromosomes section at context menu bar
 Comparison (fig. 9).
To change colors on the chart click on the legend box at
Comparing a few distributions of defined parameter at one the bottom of the tab and select other color from palette. Boxes
chart is possible by clicking  Compare button at the context have the same colors as corresponding lines at chart and they
menu followed by the appearance of one, big and empty histo- have names of the chromosomes.
gram. To add/remove data set to it just use chromosomes-like
Bioinformatics
genFoSS  java application for presentation of coding stretches of Never Born Proteins 15
Fig. 9. Comparator of selected chromosomes
Fig. 10. Codons for Homo Sapiens chromosomes 1,10,11,17 and 20
 Codon statistics tailed chart tab. To see summary codons information just use
histograms or comparator. To have all required chromosomes
At histograms, or comparator, there is a  Codons button avail- visible just press Codons button and result for them will appear
able at the context menu bar. It shows new tab with quantity (fig. 10).
information about selected chromosomes codons. There is The darker color of codon, the bigger number of its type
possibility to obtain codon statistics by pressing on active de- found in selected chromosomes. Additional options are avail-
Bioinformatics
genFoSS  java application for presentation of coding stretches of Never Born Proteins
16
able by using right mouse button on the Codons tab. Colors of nected with and data stored in database so everyone can
codons can be changed to grayscale. A set minimal value of check results. Application gives information about parameters
shown codons (all of codons with lower value are hidden) can like: sequence frequencies, percent identity, percent similarity
be marked. Details about codon quantities in codon s boxes of alignments and E-value. Application allows comparing dif-
can also be viewed. The vertical bar at the right side of codons ferent portion of genomes and checking codon frequencies of
map shows the scale and maximum value. When you click the nucleotide sequences that were found as stretch of NBP. What
codon s box, the black arrow will show where selected quantity is very important about genFoSS is it s available for free. There
is situated on axis. is a plan of creating web service by which particular tasks will
have to be sent on grid infrastructure or local clusters struc-
ture. Resources in size of about 300 CPU enabled us to com-
 Saving Results plete all computations in a couple of hours or days depending
on amount of genomic data (Having the framework prepared,
To save results for later purpose press  Save button (on the we were able to carry out the whole experiment in 38 hours.
bottom of context menu bar). All entries have  more hyperlink, Average usage of resources was 126 CPUs). Additionally, the
which can help to decide if the tab is necessary or not (fig. 11). results of our work was a framework, that would be used eas-
When all tabs to export are selected, choose the format ily for any further computations in which BLAST package and
from combo box pressing Save to file button. Then information selection of gather DNA material used.
about file destination will appear.
References
Conclusion and Perspectives
[1]. Piwowar M., Szepienie T., Roterman-Konieczna I.: Massive
GenFoSS application helps to analyse and visualise stretches identifi cation of similarities in DNA materials organized in
of genomic regions that can encode NBP. Application is easy Grid environment. BAMS, 5, (2007).
to use with standard and context menu. GenFoSS is still con- [2]. Malawski M., Szepienie T., Kochanczyk M., Piwowar M.,
Roterman-Konieczna I.: An approach to protein folding on
the grid EUChinaGrid experience. BAMS, 5, (2007).
[3]. Roterman-Konieczna I., Kochanczyk M., Malawski M., Pi-
wowar M., Szepienie T.: The Quest for Pharmacology Ac-
tive  Never Born Proteins within EUChinaGRID Project,
Grid Workshop; Kraków; October 15-18 2006.
[4]. Brylinski M., Konieczny L., Roterman I.: Fuzzy-oil-drop
hydrophobic force fi eld  a model to represent late-stage
folding (in silico) of lysozyme. J Biomol Struct Dyn., 23,
519-528, (2006).
[5]. Altschul S.F., Gish W., Miller W., Myers E.W. & Lipman
D.J.: Basic local alignment search tool. J. Mol. Biol. 215:
403-410, (1990).
[6]. The Genetic Codes: http://www.ncbi.nlm.nih.gov/Taxono-
my/Utils/wprintgc.cgi?mode=c
[7]. Chiarabelli C., Vrijbloed J.W., De Lucrezia D., Thomas
R.M., Stano P., Polticelli F., Ottone T., Papa E., Luisi P.L.:
Investigation of de novo Totally Random Biosequences.
Part II. On the Folding Frequency in a Totally Random
Library of de novo Proteins Obtained by Phage Display.
Chemistry & Biodiversity, 3, 840-859, (2006).
[8]. National Center of Biotechnology Information (ftp.ncbi.nih.
gov).
[9]. Chiarabelli C., Vrijbloed J. W., Thomas R. M., Luisi P. L.:
Investigation of de novo totally random biosequences, part
I: A general method for in vitro selection of folded domains
from a random polypeptide library displayed on phage.
Chem Biodivers 3, 827-839, (2006).
Fig. 11. Export results from first and third tab
Bioinformatics
genFoSS  java application for presentation of coding stretches of Never Born Proteins 17
GENETICS COMPUTER SCIENCE
STATISTICS MEDICINE
Bioinformatics
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 19-23
SIMULATION OF RATIO OF OLD TO YOUNG PEOPLE IN COUNTRIES
LIKE POLAND
DIETRICH STAUFFER1 AND KRZYSZTOF KUA
AKOWSKI2
Faculty of Physics and Applied Computer Science, AGH University of Science and Technology,
Mickiewicza 30, PL-30059 Krakow, Poland
1
On leave from Institute of Theoretical Physics, Cologne University, D-50923 KQ
ln, Germany.
Visit supported by COST P10. E-mail: dstauff@thp.uni-koeln.de
2
E-mail: kulakowski@novell.ftj.agh.edu.pl
Abstract: Countries like Poland with a recent sharp drop in birth rates still have some time to prepare for the problems of an
ageing society. After the year 2030 they can become increasingly serious.
Retirement gets difficult to support if everybody lives lon- the working population will not be much higher than it was in
ger, the birth rates go down, and retirement age and immigra- the past. Thus one has some years time to think, discuss and
tion/emigration remain constant. This demographic change agree on how to solve the future problems of demographic
[1] influences the ratio of people in retirement age to those of change. The upper curve there shows the effects of a net
working age. These effects have been simulated by statistical emigration of 0.1 percent per year, starting in 2010 (e.g. from
offices in many countries, but also with detailed assumptions in Poland to Western Europe; some statistical data give higher
journals and a book [2-6]; the latter two references give a com- values already now). In the past this emigration was negligibly
plete Fortran program. small, about 0.04 percent per year. The lowest curve in Fig.4
These methods, with minor adjustments, were used to pre- shows the resulting decay of the total population.
dict the future growth of this ratio for countries with low birth To get more optimistic curves, the retirement age was as-
rates like Germany [2,3,7], intermediate birth rates like Algeria, sumed to increase from 63 by one year and the births by 0.2
[4] and high birth rates like the Palestinian territories [6], always in the years 2020, 2025, 2030 and 2035; increase in retirement
ignoring special historical events which are particular for these age alone did not help much. These assumptions give the x
countries. Now we apply this method to a country like Poland symbols in Fig. 4. The two lines on both sides of the x curve
where massive immigration may be less realistic; instead we correspond to changes 0.25 (higher line) and 0.15 (lower line)
simulate increases of birth rate and retirement age. Simulation in the births and indicate the order of magnitude of the extrapo-
details are shifted to an appendix. lation errors. Fig.5 shows the ratios of people in retirement age
The decay of the birth rate (more precisely, the average to people in working age, corresponding to the same simula-
number of children per woman, unfortunately called the total tions as in Fig.4.
fertility rate) came in Poland later but sharper than in Germany In the above simulations the Gompertz slope was taken
and is approximated by 2.3-0.55 [1+tanh(0.15*year  1993)]. as b=0.093, with the mortality function increasing for adults as
The total simulated population, normalized by the actual Polish exp(bx) with increasing age x, as in [1,2], including the pos-
population in 2002, agrees well with what the official Polish au- sible change around the year 1972 [8-11] (see in particular Fig.
thority, www.stat.gov.pl, predicted: Fig.2. Fig.3 shows the ratio 6 in [10]). The actual Polish value is near 0.08 similar to Algeria
of people above retirement age of 63 to the people between [4]. Using b = 0.08 instead of 0.093 we get Figs. 6 and 7, which
20 and retirement age (lower curve). (Our time units are years overlap with the results of Figs. 4 and 5.
throughout.) Thus enhanced birth rates as simulated here and massive
The future looks less problematic if the number of people immigration as simulated for countries like Germany [2,5] could
up to age 20 is added to those above 63, both groups need- reduce the shrinking of the population and the burden of the
ing public support. Then the ratio is approaching a minimum: working population to support the retired people. That burden
middle curve in Fig.3. Thus for some time, according to Fig.3 would also be alleviated by increasing retirement age; German
until about 2030, the fraction of people needing support from parliament adopted in 2006 a law regulating these future in-
Modeling and Simulation
Simulation of ratio of old to young people in countries like Poland
20
Fig. 1. Real and assumed births. The + come from the offi cial Polish statistics,
the line is the tanh approximation used in the present simulations.
Fig. 2. Official (line) and present (+ symbols) extrapolations of Polish population.
Fig. 3. Ratio of old to working-age people. For the two upper curves the young people were added to the old ones.
Border ages are 20 and 63.
Modeling and Simulation
Simulation of ratio of old to young people in countries like Poland 21
Fig. 4. Help from increased births. The lowest curve uses the present birth rate,
the higher curves assume more or less strong increases of births in the future.
Fig. 5. Help from increased births and increased retirement age. Same simulation as in the previous figure;
the rise of retirement age from 63 to 67 does not affect the total population shown in the previous figure.
Fig. 6. Ratio of old to working-age people for different Gompertz slopes 0.093 (plus signs, Germany)
and 0.08 (other curves, Polish men).
Modeling and Simulation
Simulation of ratio of old to young people in countries like Poland
22
Fig. 7. Populations for different Gompertz slopes, same simulations as in previous figure.
creases. France recently increased the births from 1.7 to 1.9 References
within a decade. Other countries in the European Union, like
Bulgaria, Romania or the three Baltic states, may be in a situa- 1. Prioux F.: L �volution d�mographique r�cente en France:
tion similar to Poland. l esp�rance de vie progresse toujours, Population-F 63:
Simulations like these took less than a second each, in 437-478, 2008.
contrast to more sophisticated methods [12], and readers can 2. Stauffer D.: Simple tools for forecasts of population ageing
change parameters to check for the effects of different as- in developed countries based on extrapolations of human
sumptions. Countries corresponding to these simulations still mortality, fertility and migration, Experimental Gerontology
have time to adjust to the future problems of the demographic 37: 1131-1136, 2002.
change. The ageing problems seem to become very serious 3. S� Martins J. S., Stauffer D.: Should retirement age be
after the year 2050. coupled to life expectancy? Ingenierias (Univ. Nuevo Leon,
Mexico) 7: 35-38, 2004.
4. Zekri L., Stauffer D.: Sociophysics simulations III: retire-
Appendix ment demography, AIP Conference Proceedings 779: 69-
74, 2005.
The simulations calculate the age distribution of the population 5. Stauffer D., Moss de Oliveira S., de Oliveira P. M. C., S�
in one year from that in the preceding year, neglecting possible Martins J. S.: Biology, Sociology, Geology by Computa-
correlations between a mother and her daughters. Fathers and tional Physicists. Elsevier, Amsterdam 2006.
sons are neglected. For extrapolations over a few generations 6. Sumour M. A., El-Astal A. H., Shabat M. M., Radwan M.
this approximation should be good enough. A complete Fortran A.: Simulation of demographic change in Palestinian ter-
program is given in [5,6], and it is available in the electronic ritories, Int. J. Mod. Phys. C 18: 1717-1723, 2007.
version of this journal. 7. Bomsdorf E.: Life expectancy in Germany until 2050, Ex-
Women give birth from ages 21 to 40, of daughters with perimental Gerontology 39: 159-163, 2004.
half of the birth number given in the text, spread evenly over 8. Yashin A. I., Begun A. S., Boiko S. I., Ukraintseva S. V.,
these 20 years. Sons represent the other half and can be ne- Oeppen J.: The new trends in survival improvement re-
glected. The population increases by immigration or decreases quire a revision of traditional gerontological concepts, Ex-
by emigration, affecting equally all ages from 6 to 40. The mor- perimental Gerontology 37: 157-167, 2001.
tality is assumed as 7b exp[b(x-X)], giving a survival probabil- 9. Wilmoth J. R., Deegan L. J., Lundstr�m H., Horiuchi S.:
ity S(x) from birth to age x as S = exp[-7 exp(-bX)(exp(bx)-1)]. Increase of maximum life-span in Sweden 1861-1999, Sci-
Thus after births and migration have been dealt with, the popu- ence 289: 2366-2368, 2000.
lation P(x) at age x is calculated from P(x) = P(x-1) S(x)/S(x-1). 10. Edwards R. D., Tuljapurkar S.: Inequality in life spans and
Here the Gompertz slope b is assumed to increase over 150 a new perspective on mortality convergence across indus-
years from b = 0.07 to b = 0.093 (or 0.08) until the year 1971, trialized countries, Population and Development Review
and then to stay constant at this maximum value. The char- 31: 645-674, 2005.
acteristic age X, in contrast, is assumed to stay constant at 11. Kannisto V., in: Human longevity, individual life duration,
103 until 1971, and thereafter to increase by 0.15 each year. and the growth of the oldest-old population, edited by Rob-
This change of trends around 1971 was seen in some empiri- ine J.-M., Crimmins E. M., Horiuchi S., Zheng Y. Springer,
cal studies of the last years [8-11]. Heidelberg 2006, p. 111.
12. Bońkowska K., Szymczak S., Cebrat C.: Microscopic mod-
eling the demographic changes, Int. J. Mod. Phys. C 17:
1477-1484, 2006.
Modeling and Simulation
Simulation of ratio of old to young people in countries like Poland 23
DEMOGRAPHY MEDICINE
STATISTICS
Modeling and Simulation
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 25-34
MAGNETOCARDIOGRAPHY  MEASUREMENT CONDUCTION,
DATA FORMATS AND DIAGNOSTIC APPLICATIONS
KAMILA BARON-PAA
UCKA
Chair of Automatics
AGH University of Science and Technology, Mickiewicza 30, 30-059 Krakow
Abstract: Magnetocardiography (MCG) is a noninvasive diagnostic method that builds on properties of a magnetic field
generated during cardiac cycle. This paper covers basic rules of measurement conduction, interpretation of recorded data
sets, as well as data formats that are commonly used for visualization. MCG diagnostic applications are presented, not only by
comparing MCG and ECG, but mainly by emphasizing certain advantages and disadvantages of MCG examination. Diagnostic
parameters used for MCG results evaluation are presented  the ones created basing on time runs as well as those originating
from maps. Moreover, paper describes content of experimental database, which author plans to use for further research on
computer interpretation techniques of MCG data sets in decision support systems for heart disease diagnostics.
Key words: magnetocardiography, MCG, MF maps, PCD maps, heart diseases diagnostics, diagnostic parameters
1. Introduction 2. MCG data formats
Magnetocardiographic examination (MCG) registers intensity of Magnetocardiographic examination registers intensity of the
magnetic field generated during cardiac electrical activity and magnetic field generated during cardiac electrical activity. Elec-
can be considered magnetic equivalent of electrocardiographic trical impulses, flowing through human heart and causing the
examination (ECG). This method arouses interest of scientists contraction and relaxation of atria and chambers, are accord-
for more than 40 years, but thanks to recent technological de- ing to Maxwell equations the source of the magnetic field that
velopment it finally became diagnostic method accepted by is oriented perpendicularly to the electrical field. The values of
medical profession as well. This acceptance was granted not the intensity of the magnetic field that are registered in course
only because of well known advantages of MCG, such as non- of a magnetocardiographic examination can be collected in
invasiveness (lack of contact between patient s skin and sen- one of the following formats.
sors), ability to diagnose diseases that cannot be discovered
with ECG or its suitability for such specialized applications as 2.1. Time runs
monitoring fetal heart s activity in last weeks of pregnancy, but
mostly because of the fact that technological advancement of The values of the intensity of the magnetic field can be regis-
MCG devices allowed them to be moved out from magnetically tered in form of time runs  each measuring point over patient
shielded rooms and be placed next to patients beds. torso is associated with one time run. Morphological features
When MCG devices that allow  next-to-bed working envi- of MCG and ECG time runs are lot alike  on MCG time runs
ronment became commercially available, MCG systems instal- there are parts similar to P wave, QRS complex and T and
lations started in many leading medical centers  both in Eu- U waves from ECG time run  there is also timing correlation
rope and USA. This caused a series of publications, which not between those elements. The significant difference is the fact,
only reaffirmed high expectations tied in with MCG, but also that in measuring points placed over lower thorax, in proximity
presented new areas of MCG application and paved the way of the midsternal plane, time runs with normal orientation of
to implementations. R and T waves are registered, whereas in measuring points
placed over upper left thorax, time runs have reversed orienta-
tion (compare in Figure 1: time runs from box 3 and 4 with time
runs from box 1) .
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications
26
tion of the measured magnetic field intensity values for each
channel (measuring point) in the same time point, as well as of
the interpolated values for points placed between sensors. As
a result of this operation, contour map or equivalent map with
artificial coloring scheme is developed. Figure 2 demonstrates
correlation between electrical vectors of the heart in particular
phases of cardiac cycle and MF maps.
The visualization method that is frequently used for pre-
sentation of data collected in course of the MCG examination
is a development of a sequence of the maps computed for
consecutive moments of the cardiac cycle. Such visualization,
often in form of animation, allows observer to capture temporal-
spatial dynamics of the alterations on MF maps related to heart
functioning. The example comparison of MF maps sequences
for healthy (A) and sick (B) hearts of patients is presented in
Figure 3.
It is worth mentioning that analysis of the pathological al-
terations on the MF maps sequences requires specialized
knowledge due to the fact, that map fragments with the largest
values of the magnetic field intensity do not equal the areas of
the heart that are the most active at the moment. This means
that from medical user point of view MF maps are not intuitively
readable, because they do not reflect the sequence of the ac-
Fig. 1. MCG time runs from 9 channels, collected sequentially in
tivation of particular fragments of the heart, and as a result do
four consecutive measuring positions. In measuring points over
not directly show pathological functioning of the particular heart
lower thorax (3 and 4) in proximity of the midsternal plane, regis-
areas (compare Figure 5). This disadvantage was a direct rea-
tered signals have normal orientation of R and T waves, whereas
son for the development of the alternative visualization method
in points over upper left thorax (1) R and T waves are reversed.
for MCG data  pseudo-current density maps (PCD maps).
On the basis of MCG, CardioMag Imaging application.
2.3. Pseudo Current Density Maps (PCD maps)
2.2. Magnetic Field Maps (MF maps) PCD maps (or alternatively Arrow Maps) were introduced by
Cohen in 1976 in order to enable such representation of mag-
The alternative format of data collected in course of magne- netic field intensity values, which would reflect the source of
tocardiographic examination is the MF map (Magnetic Field measured values  the distribution of the density of the heart s
Map). This format is created by computing spatial distribu- currents. [17].
Fig. 2. Correlation between electrical vectors of the heart in particular phases of cardiac cycle and MF maps.
Electrical vector is depicted with black arrow whereas vector of the magnetic field is depicted with white arrow.
In all phases, as expected, both vectors are mutually orthogonal [9].
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications 27
A B
Fig. 3. MF maps sequences in line with the end of T wave of healthy 53 years old male (A) and 54 years old male after stroke (B). On the
basis of MCG, CardioMag Imaging application.
The induction of the magnetic field is related to the density
r "Bz r "Bz r
of the current that generates this field in accordance with the c = �" ex - �" ey
"y "x
following equation: (2)
r
r
(1) where:
rot B = � �" j
r
where: c - pseudo current density
r
r
rot B - rotation of the magnetic field induction The map that originates from c values calculated for all
points of examination area is called PCD map. In its original
r
"By "Bx r
�# "Bz "By r "Bx "Bz r
ś# �# ś#
�# ś#
version of year 1976, PCD map consisted of arrows, which
r
ś# �" ey + - ź# �" ez
ś#
rot B = - ź# �" ex + - ź#
ś#
ś# ź# ś# ź#
"y "z "z "x "x "y lengths were coding values of the amplitude of c in a given
�# #
�# # �# #
r
point. Nowadays, amplitude of c in a given point is depicted
r r r
ex,ey ,ez - versors of the coordinate system by a color from the imposed color scale. The example of the
r contour MF map and its equivalent PCD map is presented in
j - density of the current the Figure 4.
Application of PCD maps in visualization of MCG data re-
If in course of the MCG examination all three components sults in maps where localization of the point with the largest
of magnetic induction vector were registered, it would be pos- signal amplitude is equivalent to the localization of the elec-
r
sible to obtain distribution of the density of the current that trical dipole of the heart p , and orientation of this point s
r r
generates this field. Unfortunately, typically only Bz component c vector is in accordance with orientation of p . Therefore,
is registered. Applying to it HC transformation (Hosaka-Cohen PCD map is intuitive for doctor s interpretation, since it reflects
transformation) allows to obtain the value of so-called pseudo which areas of the heart are active at the moment. The relation
current density. This transformation is described by following between PCD map and activity of the heart in comparison with
equation: analogical relation for MF map is depicted in Figure 5.
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications
28
A B
r
(p = 1� Am)
Fig. 4. Contour MF map (A) and PCD map (B) for Bz component, calculated basing on Biot-Savart s law for dipole
placed 10cm below map surface [17].
The analysis of the above figures confirms, that PCD map 3. Forward and inverse problem
intuitively shows the progress of the right atrium s activation
solutions in magnetocardiography
(Fig. 5.B), whereas similar interpretation of the MF map (Fig.
5.A) requires additional knowledge.
The additional advantage of the PCD maps application is MCG allows to register the values of the intensity of the heart s
the fact, that as contrasted with MF maps, they are indepen- magnetic field, but in the limelight remains the question of
dent of the sensor configuration that was used for the registra- how, basing on MCG data, can one recreate localization of
tion of the magnetic field (magnetometers, flat gradiometers, the source of the electrical activity of the heart. Answering this
SQUIDs). This enables simple comparison of the maps even question is crucial, since observation of the electrical impulses
when they were created on different platforms and in different flowing through the heart allows the observer to discover many
research centers [17]. pathologies such as ischemic disease or infarction scars. In this
A B
Fig. 5 A  Atrium activation visualized with MF map. The difference between consecutive contour lines is 0.5 pT (red: positive values,
blue: negative, black: Bz=0); B  PCD map equivalent to 5.A map [17]
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications 29
Fig. 6. Straight (A) and curved (B) current paths as the sources generating maps of the electrical fi eld (C and D) and maps of the mag-
netic fi eld (E and F) registered 1cm above the source and expressed in arbitrary units. [8]
context, the fact that the map of the magnetic field, as opposed be computed depends on the approximations that were made
to the map of the electrical field, carries the information about within the  model of the source articulating the activity of the
the curvature of the path of the current that was the source heart in context of the currents it generates, and within  model
generating both types of maps [8, 16] becomes a valuable fea- of the conductive volume that describes the conductive prop-
ture of MCG maps. Maps of the magnetic fields generated as erties of the chest. Finally, it has to be decided in which place
a result of the identical currents flowing through straight and the magnetic field will be calculated, in relation to the location
curved trajectories are significantly different, whereas analogi- of the source of the current and conductive volume, as well as
cal maps of the electrical fields do not show any difference at the choice of the component of the field to be computed has to
all. This difference is presented in Figure 6, where electrical be made.
and magnetic maps created for two different current paths are
compared.
MF maps, besides reflecting the differences in the path of 3.2. Inverse problem solution
the current flow, allow computing the distribution of the currents
generated by active cells of the heart muscle. Solution of such To solve the inverse problem, especially interesting in context
analytical task is called the inverse problem solution, but to de- of the magnetocardiography, is to determine the distribution of
scribe the process of obtaining such solution it is necessary the currents related to the electrical activity of the heart, on the
to explain in the first place the process of obtaining forward basis of the measured values of the intensity of the magnetic
problem solution. field. The solution of such problem is called the inverse prob-
lem solution. It is worth mentioning, that the inverse problem
does not have an unambiguous solution  it is possible that dif-
3.1. Forward problem solution ferent current distributions give rise to creation of the identical
magnetic field intensity maps. In order to determine the locali-
Computing the intensity of the magnetic field basing on meas- zation, orientation and intensity of source of the current, basing
ured distribution of the electrical field is called the forward prob- on MF maps obtained measurably, it is necessary to solve the
lem solution. While searching for the solution of such problem, forward problem iteratively  the localization, orientation and
it is necessary to take into the consideration the nature of the intensity of the hypothetical source of the current is improved
physical phenomenon occurring in course of the electrical ac- iteratively, so that the map generated by it would be as close to
tivity of the heart as well as details of the geometry of the torso. the one obtained measurably as it is only possible.
The accuracy with which the intensity of the magnetic field will
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications
30
3.3 Models cially in cases when dipoles are placed deeply and in cases of
data with high level of interference.
There are many different models of the source  from the sim-
plest one which models the whole heart activity with just single
electrical dipole, to more complicated models which contain 4. Diagnostic parameters
many dipoles, dipole in the movement or even many dipoles
in movement that create the layer representing the action po- The analysis of the magnetocardiographic examination results
tential wavefront. allows diagnosing many cardiac diseases. Depending on the
Similar distinction is observed in case of torso models experience of the performer, the process of making diagnose
 the simplest model in use is a homogeneous semi-infinite can be based solely on visual assessment of MF or PCD map
medium, but there are also more sophisticated models, which sequences, but it can as well comprise of the phase in which
not only encompass realistic geometry of human torso but also auxiliary diagnostic parameters are evaluated. Such param-
take into account internal inhomogeneities representing lungs eters can be calculated basing on MCG time runs or MF/PCD
and intraventricular blood masses. It is worth mentioning that maps.
advanced models enable obtaining potentially more accurate Diagnostic parameters are calculated for the time frames
results, but with the increase of the model s complexity, the risk that correspond to the type of the disease being diagnosed 
of potential instability of the solution increases as well. for example in case of ischemic heart diseases (IHD) only ven-
tricular repolarization time runs and maps are taken into con-
sideration. In case of some diseases both exercise and resting
3.4. Solutions tests have to be performed in order to make a diagnose.
Parameters that are most frequently used for diagnose of
Depending on the configuration of the chosen models, differ- the ischemic heart diseases include [1, 2, 3, 4, 5]:
ent types of solutions are obtained  Equivalent Current Dipole 1. Parameters calculated based on MF maps:
(ECD) or Current Density Estimation (CDE). " ą angle  tilt angle of the vector that joins points of the
If a single electrical dipole was chosen as a model of the largest (+) and the smallest (-) amplitude on MF map (for
source, and homogenous semi-infinite medium as a model of healthy man: ą "[-1100, 200])
a torso, then as a solution of the inverse problem Equivalent " rotation of ą angle in 30ms (for healthy man: ą angle should
Current Dipole will be obtained, which can be considered the rotate by less than 45� in 30ms )
first approximation of the parameters of the source of electrical " variation of the distance between (+) and (-) points in 30ms
activity of the heart. This solution is accurate enough for many (for healthy man this distance should vary by less than
purposes  for example estimation of the location of the initial 20mm in 30ms)
activation of the ventricle via an accessory pathway in patients " variation of the ratio between the field strength in points (+)
with WPW syndrome, or localization of the onset of an ectopic and (-) in 30ms (for healthy man this ratio should vary by
beat originating from an arrhythmogenic region of the heart in less than 0,3 in 30ms)
patients suffering from periods of sustained ventricular tachy- " repolarization stabilization interval after exercise (RSI) 
cardia. Unfortunately, in case of many cardiac pathologies, this time interval between Q wave onset and the stabilization
simple solution is not capable of showing complicated electro- time, which is defi ned as the moment in which orientation
physiological processes that underlie the disease. of the vector joining (+) and (-) points reaches within 50� of
If electrical activity of the heart cannot be narrowed down the orientation at the peak of the T wave (for healthy man
to small area, then application of the model of the source in the normal values depend on gender and age, for instance
form of one single dipole does not make sense. In such cases for men below 50 years old RSI should be within range of
more advanced model, with larger number of dipoles placed 152ą35 [ms])
in chosen areas of the heart, is applied. Problem to be solved 2. Parameters calculated based on MCG time runs, are equiv-
is then defined as of how to choose which dipoles should be alents of analogical parameters defi ned for ECG signal:
active at the moment, so that generated MFM map would be " ST amplitude  averaged value of MCG signal within 5ms
the equivalent of the map measured in course of the MCG ex- around point that is situated 60ms after J point of MCG
amination. Solution of that problem is called Current Density time run (for healthy man normal values depend on gender
Estimation (CDE) and finding such solution for consecutive MF and age, for instance for men below 40 years old ST ampli-
maps, opens up possibilities of diagnosing cardiac diseases tude should be within range of 2.2ą1.9 [pT])
that are otherwise hard to detect. The system of equations that " ST slope  the slope of the linear regression line fi t to the
join large amount of dipoles placed on epicardium with mea- interval from 20 to 60ms after J point of MCG time run (for
suring points located outside of the volume of the heart, does healthy man normal values depend on gender and age, for
not have an unambiguous solution, therefore some additional instance for men below 40 years old ST slope should be
constraints are needed. One of the classes that originate from within range of 16ą7 [pT s-1])
imposing additional restriction is a group of solutions which as- " T wave amplitude  value of the MCG signal at the peak of
sume that the length of the vector that corresponds to the sum the T wave (for healthy man normal values depend on gen-
of all active dipoles has to be minimal. Such solution is called der and age, for instance for healthy men below 40 years
Minimal Norm Estimation (MNE). Additional knowledge, based old T wave amplitude should be within range of 12ą6 [pT])
on heart physiology, specifies allowed range of the strength " ST-T integral  integrated MCG signal from the J point to
and position of dipoles  it helps to stabilize the solution, espe- the T wave offset (for healthy man normal values depend
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications 31
on gender and age, for instance for men below 40 years old parameters of the QRS complex [13, 28, 29]. It is also possible
ST-T integral should be within range of 1170ą520 [fT s]) to examine fetus in order to detect QT prolongation [30].
Latest applications of MCG that can be found in the
5. Diagnostic applications of MCG
literature [4, 18]:
" monitoring of the graft reaction after heart transplantation
Myocardial Ischemia " early diagnosis of arrhythmogenic right ventricular dyspla-
sia
One of the most important MCG applications is detection of " detection of acute myocarditis
the Ischemic Heart Diseases (IHD) [4]. Alterations in electro- " assessment of the risk in patients with Brugada-like ECG
physiology of the heart that were caused by ischemic disease patterns
are often not visible on ECG, thus in order to confirm suspicion " monitoring of the cardiac activity in animals subject to new
of ischemic disease exercise ECG is performed and when it s drugs testing
inconclusive, patient has to be exposed to more invasive tests " reconstruction of the patient s heart anatomy basing on
such as stress-echocardiography or coronary angiography. analysis of 3D map of current density
Therefore MCG non-invasiveness in combination with its high
accuracy in diagnosing IHD is one of its greatest advantages.
Similarly to ECG, MCG examination can be preceded by 6. MCG and ECG comparison
exposing patient to physical or pharmacological stress. In re-
searches carried out so far, both specially designed nonmag- Regardless of the fact that the source generating ECG and
netic ergometers [10] as well as standard ergometers used for MCG signals consists of the same currents flowing through the
exercise ECG [7] were utilized. The exact listing of MCG stud- heart, information content carried by both examinations is not
ies, performed with and without stress, can be found in [12]. identical  moreover, comparison of the value of those infor-
mation is favorable for MCG examination.
Localization of pre-excitation and different sources in the MCG examination is more sensitive in detection of mag-
netic field generated by tangential currents than ECG is in de-
heart
tection of electrical field generated by the same currents [6,
MCG examination enables localization of the accessory path- 9, 12]. Furthermore, vortex currents generate magnetic field
ways which trigger pre-excitation [12], for instance in patients detected by MCG, but they are not the source of any electri-
with Wolff-Parkinson-White syndrome [14]. Other similar appli- cal field that could be detected by ECG. Those features make
cations include localization of tachycardia points of origin [19] MCG more suitable for detection of the diseases that are
or localization of premature ectopic complexes [20]. It is also caused by alteration of the direction of the electrical currents
possible to perform three-dimensional localization of amag- flowing through the heart. Since in healthy heart main direc-
netic tip of pacing catheter [15, 21]. tion of the activation wavefront is radial, from endocardium to
epicardium, then MCG enables more precise detection of the
Arrhythmia risk stratification ischemic changes in direction of depolarization and repolariza-
tion than ECG does.
MCG allows to estimate the risk of life-threatening arrhythmia It is worth to mention that MCG especially sensitively re-
occurrences in patients who in the past suffered from myocar- sponds to intra  and extra  cellular currents whereas ECG
dial infarction [12, 4]. Assessment of the risk of sudden death electrodes placed on patient s thorax measure the difference
caused by arrhythmia is based on the detection of possible dis- in the potentials caused by secondary (volume) currents flow-
continuities in activation of the heart muscle during ventricular ing right beneath the skin. This difference in substantial in at-
depolarization, abnormal inhomogeneity of VR or abnormal tempts to measure cardiac activity of the fetus in last weeks of
heart rate variability. This assessment can be based on analy- the pregnancy, and was the source of the thrive of fMCG (fetal
sis of the late fields of MCG map series [22, 23] as well as MCG) which is an examination dedicated to measuring the in-
on magnetocardiographic intra-QRS fragmentation [24] or QT tensity of the magnetic field generated by fetal heart.
dispersion [25]. The important advantage of MCG examination is, as
previously mentioned, its noninvasiveness. In course of the
Detection of LV hypertrophy examination patient s skin remains not only intact but there is
also no physical contact between patient s skin and sensors.
MCG allows to identify patients suffering from left ventricle This can be considered serious ascendance over ECG since
hypertrophy (LV hypertrophy) [26] as well as to estimate the it eliminates all problems related to interferences caused by
progression of the disease [27]. skin-electrode contact. Moreover, thanks to noninvasiveness,
MCG can be performed on hyperexcitable patients, allowing to
Assessment of the fetus heart beat rhythm monitor the risk of sudden cardiac death in patients with Rett
syndrome [11] or other patients that are hard to enter into com-
MCG examination enables monitoring of the fetus cardiac ac- munication. Lack of necessity to undress patient or place elec-
tivity even in the last weeks of the pregnancy. The quality of trodes on patient s skin shortens the time needed for perform-
the registered signal is good enough to capture parameters ing examination. In conjunction with decreased susceptibility
such as AV conduction, repolarization period or morphological to movement artifacts it allows to apply MCG in monitoring of
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications
32
cardiac activity of small animals in tests of new drug influence 7. Experimental database
on heart  ECG application required animal sedation, therefore
introducing additional influence on animal s cardiac activity Available research material in form of MCG data records was
[18]. acquired in course of diagnostic examinations performed in
In contrary to ECG, configuration of MCG sensors is per- The John Paul II Hospital in Krakow, with the use of CMI 2409
manent and is not modified between consecutive examina- magnetocardiograph, CardioMag Imaging Inc. Examinations
tions. It allows to obtain high level of reproducibility, which is of were performed on patients already diagnosed with other tech-
great significance in case of therapy performed after restenosis niques and remaining under constant medical consultancy.
or transplantation or in different kinds of clinical tests [8]. Database contains results of 466 examinations of which 415
Distinction between types of the injury currents is another include commentary about patient s health state at the moment
area of MCG application. Difference between potentials of the of the examination. Following groups can be distinguished:
ischemic and normal cells at the rest state cause the flow of " patients without a cardiac infarction
diastolic injury current, whereas the difference in action po- " patients without a cardiac infarction but with hypertension
tentials between ischemic and normal cells cause the flow of " patients with a cardiac infarction
systolic injury current. ECG examination, in course of skin in- " patients with a cardiac infarction and hypertension
terference filtration filters TQ base line, thus it cannot detect " ablation patients
its displacement. As a result, ECG is unable to differentiate " patients after angioplasty
between ST segment displacement caused by systolic injury Among available data, four groups were chosen and ad-
current and the one caused by diastolic injury current. Unlike ditionally sorted according to gender and age. Following table
ECG, MCG can register TQ base line displacement, thus it al- presents details of MCG examination data set.
lows to differentiate between those two types of injury currents
[12].
Apart from all of the above advantages, MCG examina- 8. The course of planned research
tion does have some weaknesses. The greatest ascendance
of ECG over MCG is the existence of numerous publications The most common way to analyze diagnostic capabilities of
describing application and clinical suitability of ECG  there magnetocardiography involves the usage of MCG time runs
is still not enough of literature serving the same purpose for and computing morphological parameters that base on simi-
MCG. Even though the number of such publications is increas- larities between MCG and ECG signals. The other way is to
ing, there is still a lack of researches conducted on large group attempt to directly utilize MF map sequences.
of patients, which could grant for MCG full acceptance of medi- Instead of using well known parameters, like the alteration
cal profession. Nevertheless, it seems that emergence of such in the direction of the magnetic induction vector, it could be in-
publications is only a matter of time. teresting to find and use different ones, more related to the flow
Table 1. MCG examination data set
Gender 20-29 years 30-39 years 40-49 years 50-59 years 60-69 years 70-79 years 80-89 years
PATIENTS WITH A CARDIAC INFARCTION
Women 0 102110
Men 0055540
PATIENTS WITH A CARDIAC INFARCTION AND HYPERTENSION
Women 0 2 0 6331
Men 0 3 5 15 10 2 0
PATIENTS WITHOUT A CARDIAC INFARCTION
Women 0 0 10 55 10 30
Men 1 1 19 23 7 4 0
PATIENTS WITHOUT A CARDIAC INFARCTION BUT WITH HYPERTENSION
Women 0 1 8 50 23 11 1
Men 0 1 13 35 14 0 0
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications 33
of electrical impulses in the heart. For instance, one could try Laboratory for Interventional and Intensive Cardiac Care.
to establish whether it is possible to find on MF or PCD maps Lecture Notes in Computer Science: 1009-1019, 2003.
trajectories of points with the largest amplitude, which would 8. Koch H., Haberkorn W.: Magnetic fi eld mapping of cardiac
be characteristic for each of the described test groups. Finding electrophysiological function. Phil. Trans. R. Soc. Lond.
such trajectories on PCD maps would be especially interes- A 359: 1287-1298, 2001.
ting since following points with the largest field intensity values 9. Steinberg B., Roguin A., Watkins III S.: Magnetocardiogram
corresponds with monitoring the direction of activation wave- recordings in a nonshielded environment  reproducibility
front. The attempt to find and analyze new set of parameters and ischemia detection. A. N. E. 10(2): 152-160, 2005.
computed based on MF or PCD maps will be the direction of 10. Takala P., Hanninen H., Montonen J.: Magnetocardio-
author s research, whereas the main goal will be to develop graphic and electrocardiographic exercise mapping in
new methods of medical images interpretation (MF and PCD healthy subjects. Annals of Biomedical Engineering, Vol.
maps) in the way that could support diagnostics of the chosen 29: 501-509, 2001.
heart diseases (infarctions and arterial hypertension). 11. Brisinda D., Meloni A., Hayek G.: Magnetocardiographic
imaging of ventricular repolarization in Rett syndrome.
Lecture Notes on Computer Science: 205-215, 2005.
9. Conclusion 12. H�nninen H.: Multichannel Magnetocardiography And Body
Surface Potential Mapping In Exercise-Induced Myocardial
This paper describes chosen aspects of new, noninvasive Ischemia. Academic Dissertation, Helsinki 2002.
method of heart diseases diagnostics  magnetocardiography. 13. Brisinda D., Comani S., Meloni A.: Multichannel mapping
In recent years MCG started to play role of support tool for of fetal magnetocardiogram in an unshielded hospital set-
other diagnostic techniques that are more established in medi- ting. Prenatal Diagnosis, Vol. 25: 376-382, 2005.
cal profession. Fast technological development suggests that 14. Fenici R., Brisinda D., Nenonen J.: Noninvasive Study of
it will gain more importance. MCG data visualized in form of Ventricular Preexcitation Using Multichannel Magnetocar-
magnetic field map or pseudo-current density map allow to di- diography. PACE, Vol. 26: 431-435, 2003.
agnose diseases that could be missed by ECG, moreover non- 15. Fenici R., Brisinda D., Nenonen J.: Phantom validation of
invasiveness of MCG enables it to be used in situations where multichannel magnetocardiography source localization.
ECG could not be used. Indisputably MCG is a promising tech- PACE, Vol. 26: 426-430, 2003.
nique, but rapid development of medical devices was not fol- 16. Kosch O., Meindl P., Steinhoff U.: Physical aspects of car-
lowed by as dynamical development of computer methods of diac magnetic fi elds and electric potentials. Biomag 2000.
data analysis. Therefore the main goal of author s research will Proceedings of the 12th International Conference on Bio-
be to develop new algorithms supporting interpretation of MCG magnetism: 553-556, 2001.
data that could in future assist in advancement of CAD and 17. Haberkorn W., Steinhoff U., Burghoff M.: Pseudocurrent
PACS systems. density maps of electrophysiological heart, nerve or brain
function and their physical basis. BioMagnetic Research
and Technology 4: 5, 2006.
Literature 18. Koch H.: Recent advances in magnetocardiography. Jour-
nal of Electrocardiology, Vol. 37, Supplement: 117-122,
1. Makela T., Pham Q. C., Clarysse P.: A 3-D model-based 2004.
registration approach for the PET, MR and MCG cardiac 19. Moshage W., Achenbach S., Gohl K.: Evaluation of the
data fusion. Medical Image Analysis 7: 377-389, 2003. non-invasive localization accuracy of the cardiac arrhyth-
2. Chen J., Thomson P., Nolan V.: Age and sex dependent mias attainable by multichannel magnetocardiography
variations in the normal magnetocardiogram compared (MCG). Int J Cardiac Imaging 12: 47-59, 1996.
with changes associated with ischemia. Annals of Biomedi- 20. Oeff M., Burghoff M.: Magnetocardiographic localization of
cal Engineering, Vol. 32, No. 8: 1088-1099, August 2004. the origin of ventricular ectopic beats. Pacing Clin Electro-
3. Fenici R., Brisinda D., Nenonen J.: Bedsite Multichannel physiol 17: 517-522, 1994.
Magnetocardiography in Clinical Practice. Biomag2002, 21. Pesola K., Nenonen J., Fenici R.: Bioelectromagnetic lo-
VDE Verlag Gmbh: 568-570, 13th Int. Conf. on Biomag- calization of a pacing catheter in the heart. Phys Med Biol
netism, Jena 10.-14.8.2002. Berlin 2002. 44: 2565-2578, 1999.
4. Fenici R., Brisinda D., Meloni A.: Clinical application of 22. M�kij�rvi M., Montonen J., Toivonen L.: Identification of pa-
magnetocardiography. Expert Review of Molecular Diag- tients with ventricular tachycardia after myocardial infarc-
nostics: 291-305, 2005. tion by high-resolution magnetocardiography and electro-
5. Fenici R., Brisinda D., Meloni A.: Clinical validation of ma- cardiography. J Electrocardiol 26: 117-124, 1993.
chine learning for automatic analysis of multichannel mag- 23. Korhonen P., Montonen J., M�kij�rvi M.: Late fi elds of the
netocardiography. Lecture Notes on Computer Science: magnetocardiographic QRS complex as indicators of pro-
143-152, 2005. pensity to sustained ventricular tachycardia after myocar-
6. Hanninen H., Takala P., Makijarvi M.: Detection of exercise- dial infarction. J Cardiovasc Electrophysiol 11: 413-420,
induced myocardial ischemia by multichannel magnetocar- 2000.
diography in single vessel coronary artery disease. A.N. E. 24. Korhonen P., Montonen J., Endt P.: Magnetocardiographic
5 (2): 147-157, 2000. intra-QRS fragmentation analysis in the identification of pa-
7. Fenici R., Brisinda D., Meloni A.: First 36-channel Magne- tients with sustained ventricular tachycardia after myocar-
tocardiographic Study of CAD Patients in an Unshielded
Biomedical Engineering
Magnetocardiography  measurement conduction, data formats and diagnostic applications
34
dial infarction. Pacing Clin Electrophysiol 24: 1179-1186, 27. Karvonen M., Oikarinen L., Takala P.: Magnetocardio-
2001. graphic indices of left ventricular hypertrophy. Journal of
25. Oikarinen L., Viitasalo M., Korhonen P.: Postmyocardial Hypertension, Vol. 20, No. 11: 2285-2292, 2002.
infarction patients susceptible to ventricular tachycardia 28. Van Leeuwen P., Hailer B., Bader W.: Magnetocardiogra-
show increased T wave dispersion independent of delayed phy in the diagnosis of fetal arrhythmia. Br J Obstet Gy-
ventricular conduction. J Cardiovasc Electrophysiol 12: naecol 106: 1200-1208, 1999.
1115-1120, 2001. 29. Men�ndez T., Achenbach S., Beinder E.: Usefulness of
26. Fujino K., Sumi M., Saito K.: Magnetocardiograms of pa- magnetocardiography for the investigation of fetal arrhyth-
tients with left ventricular overloading recorded with a sec- mias. Am J Cardiol 88: 334-336, 2001.
ond-derivative SQUID gradiometer. J Electrocardiol 17: 30. Men�ndez T., Achenbach S., Hofbeck M.: Prenatal diag-
219-228, 1984. nosis of QT prolongation by magnetocardiography. Pacing
Clin Electrophysiol 23: 1305-1307, 2000.
PHYSICS RADIOLOGY
Biomedical Engineering
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 35-41
AN ALGORITHM FOR DETECTING LESIONS IN CBF AND CBV
PERFUSION MAPS
TOMASZ HACHAJ
Chair of Automatics
AGH University of Science and Technology, Mickiewicza 30, 30-059 Krakow
Abstract: This paper presents an algorithm that enables detection of lesions in CBF and CBV perfusion maps. There are two
steps of the algorithm: at the first step the algorithm detects symmetry axis of an image (between left and right hemisphere), in
the second stage the level of asymmetry in cerebral blood flow and cerebral blood volume is measured by detection of regions
with different perfusion in both brain hemispheres. Test of the algorithm were performed on a set of 56 different CBF and CBV
images showing or not cerebral blood flow and volume anomalies. The paper also describes the method for estimating cerebral
blood fl ow with a non diffusing contrast agent based on the Meier  Zierler convolution model as well as CBF, CBV, MTT and
TTP perfusion maps.
Key words: dynamic CT perfusion, perfusion maps, CBF, CBV, lesions detection, symmetry detection, Meier  Zierler
convolution model
Introduction of cerebral blood flow (CBF) and cerebral blood volume maps
(CBV). Despite the fact that the norm for each of the perfu-
Brain perfusion imaging becomes more and more popular in sion parameter has been computed, the diagnosis is based on
head injuries, epilepsy, brain vascular disease and especially comparison of relative values of symmetric regions of interest
in stroke diagnosing ([7], [13], [15], [16], [18], [19], [24]). Dy- (ROI) of blood perfusion between left and right hemisphere.
namic CT / MR perfusion treatment is a modern and broadly ([7]).
used neuroradiology technique that enables to evaluate total The algorithm presented in the following paper is mainly
and regional blood flows in time unit. P-CT treatment enables based on mathematical morphology which is commonly used
recognition of structural changes of ischemia and shows the as reliable tool for image analysis ([1]). Mathematical morphol-
difference between ischemic stroke and hemorrhagic stroke. In ogy is successfully used not only in medical image analysis
the high  industrial countries of Western Europe brain stroke but also in materials engineering ([2]) and on many other
is the third  due to its frequency  reason of death (just after fields of science ([3]). The algorithm derived by the author has
heart attack and cancer) and the most frequent cause of death been implemented in Matlab environment with using Imaging
in elderly age ([10]). There are different methods of treatment of processing toolbox ([25]) package and mathematical libraries
ischemic and hemorrhagic stroke and there is a short period of ([26]). The test set for the algorithm consisted of 28 CBF perfu-
time when treatment can be used, so it is important to early find sion images and 28 CBV perfusion images that are results of
the cause of illness. Analyzing the perfusion maps can do it. researches on 8 patients. Maps were generated with Syngo
There are few algorithms for detecting lesions on CT imag- Neuro Perfusion CT software by Siemens ([23]).
es (i.e. e. [5]) but there are no common methods for analyzing
series of dynamic perfusion images like brain perfusion maps.
Commercial software for perfusion imaging that is broadly 2. The Meier  Zierler model
used in hospitals (i.e. Syngo Neuro Perfusion CT by Siemens)
doesn t contain a mechanism that enables automatic detection There are four types of basic perfusion maps: cerebral blood
of perfusion anomalies that can be seen on that kind of maps. flow perfusion map (CBF), cerebral blood volume (CBV), mean
This paper presents a method designed by the author which time transit (MTT) and time to peak (TTP). In order to generate
enables detection of potential lesions that can be visualized dynamic CT / MR perfusion maps the Paul Meier and Kenneth
Pattern Recognition
An algorithm for detecting lesions in CBF and CBV perfusion maps
36
L. Zierler model is commonly used ([7]). In this radiology treat- MTT). Each pixel corresponds to the value of perfusion map in
ment the contrasting material is injected in to the cardiovas- the given point. Usually a map has a resolution of 512 x 512
cular system relatively quickly (impulse injection). CT scanner pixels in the gray scale (because of DICOM format there can
measures the contrast material that remains in the capillary be more than 256 gray shades). The symbols used in the fol-
network. As a result the time density curve (TDC, also time in- lowing algorithm are presented in Table 1.
tensity curve TIC) for brain arteries and tissues is obtained. In
R(t)
Meier  Zierler model the impulse residue function (IRF, ) Table 1. Notation used in the description of the algorithms (based
is also defined. This function indicates the amount of contrast on notation from [1]).
in the brain tissues as the function of time. The value of pixels
Notation in algorithm Description
in TTP perfusion map is measured as maximal value of TDC
curves. This map doesn t need any further calculation. It is not
Assignment A to B.
A := B
possible to calculate R(t) directly, because there is no possibil-
ity to identify in non  invasion way the arteries that transmit
The binary image.
I
blood to separate regions of the brain. Instead of that contrast
material is injected that remains in the capillary network and
Value (color) of pixel of binary
can be measured by CT scanner (the mass of the contrast in
I(x, y)
image with coordinates (x, y) .
tissues Q(t) ). If the blood flow is constant and contrast en-
hancement is linear dependent to contrast concentration the
Gray shades image.
i
Meier  Zieler model can be used:
Value of pixel of gray shades with
Q(t) = CBF �"[CA(t) " R(t)]
i(x, y)
(1)
coordinates (x, y) .
Q(t) CA(t)
Asymmetry map (CBF / CBV).
Where and are the contrast enhancement a
of brain tissues and arteries (values can be obtained from CT
Value of pixel of asymmetry map
"
/ MR scans), is a convolution operator. According to Axel
a(x, y)
with coordinates (x, y) .
[12] blood volume that flows through capillary network (CBV)
can be obtained as:
Returns the largest integral value
Floor that is not greater than number
"
given.
+"Q(t)dt
0
 Filling the holes in the binary
CBV =
Fill
"
image.
(2)
A
+"C (t)dt
Labels all the components in the
0
Label
binary image.
If impulse residuum function ( R(t) ) is known one can
Binarization with bottom threshold
calculate MTT as follows: BinX
X.
"
Erosion with semi  circular struc-
+"R(t)dt
IŚEX
tural element with diameter X.
(3)
0
MTT =
Dilation with semi  circular struc-
Rmax
I �" EX
tural element with diameter X.
Rmax R(t)
Where is the maximal value of curves.
Opening with semi  circular struc-
Cerebral blood flow (CBF):
I o EX
tural element with diameter X.
"
Closing with semi  circular struc-
Rmax �"
+"Q(t)dt
CBV I " EX
tural element with diameter X.
0
CBF = = (4)
" "
MTT
Median filter sized YxY.
A
+"R(t)dt �"+"C (t)dt
MedYxY
0 0
Logical multiplication of images
Q(t)
Concentration of contrast in tissues is correlated
A )" B
A and B.
with concentration of contrast in arteries CA(t) and impulse
residuum function R(t) as it states in equation (1) so
R(t)
Logical sum of image.
A *" B
can be calculated as a product of deconvolution. The most
common way of solving that problem is to change it into the
Mirror image of image I towards
matrix equation and after it solves it with singular value decom-
I
vertical symmetry axis.
position method (SVD) ([13], [14], [20], [21], [22]).
The values obtained this way are used for construction
proper perfusion maps (separate map for TTP, CBF, CBV and
Pattern Recognition
An algorithm for detecting lesions in CBF and CBV perfusion maps 37
3. Detecting the symmetry axis on CBF detection of symmetry axis several fi rst and last centers of
mass should be omitted (empirically tested, that it should
and CBV images
be 3 fi rst and 3 last centers). This is because the irregular-
ity of borders of brain image of CBF and CBV maps. The
In order to do further perfusion image analysis it is necessary symmetry axis found by the algorithm is marked on Fig
to find lines that separate left brain hemisphere from right (in 1 by the white straight line. The improvement in the final
the following text it is called  symmetry axis ). In many CBF calculation of symmetry axis can be obtained by perform-
and CBV image the head of the patient is slightly rotated on ing on each of the  stripes of image morphological closing
small angle towards bed of tomograph (the symmetry axis with structural semi  circular element with relatively large
isn t orthogonal towards OX axis). The symmetry axis of the size (best results was for 25 pixels diameter).
image isn t always the same line that separates the brain
hemispheres, what makes problem more complicated. In com- The image is rotated around point that lays in the arithmeti-
mercial software the symmetry axis selection is made manu- cal canter of mass of brain so that the symmetry axis is now
ally (i.e. [23]). It is difficult to propose the certain algorithm for orthogonal to OX axis. The constant point of the rotation is cal-
automatic determination of symmetry axis because CT scans culated as:
are made in many different plains and because of that any
x2 + x1
assumption about the shape of the image cannot be stated. xc =
2
Those facts also obstruct the structural-based image analysis (7)
algorithms (algorithms of that kind often find good solutions in
y2 + y1
medical images describing [4]). yc =
2
The author s method of symmetry axis detection for CBF
and CBV maps is based on basic observation:  For a uni- Points and (x2, y2) on Fig 1.a are the first
(x1, y1)
form body it s center of mass is also its center of volume [6]. (when looking from the top of the picture) and the last points
If a body has got a symmetry axis, that symmetry axis come that is common for the symmetry axis and the image of the
through the center of volume, so for a uniform body the center brain.
of mass lays on a symmetry axis if fit exists. So assuming uni-
formity and symmetry of a body we can determine the symme-
try axis finding its center of volume. The algorithm of symmetry
axis derivation comes as follow:
" Find the square window in the area of image (ROI) in which
pixels corresponding to brain are located:
xmin = minx : I(x, y) > 0
xmax = maxx : I(x, y) > 0
ymin = miny : I(x, y) > 0
(5)
ymax = max : I(x, y) > 0
y
ymax - ymin ), sepa-
" Compute number of rows of ROI (
rate the ROI on identical number of horizontal  slices of
the same width as basic ROI (best results very obtained for Fig 1. (a) A CBF image with centers of mass marked (white cir-
20 regions). For each of the regions compute coordinates cles). A white line towards them is a potential symmetry axis of
of the center of mass as follows: the brain (the line was approximated with least square method).
(b) The same image after rotating the image around arithmetical
+"� �" x �" dV
center of brain. Potential symmetry axis (white line) is now ortho-
V
xm = gonal to OX axis.
m
(6)
+"� �" y �" dV
4. The algorithm for lesions detection
V
ym =
in CBF and CBV perfusion maps
m
Where:
The same algorithm for detection potential lesions (asym-
0 gdy I(x, y) = 0
ż#
� = metries with the sufficient size) was used for processing both
�#1 gdy I(x, y) > 0
kinds of maps. The values of the factors used as the param-
�#
eters in the algorithm were empirically determined. The algo-
White circles on Fig 1.a mark centers of mass of the ex- rithm goes as follows:
ample image. " Loading the image from the DICOM fi le. In order to make
" Centers of mass obtained in the previous step were used analysis of the algorithm easier each image was colored
for calculation of the symmetry axis of the figure. The an- with color palette that can be seen on Fig 2. Black color
gular coeffi cient and the translation coeffi cient were both corresponds to smallest values of CBF / CBV, red color
estimated with least square method. In order to improve corresponds to the highest ones.
Pattern Recognition
An algorithm for detecting lesions in CBF and CBV perfusion maps
38
Relative asymmetry of regions is computed:
If i(x, y) > i(x, y)
�# ś#
�# ś#ź#
i(x, y)
a(x, y) := Floorś#10 �" ś# ź#ź#
ś# ź#
ś#
i(x, y)
�# #
�# #
In other cases:
�# ś#
�# ś#ź#
i(x, y)
ś# ź#ź#
a(x, y) := Floorś#10 �"
ś# ź#
ś#
Fig 2. Color palates that were used for coloring the CBF (a) and
i(x, y)
�# #
�# #
CBV (b) perfusion maps.
If one of the pixels i(x, y) or i(x, y) has 0 value then
1 is assigned.
" Detection of the symmetry axis with this algorithm that " Detection of potential asymmetry. The binarization with
was described in paragraph 3. The image is also rotated bottom threshold 60 is performed.
around constant point calculated according to equation (7).
A := Bin60(a)
After that operation the symmetry axis should be orthogo-
nal to OX axis. " The symmetry axis obtained in (b) does not necessarily
" Reduction of number of colors (gray shades) on CBF and separate brain into symmetry regions. On the asymmetry
CBV map. Reduction is necessary because it improves re- map A the order of the brain region can be detected as
sults of median fi ltration (see (d)). After this operation the potential asymmetry. In order to eliminate this effect the bi-
homogeneous areas on both sides of brain map are ob- nary mask of  potentially biggest brain region P is created.
tained. The mask is the same size as mask A and it is consisted
of pixels for which i(x, y) > 0 or i(x, y) > 0 . All the
0 when i(x, y) e" 0, i(x, y) < 15
ż#
 holes in the interior region of map are removed. The bor-
�#1 when i(x, y) e" 15, i(x, y) < 30 der of the mask is also removed by morphological opening
�#
i(x, y) :=
�#2 when i(x, y) e" 30, i(x, y) e" 45 with semi  circular structural element of diameter 5.
�#
P(x, y) := Bin1(i *" i)
�#
...
�#
P := Fill(P)
In another words:
P := PŚE5
i(x, y) := Floor((i(x, y) /15)*15)
" Calculation of logical multiplication of the asymmetry mask
" Median fi ltration with the window size of 15x15 pixels. Me- A and the brain mask P.
dian average fi lter doesn t generate new values of pixels,
A := A )" P
thanks to that homogeneous areas on both sides of brain
map are obtained. The result image lacks some details (the " Elimination of narrow links between regions. Morphological
presence of them would make further analysis of image opening with semi  circular structural element of diameter
harder because they generate large number of asymmetry 3 performs that operation.
regions). If there are to much small asymmetry regions the
A:= Ao E3
proper classification turns to be impossible.
" The last step of the algorithm is eliminating all the small re-
i := Med15x15 (i)
gions. Performing morphological erosion of semi  circular
" Compare the left and the right sides of the image. As the structural element of diameter 10 accomplishes this level.
result the asymmetry map is obtained. The asymmetry On the result image there are only regions that contain
map generation goes as follows: some pixels that  survived the above operation.
l := Label(A)
i(x, y)
For each pixel on that belongs to the left part of
A':= AŚE1
0
the image, check i(x, y) > 0 or for the symmetrical pixel to-
wards symmetry axis detected in (b)) i(x, y) > 0 . If A'(x, y) > 0 add the whole region with label l(x, y)
If and i(x, y) = 0 , the background of the to the result image A.
i(x, y) = 0
image is found. " The asymmetry map A is combined with CBF / CBV map
In the other case one of the following condition is satisfied: (from point (b) of this algorithm). On the left side of the im-
If i(x, y) > 0 and i(x, y) = 0 or i(x, y) = 0 and age region with fractional decreasing of CBF / CBV can be
i(x, y) > 0 then one of the pixels belongs to the brain re- seen (the left side of the patient is on the right side of the
gion and the other can belong to the background (if the sym- image).
metry axis wasn t set properly, the region of CBF / CBV map is The results of above algorithm is presented step by step
asymmetric) or both of pixels belong to the brain region, but in on Fig 3 for CBV map (in case of CBF maps results are simi-
one of the cases CBF / CBV value is very small. lar). Fig 4 shows four pairs of CBF / CBV perfusion maps and
If i(x, y) > 0 and i(x, y) > 0 both regions belong to the same pairs (after processing them by the algorithm) with
the brain. lesions marked.
Pattern Recognition
An algorithm for detecting lesions in CBF and CBV perfusion maps 39
Fig 3. The steps of asymmetry  detection algorithm in CBV maps. This fi gure is described in the text.
Fig 4. Four pairs of CBF (top left) and CBV (bottom left) perfusion maps. Right from the arrow the same pair after process-
ing it by the algorithm. Detected borders of potential lesions regions are marked with white line. Description of the images:
(a), (b) CBF and CBV is slightly decreased on the left on the level of the top sides of lateral ventricle frontally and parietally
(c) CBF and CBV is decreased in the region of right middle cerebral artery, brain stroke was diagnosed (d) CBF and CBV is
slightly decreased on the left on the level of the top sides of lateral ventricle frontally, parietally and temporally.
Results and discussion pixels positions). In other cases deviation wasn t greater than
5 degrees from proper symmetry axis. In three cases for CBF
The algorithm for detection asymmetry on perfusion maps has maps and four cases for CBV maps improper symmetry axis
been tested on 28 CBF and 28 CBV perfusion maps from 8 causes overestimation of asymmetry regions (Fig 4.d, the im-
different patients. That test set consisted of 24 cases where proper region can be seen in the central part of the brain near
abnormal perfusion was diagnosed. The symmetry axis was confluence of the sinuses [8], [9]). The errors of that kind could
rightly detected in 42 cases (21 for CBF and 21 CBV maps, be caused by insufficient symmetry of the binary mask applied
because CBF and corresponding to it CBV map have identical on the brain. There are two reasons of this: asymmetry of CBF
Pattern Recognition
An algorithm for detecting lesions in CBF and CBV perfusion maps
40
Fig 5. The accuracy improvement of algorithm after manually choosing symmetry axis that can be seen in some cases (more ditels in
paragraph 5). Pair of CBF / CBV asymmetry map with detected borders of potential lesions regions marked with white line (the same
case as in Fig 4.d)). (a) Symmetry axis was detected automatically (overestimation of asymmetry regions), (b) After setting symmetry
axis manually overestimated region disappeared.
/ CBV perfusion maps (lack or very small perfusion in some References:
regions) or natural brain asymmetry.
The detection of asymmetry was tested on the same set of [1] John Goutsias, Sinan Batman, Morphological Methods for
images. All of the maps that don t have diagnosed asymmetry Biomedical Image Analysis, Medical Imaging, Volume 2.
were rightly classified (no asymmetry detected). On one of the Medical Image Processing and Analysis, Washington USA,
images of CBF that consisted lesions no changes were detect- 2000
ed (on the level of the top sides of lateral ventricle in the left [2] Krysztof J. Kurzydłowski, Leszek Wonjnar, Janusz Szala,
hemisphere; in medical documentation lesions was described Praktyka analizy obrazu, Polskie Towarzystwo Stereolo-
as fractional). On three CBF and CBV maps small excessive giczne, Kraków 2002
regions of asymmetry were detected that were situated close [3] Zygmunt Wróbel, Robert Koprowski, Praktyka przetwa-
to automatically detected symmetry axis (for this maps sym- rzania obrazów w programie Matlab, Akademicka Oficyna
metry axis wasn t rightly detected and it was slightly rotated Wydawnicza Exit, Warszawa 2004
towards the true axis that separate brain hemispheres). In one [4] Marek R. Ogiela, Strukturalne metody rozpoznawania ob-
case in CBV map algorithm detected a large asymmetry region razów w kognitywnej analizie zobrazowań medycznych,
(it was enlarged by region near automatically detected sym- Uczelniane Wydawnictwa Naukowo-Dydaktyczne, Kraków
metry axis). 2004
The algorithm presented in this publication has achieved [5] Adam Sędziwy, Automatyczne wykrywanie i analiza zmian
satisfactory results. 85,7% maps were properly diagnosed patologicznych w obrazach MR i CT struktur mózgowych,
(85,7% for CBF and 85,7% CBV). 75% errors in CBF maps rozprawa doktorska, Kraków 2003
and 100% errors in CBV maps were caused by over  detec- [6] Encyklopedia fi zyki, PWN, Warszawa 1972
tion of asymmetry regions. Errors were eliminated in cases [7] Jerzy Walecki, Tomasz Bulski, Katarzyna Sklinda, Choroby
when symmetry axis was manually selected (Fig 5). After that naczyniopochodne ośrodkowego układu nerwowego, Po-
96,4% CBF maps and 100% CBV maps were rightly diag- stępy neuroradiologii, Warszawa 2007, 472-512
nosed. In order to eliminate that kind of errors new algorithm [8] Jolanta Sikorska, Katarzyna Sklinda, Tomasz Ochrynik,
for symmetry axis detection should be developed. Only in one Anatomia radiologiczna OUN, Postępy neuroradiologii,
case for CBF maps fractional lesions were not detected (that Warszawa 2007, 151-166
is 3,6% of terror diagnosis for population of CBF maps and 1,8 [9] Jerzy Gielecki, Anna Żurada, Grzegorz Gajda, Wiesław
for population of all map that was tested). Important remark is Cybulski, Ośrodkowy i obwodowy układ nerwowy, Górnicki
that the same algorithm can produce proper asymmetry maps Wydawnictwo Medyczne 2008, CD-R
for both CBF and CBV perfusion maps. Further researches by [10] Hans  Christopher Diener, Michael Forsting, Udar mózgu.
the author will be directed on improving the algorithm shown, Podręczny atlas, Wydawnictwo Medyczne, Wrocław 2004
adaptating it to other perfusion maps (TTP and MTT) and pro- [11] K.L. Zierler, Equations for measuring blood fl ow by exter-
posing methods for automatic creation of diagnosis prognoses nal monitoring of radioisotopes, Circ Res 1965; 16: 309-21
based on information derived by the detection algorithms.
Pattern Recognition
An algorithm for detecting lesions in CBF and CBV perfusion maps 41
[12] L. Axel, Cerebral blood fl ow determination by rapid-se- [19] James D. Eastwood, Michael H. Lev, Max Wintermark,
quence computed tomography: a theoretical analysis, Ra- Clemens Fitzek, Daniel P. Barboriak, David M. Delong,
diology 1980; 137: 679-686 Ting-Yim Lee, Tarek Azhari, Michael Herzau, Vani R. Chi-
[13] Makoto Sasaki, Kohsuke Kudo, Hirobumi Oikawa, CT lukuri, James M. Provenzale, Correlation of early dynamic
perfusion for acute stroke: Current concepts on technical CT perfusion imaging with whole-brain MR diffusion and
aspects and clinical applications, International Congress perfusion imaging in acute hemispheric stroke, AJNR 24
Series 1290 (2006), 30-36 (2003), 1869-1875.
[14] R. Wirestam, E. Ryding, A. Lindgren, B. Geijer, S. Holt�s, [20] XingFeng Lia, Jie Tiana, EnZhong Lia, XiaoXiang Wanga,
F. St�hlberg, Absolute cerebral blood fl ow measured by JianPing Daib, Lin Aib, Adaptive total linear least square
dynamic susceptibility contrast MRI: a direct comparison method for quantification of mean transit time in brain per-
with Xe-133 SPECT, Magnetic Resonance Materials in fusion MRI, Medical Image Processing Group, Institute of
Physics, Biology and Medicine 11 (2000), 96-103 Automation, Chinese Academy of Sciences, Beijing, China,
[15] Michael H. Lev, Alan Z. Segal, Jeffery Farkas, Syeda T. Department of Radiology, Tian Tan Hospital, Beijing, Chi-
Hossain, Christopher Putman, George J. Hunter, Ronald na. Received 27 September 2002; accepted 24 January
Budzik, Gordon J. Harris, Ferdinando S. Buonanno, Mus- 2003
tapha A. Ezzeddine, Yuchiao Chang, Walter J. Koroshetz, [21] H.-J. Wittsack, A.M. Wohlschl�ger, E.K. Ritzl, R. Kleiser,
R. Gilberto Gonzalez, Lee H. Schwamm, Utility of perfu- M. Cohnena, R.J. Seitz, U. M�dder, CT-perfusion imaging
sion-weighted CT imaging in acute middle cerebral artery of the human brain: Advanced deconvolution analysis us-
stroke treated with intra-arterial thrombolysis: prediction ing circulant singular value decomposition, Computerized
of fi nal infarct volume and clinical outcome, Stroke 32 Medical Imaging and Graphics 32 (2008), 67-77
(2001), 2021-2028 [22] T.S. Koh, C.K. Markus Tan, L.H. Dennis Cheong, C.C.
[16] James D. Eastwood, Michael H. Lev, Tarek Azhari, Ting- Tchoyoson Limc, Cerebral perfusion mapping using a ro-
Yim Lee, Daniel P. Barboriak, David M. Delong, Clemens bust and efficient method for deconvolution analysis of dy-
Fitzek, Michael Herzau, Max Wintermark, Reto Meuli, namic contrast-enhanced images, NeuroImage 32 (2006),
David Brazier, James M. Provenzale, CT Perfusion Scan- 643-653
ning with Deconvolution Analysis: Pilot Study in Patients [23] Siemens AG, Clinical Applications. Application Guide.
with Acute Middle Cerebral Artery Stroke, Radiology 2002; Software Version syngo CT 2007A, Siemns Medical,
222: 227-236 06/2006
[17] Matthias Koenig, Michael Kraus, Carmen Theek, Ernst [24] Kohsuke Kudo, Perfusion Mismatch Analyzer (PMA),
Klotz, Walter Gehlen, Lothar Heuser, Quantitative assess- Acute Stroke Imaging Standardization Group (Online
ment of the ischemic brain by means of perfusion-related only), http://asist.umin.jp
parameters derived from perfusion CT, Stroke 32 (2001), [25] Matlab 6  Imaging processing toolbox, User s Guide, The
431-437 MathWorks, Inc., March 2008 (Online only), http://www.
[18] Richard E. Latchaw, Howard Yonas, George J. Hunter, mathworks.com/access/helpdesk/help/pdf_doc/images/
William T.C. Yuh, Toshihiro Ueda, A. Gregory Sorensen, images_tb.pdf
Jeffrey L. Sunshine, Jose Biller, Lawrence Wechsler, Ran- [26] Matlab 7  Mathematics, The MathWorks, Inc., March
dall Higashida, George Hademenos, Guidelines and rec- 2008 (Online only), http://www.mathworks.com/access/
ommendations for perfusion imaging in cerebral ischemia, helpdesk/help/pdf_doc/matlab/math.pdf
Stroke 34 (2003), 1084-1104
COMPUTER SCIENCE RADIOLOGY
Pattern Recognition
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 43-50
THE REGISTRATION AND ATLAS CONSTRUCTION OF NOISY
BRAIN COMPUTER TOMOGRAPHY IMAGES BASED ON FREE FORM
DEFORMATION TECHNIQUE
TOMASZ HACHAJ
Chair of Automatics
AGH University of Science and Technology, Mickiewicza 30, 30-059 Krakow
Abstract: This article presents the algorithm for nonlinear registration images of brain computer tomography. The registration
process is performed by affine transformation optimized by Nelder  Mead simplex method and non-rigid free form deformation
model based on B  splines. That algorithm is also used for creation of simple intensity driven deformable brain atlas that can
be used for indexing different parts of patient s brain. Algorithm was tested on 28 CT images, with resolution 512 x 512 pixels,
obtained for 8 different patients.
1. The registration of image define a mathematical measure of intensity similarity between
the deforming scan and the target. The second approach was
Registration is the process of finding a transformation T that adopted by the author because it does not require the segmen-
best matches two images according to a criterion of similar- tation or preprocessing of registrated images.
ity [7]. One image is the reference, which remains fixed dur- Many brain images registration algorithms are developed
ing the registration process, while the other is deformed in and tested for magnetic resonance (MR) and require prior seg-
the geometric space of the reference. The reference image is mentation of the image ([20], [21], [22]). There are some cases
also called  fixed or target image and the transformed image when there are only brain CT images (i.e. during dynamic brain
is called the  moving image . After registration of the image perfusion CT ([13], [14])) that can be used for proper diagnosis
( making it similar to other known images  so called cases ) statement. Dynamic brain perfusion CT generates set of blood
there is an opportunity to use it in system of automatic image perfusion maps where due to detecting asymmetries important
understanding ([11], [12]) and describe them more succes- brain lesions can be detected ([15]). It is important to find the
sively without registration. The researches described in this size of this lesions and to describe their position in the image.
paper were performed on the set of brain computer tomog- As perfusion maps are product of set of CT scans there is
raphy (CT) images. The main goal was to register that set of a possibility to use one of these CT scans (often the first or one
images in order to enable atlas-based description of image s of the first images where the injected contrast agent is still not
content. Deformable brain atlases are adaptable brain tem- visible) as the  fixed or target image and the brain atlas im-
plates that can be individualized to reflect the anatomy of new age as  moving image . After a registration a brain atlas image
subjects ([9]). This allows an automated labeling of structures on the new CT scan the map describing the individual parts of
in new patients scans. Nonlinear registration approaches are brain is created.
commonly classified into two major types, intensity-based and
model-based, depending on the type of information that drives
them. Model-driven algorithms first build explicit geometric Global spatial transformation
models, representing separate, identifiable anatomic elements
in each of the scans to be matched. Model-driven approaches
contrast with intensity-driven approaches. Intensity driven ap- One of the simplest methods of registration images is affine
proaches aim to match regional intensity patterns in each scan transformation ([19]). The affine transformation captures only
based on mathematical or statistical criteria. Typically, they the global motion of the image.
Pattern Recognition
The registration and atlas construction of noisy brain computer tomography images&
44
The basic affine transformation in R3 is:
(Amn - A)�"(Bmn - B)
""
x' x
Ą# ń# Ą# ń#
m n
(2.9)
r =
ó# ó#
y'Ą# = A�" yĄ# + b
�# ś# �# ś#
2 2
ó# Ą# ó# Ą#
ś# (Amn - A) �"ś# (Bmn - B) ź#
ź#
"" ""
ó# z'Ą# ó# zĄ#
�# m n # �# m n #
Ł# Ś# Ł# Ś#
(2.1)
Where: Correlation coefficient can be used for measurement the
quality of the registration images (i. e. [4]).
Ą# ń#
a11 a12 a13
b1
Ą# ń#
 mean value of all elements of A or B
ó# Ą# A, B
A = a22 a23Ą# , b = ó#b2 Ą#
(2.2)
ó#a21 Iref  reference image of registration.
ó# Ą#
ó#a31 a32 a33 Ą#
ó#b3 Ą#
Ł# Ś#
Ł# Ś#
In the author s researches the Nelder  Mead simplex
method was used ([1], [2]) that was proposed as a good solu-
If b = 0 we simply get the equation of linear transforma- tion ([3]) for finding optimal vector (2.7). The initial vector for
tion. the process of optimization was d = [1,1,0,0,0] (void trans-
Several common spatial transformations applied to images formation).
can be expressed in terms of an affine equation. In R2 the
matrix responsible for scaling will be:
3. Local spatial transform, non  rigid
s11 0
Ą# ń#
AS = , b = 0 (2.3)
free form deformation transformation
ó#
0 s22 Ą#
Ł# Ś#
x'= s11 �" x
The affine transformation captures only the global motion of
y'= s22 �" y
the brain. An additional transformation is required, which mod-
Rotation: els the local deformation of it. The nature of the local defor-
mation of the brain can vary significantly across patients and
cos� sin�
Ą# ń#
, b = 0 (2.4) with age. Therefore, it is difficult to describe the local deforma-
AR =
ó# Ą#
tion via parameterized transformations. Instead, the free form
Ł#- sin� cos� Ś#
deformation (FFD) model, based on B  splines, which was
x'= x �"cos� - y �"sin�
successively used i.e. in registration breast MR images [4], CT
liver images [5] or tracking and motion analysis in cardiac im-
y'= x �"sin� - y �"cos�
ages [6]. The basic idea of FFD s is to deform an object by ma-
Translation: nipulating an underlying mesh of control points. The resulting
deformation controls the shape of the 3  D or 2  D object and
1 0
Ą# ń#
At = (t1,t2) (2.5) produces a smooth and continuous transformation.
ó#0 1Ą# , bt =
One familiar technique to represent a nonrigid transforma-
Ł# Ś#
tion is to employ spline functions such as B  splines, because
x'= x + t1
B  splines are locally controlled, they are computationally ef-
y'= y + t2
ficient compared to the other globally controlled splines [5].
All of these transformations can be simply combined by B  spline free form deformation is a non-rigid registration
multiplying transformation matrixes, in example if we want to algorithm that has been widely used in the past few years ([7]).
perform scaling rotation translation: The transformation model is a free-form deformation (FFD) that
is described by a cubic B-spline. For any point of image x, y,
x' x
Ą# ń# Ą# ń#
the B  spline transformation is computed from the positions of
= AR �" At �" AS �" + bt
ó# ó#
the surrounding 4x4 control points. The mesh of control points
y'Ą# yĄ#
(2.6)
Ł# Ś# Ł# Ś#
controls the deformation of the whole image. The basic idea of
Those three transformations in R2 have all together five FFD in image registration process is to manipulate the position
degrees of freedom: of points in mesh grid in order to minimize the  difference (er-
(2.7) ror function) between new and reference images.
d =[s11, s22,� ,t1,t2]
The transformation model of a free-form deformation that
Let I' = Aff (d, I ) be an affine transformation of image is performed by a cubic B  spline can be described by the
I. In order to find optimal affine transform with parameters equation below:
d
3 3
the function to minimized was defined as:
T (x, y) = (u) �" �m (v) �"Ći+l, j+m (3.1)
""�l
f (d ) = - r(I', Iref )
(2.8) l=0 m=0
Where: Where:
r(A, B)  correlation coefficient between A and B, i, j  the indices of the control points,
where A and B are matrices or vectors of the same size given
ó# Ą#
ó# Ą# y
by equation: x
j = - 1
i = - 1
ó#� Ą#
ó#� Ą#
ó# Ą#
y
Ł# x Ś# Ł# Ś#
, (3.2)
Pattern Recognition
The registration and atlas construction of noisy brain computer tomography images& 45
Fig 1. Adaptive mesh refi nement during iterations of algorithm. New point of mesh is placed between two old points. At the beginning
mesh with size 3x 3 (black points), in the next step 5 x 5 (new points are gray) and the final 9 x 9 mesh (white points).
Fig 2. (a) The image of the brain with mesh of control points size 6 x 6. The distance between nearest points Ć in the row / column is �.
(b) Control point Ći, j affects points only inside the neighborhood domain (the error computation during the optimization
4 �"� � 4 �"
process of the position of each of the control points is performed only inside this region).
� ,�  the spacing between the control points (in mesh
x y
(3�"t3 - 6�"t2 + 4)
grid). �1(t) = (3.5)
6
u, v  the relative positions of (x, y) inside that cell in the 2D
space,
(- 3�"t3 + 3�"t2 + 3�"t +1)
(3.6)
�2 (t) =
ó# Ą#
ó# Ą#
x x y y
, (3.3) 6
u = - v = -
ó#� Ą#
ó#� Ą#
� �
ó# Ą# t3
x Ł# x Ś# y y
Ł# Ś#
�3(t) = (3.7)
Ći, j  set of deformation coefficients that is defined on 6
a sparse, regular grid of control points placed over the moving In the discrete case the return coordinates must be
image. changed in to integer values. After calculation the coordinates
The functions �0 through �3 are the third  order spline of the pixel (xnew, ynew ) = T (x, y) that will be trans-
polynomials: formed to the current x, y pixel the value of color (gray level)
(- t3 + 3�"t2 - 3�"t +1)in the pixel must be computed. In the author s algorithm the
�0 (t) = (3.4) shown below method of computation of gray level is used:
6
Pattern Recognition
The registration and atlas construction of noisy brain computer tomography images&
46
The algorithm is a modified version of the B  spline reg-
(x , ylocal ) := T (x, y) - (x, y)
local
Ł#T Ś# (3.8) istration algorithm in [4]. It does not include the control grid
smooth optimization, instead the grid refinement was used.
perc(0) = (1- xlocal ) �"(1- ylocal ) The local transformation is represented as a combination of
B  spline FFD s at increasing resolutions of the control point
perc(1) = (1- xlocal ) �" ylocal mesh. The control point mesh at level l is progressively refined
by inserting new control points between the existing ones to
perc(2) = xlocal �"(1- ylocal ) create the control point mesh at level l +1 (Fig 1.).
The new image to register is I, the reference image is Iref .
perc(3) = xlocal �" ylocal The  de  scalping (partitioning brain from the non-brain tis-
sues [23]) is computed on I and Iref brain images. In order
color(0) = I(Ł#x ) to perform  de  scalping the following fast three  step algo-
Ś#,Ł#yŚ#
rithm designed by author was used:
color(1) = I(Ł#x ) " Find bones (skull) in CT image (simple threshold opera-
Ś#,Ł#yŚ#+1
tion)
color(2) = I(Ł#x ) " Find tissues other than bones (simple threshold operation)
Ś#+1,Ł#yŚ#
" Take as the result image only those tissues that are inside
color(3) = I(Ł#x ) closed region defi ned by bones.
Ś#+1,Ł#yŚ#+1
The result of this algorithm can be seen in Fig. 3.
Where I(x, y) is the value of color (gray level) of the
input image I. The complete registration algorithm goes as follows:
3
1. Calculate the optimal affi ne transformation with using
Iout (x, y) = Nelder  Mead algorithm
"color(i) * perc(i) (3.9)
i=0 2. Set the input mesh nodes number (nRows, nColumns)
The registration error value based on pixel distance (log) nRows := 8
between two images used by the author in the algorithm: nColumns := 8
n m 3. Initialize the grid nodes of deformation mesh with size
(3.10) nRows x nColumns.
(Ai - Bi +1)
""log j j
4. Set the number of grid refi ne operation to 2 and gradient
i=1 j=1
C(A, B) =
recalculation operation to 6
n �" m
refine := 2
The registration error can be also computed in image do- recalculation := 6
main: 5. For r: = 1 to refine + 1
([k,k + o], [l,l + p]) �" ([1,n], [1,m])
Set the resize factor
l+ p
k +o
r
resize _ factor :=
(Ai - Bi +1)
""log j j
refine +1
(3.11)
i=k j=l
C(A, B) =
k �"l
Fig 3. The example results of  de  scalping algorithm ((a) before (b) after  de  scalping )
Pattern Recognition
The registration and atlas construction of noisy brain computer tomography images& 47
" Resize image according to resize_factor, use gaussian The example results of registration of three brain images
smoothing on new created images in order to make optimi- are shown in Fig 6. The brain regions detected in images (Fig
zation process faster 6. e) are marked with a white line. The final mesh of control
" Compute the value of gradient movement constant: points (size 29 x 29) for the same three cases can be see in
Fig 7. In Fig 8.The differences between atlas images and the
Irows -1 Icolumns -1
ś#
� = 0.2 �"�# + new images are presented before (Fig 8. a) and after the reg-
ś# ź#
nRows -1 nColumns -1 #
�# istration (Fig 8. b).
Where: Table 1. presents the summary results of the image regis-
Irows , Icolumns is number of rows and columns in input tration process. The error (computed with (3.10) between new
matrix I. image and atlas image before and after registration) decreases
" For z := 1 to recalculation in all cases.
 Compute the gradient of cost function (3.11)
Table 1. Error before and after registration  error function (3.10)
"C(I, Iref ,Ś1), "C(I, Iref ,Ś2),..., "C(I, Iref ,Śn)
Ą# ń#
"C =
ó#
Error after
"Ś1 "Ś2 "Śn Ą# Number Error before
Ł# Ś#
registration registration
(using the central differential quotient method), each con-
1 0.9205 0.72356
trol point affects points only inside its 4 �"� � 4 �"� neigh-
borhood domain (Figure 4).
2 0.9826 0.79034
 Recalculate the control points of mesh grid
3 0.9922 0.82229
"C
Ś := Ś + � �"
4 0.9091 0.75868
"C
5 1.2056 1.0558
� := � �" 0.85
6 1.2028 1.0506
 Recalculate nodes of deformation mesh.
7 1.2454 1.1302
nRows := 2 �" nRows +1
8 1.0754 0.97403
nColumns := 2 �" nColumns +1
9 0.9874 0.80488
 Recalculate nodes of deformation mesh with size nRows x
nColumns. Each new node appears between two old one and 10 0.8958 0.74321
its position is interpolated with spline interpolation (Fig. 1).
11 0.9196 0.77166
" Compute deformation on new image according to defor-
12 0.7982 0.68605
mation mesh
13 0.7228 0.71876
I := splineTransform(I,Ś)
14 0.6822 0.5947
15 0.7531 0.65938
4. Experiment and discussion
16 0.6308 0.58591
The experiment was performed on 28 CT images, with 512 x
17 0.7750 0.67902
512 pixels resolution, from 8 different patients. Each image was
18 0.6528 0.57475
acquired from a series of dynamic brain perfusion CT scans
(often the first or one of the first images where the injected
19 0.6397 0.60927
contrast agent is still not visible). The aim of registration was to
20 0.5617 0.44583
generate the brain maps that will be accurate enough to name
21 0.8419 0.79936
some of the  main parts of brain in new images. In order to do
that, the  reference image of the algorithm described above
22 0.7263 0.67874
is now the new image, and the  image to registrate is one of
23 0.7795 0.72103
the set of  brain atlas reference image . This algorithm was im-
24 0.6470 0.6006
plemented in the Matlab environment and the C programming
language (compiled mex functions for faster performance).
25 1.0059 0.77045
The atlas consists of 12 brain CT images (the  reference
26 0.9757 0.75389
images for the registration process) and 12 images with la-
27 0.9507 0.81015
beled regions that describe some parts of brain (Fig 4, brain
region labels based on [18]). The template images were creat-
28 0.8895 0.77162
ed from a set of CT images acquired from one healthy patient
(similarly to original Talariach atlas templates that was based
on orthogonal sections acquired post mortem from only one The algorithm presented in this publication can be used as
60-year-old female subject [9]). a reliable tool for registration not only medical images but also
Images from dynamic brain perfusion CT treatment have any type of time-series or similar in shape image classes. After
more noises and movement artifact in them then  typical brain creating a set of atlas images and altering role of reference
CT images (Fig 5.) that makes process of registration even image with  moving image the deformable brain atlas system
harder.
Pattern Recognition
The registration and atlas construction of noisy brain computer tomography images&
48
Fig 4. Images form atlas, top two rows  raw CT images, bottom two rows  map of the brain parts, each color
(gray level) symbolize different brain area: symmetry axis (229), Frontal lobe (77), parietal lobe (153), temporal
lobe (29), occipital lobe (102), diencephalon (67), insula, insular lobe (135), mesence phalon; midbrain (188),
anterior lobe of cerebellum (37)
Fig 5. (a)  Plain CT image, (b) CT images registrated during dynamic CT perfusion treatment.
Noises and movement artifacts are clearly visible.
can be obtained. Images used in process do not need any prior tween brain hemispheres). There is also a difference between
preprocessing (in the case of brain optionally  de  scalping ). atlas- and  new images because atlas images are  plain CT
The main reason of the errors that appear even after the images and  new images came from perfusion CT treatment
registration process is that images are acquired from different and has more noises and movement artifact in them. These
patients and from different modalities (there is no possibility to problems can be solved by creating some kind of  eigenbrains
create 2D brain atlas images in all possible axial slices), be- like those in [10].
sides between patient difference there are also different radius The further research of the author will be focused on test-
in which patients lays they heads so different regions of the ing different similarity measures (i.e. based on Shanon infor-
brain can be see in the  same axial slices. Another reason of mation theory [8]) that can improve the accuracy of registration
errors that can be clearly seen in Fig 6. d is that affine trans- during rigid and non  rigid algorithm level. The second goal
form often finds local minimum of error function not performing will be testing the deformable atlas in diagnosing the set of
rotation of image correctly (see the twisted  symmetry axis be- perfusion CT brain maps and comparing it with medical docu-
Pattern Recognition
The registration and atlas construction of noisy brain computer tomography images& 49
Fig 6. The example results of the algorithm: (a) reference images of brain atlas (b) new images (c) adapted
(registered) brain atlas image (d) adapted (registered) brain atlases labeled maps (e) adapted (registered)
brain atlases labeled maps combined with input images.
Fig 7. The final mesh grid (size 29 x 29) that was used for deformation brain atlases in Fig 6 c) and d).
Fig. 8 Difference (a) before and (b) after the registration between images from Fig 6 a) and b).
Pattern Recognition
The registration and atlas construction of noisy brain computer tomography images&
50
mentation of each case in order to propose algorithms proper [13] Makoto Sasaki, Kohsuke Kudo, Hirobumi Oikawa, CT
for automatic image understanding ([16], [17]). perfusion for acute stroke: Current concepts on technical
aspects and clinical applications, International Congress
Series 1290 (2006) 30-36
References: [14] Aleksa Cenic, Darius G. Nabavi, Rosemary A. Craen,
Adrian W. Gelb, and Ting-Yim Lee, Dynamic CT Measure-
[1] Jeffrey C. Lagarias, James A. Reeds, Margaret H. Wright, ment of Cerebral Blood Flow: A Validation Study, January
Paul E. Wright, Convergence properties of the Nelder  1999
Mead simplex method in low dimensions , SIAM Journal of [15] Richard E. Latchaw, Howard Yonas, George J. Hunter,
Optimization, Vol. 9, No. 1, pp. 112-147, 1997 William T.C. Yuh, Toshihiro Ueda, A. Gregory Sorensen,
[2] William H. Press, Brian P. Flannery, Saul A. Teukolsky, Wil- Jeffrey L. Sunshine, Jose Biller, Lawrence Wechsler, Ran-
liam T. Vetterling, Downhill Simplex Method in Multidimen- dall Higashida, George Hademenos, Guidelines and rec-
sions, Numerical Recipes in C, Second Edition (1992), pp. ommendations for perfusion imaging in cerebral ischemia,
408-412 Stroke 34, (2003) 1084-1104
[3] John L. Semmlow, Biosignal and Biomedical Image Pro- [16] Ryszard Tadeusiewicz, Marek R. Ogiela, Automatic Image
cessing. MATLAB  Based Applications , CRC Press 2004 Understanding  A New Paradigm for Intelligent Medical
[4] D. Rueckert, L. I. Sonoda, C. Hayes, D. L. G. Hill, M. O. Image Analysis (opening article), BIO-ALGORITHMS AND
Leach, D. J. Hawkes, Nonrigid Registration Using Free- MED-SYSTEMS, Vol. 2, No. 3, 2006, pp. 5-11
Form Deformations: Application to Breast MR Images, [17] Ryszard Tadeusiewicz, Marek R. Ogiela, Structural Ap-
IEEE Transaction on Medical Imaging, Vol. 18, No. 8, Au- proach to Medical Image Understanding, Bulletin of the
gust 1999 Polish Academy of Sciences  Technical Sciences, Vol.
[5] Fumihiko Ino, Kanrou Ooyama, Kenichi Hagihara, A data 52, No. 2, 2004, pp. 131-139
distributed parallel algorithm for nonrigid image registra- [18] Antoine Micheau, Denis Hoa, Atlas of brain MRI cross-
tion, Parallel Computing, 2005 sectional anatomy, Anatomy of the whole human body:
[6] E. Bardinet, L. D. Cohen, N. Ayache, Tracking and motion medical imaging and pictures, 2008, http://www.imaios.
analysis of the left ventricle with deformable superquadrics, com/en/e-Anatomy
Med. Image Anal., Vol. 1, No. 2, pp. 129-149, 1996 [19] L. Ramirez, N. G. Durdle, V. J. Raso, Medical image regis-
[7] Adriane Parraga, Johanna Pettersson, Altamiro Susin, tration in computational intelligence framework: a review,
Mathieu De Craene , Benoit Macq, Non-rigid registration Electrical and Computer Engineering, 2003. IEEE CCECE
methods assessment of 3D CT images for head-neck ra- 2003. Canadian Conference on Volume 2, 4-7 May 2003,
diotherapy, Medical Imaging 2007: Image Processing, Vol. 2, pp. 1021-1024
2007 Macq [20] P.J. Phillips, Y. Vardi, S.M. Dunn, M.S. Buchsbaum, J.L.
[8] Julia A. Schnabel, Daniel Rueckert, Marcel Quist, Jane Spiegel-Cohen, An automatic MR-PET registration algo-
M. Blackall, Andy D. Castellano-Smith, Thomas Hartkens, rithm, Pattern Recognition, 1996, Proceedings of the 13th
Graeme P. Penney, Walter A. Hall, Haiying Liu, Charles International Conference on Volume 3, 25-29 Aug. 1996,
L. Truwit, Frans A. Gerritsen, Derek L. G. Hill, David J. Vol. 3, pp. 533-537
Hawkes, A Generic Framework for Non-Rigid Registration [21] Guorong Wu, Feihu Qi, Dinggang Shen, Learning-based
Based on Non-Uniform Multi-Level Free-Form Deforma- deformable registration of MR brain images, Medical Im-
tions, Medical Image Computing and Computer-Assisted aging, IEEE Transactions on Volume 25, Issue 9, Sept.
Intervention (MICCAI) 2001 2006, pp. 1145-115
[9] P. Thompson, M. Mega, K. Narr, E. Sowell, R. Blanton, A. [22] Simon K. Warfi eld, Jan Rexilius, Petra S. Huppi, Terrie E.
Toga, Brain image analysis and atlas construction, Hand- Inder, Erik G. Miller, William M. Wells III, Gary P. Zientara,
book of Medical Imaging, chapter 17. Pages 1066-1119, Ferenc A. Jolesz, Ron Kikinis, A Binary Entropy Measure
SPIE. 2000 to Assess Nonrigid Registration Algorithms, Fourth Inter-
[10] Usha Sinha, Hooshang Kangarloo, Principal Component national Conference on Medical Image Computing and
Analysis for Contentbased Image Retrieval, RadioGraph- Computer-Assisted Intervention, 2001 Oct 14-17, Utrecht,
ics 2002; 22: 1271-1289 The Netherlands, pp. 266-274
[11] M. R. Ogiela, R. Tadeusiewicz, Modern Computational [23] A. Purwar, R. Gupta, M. K. Sarma, G. Bayu, A. Singh, D.
Intelligence Methods for the Interpretation of Medical Ima- K. Rathore, S. Saksena, R. Trivedi, A. Mishra, M. Haris,
ges, Springer-Verlag, Berlin-Heidelberg, 2008 P. Mohan, R. Rathore, De-scalping of the brain in echo
[12] R. Tadeusiewicz, M. R. Ogiela, Medical Image Understan- planar DT-MRI, Proceedings of International Society of
ding Technology, Springer Verlag, Berlin-Heidelberg, 2004 Magnetic Resonance in Medicine 14, 2006
COMPUTER SCIENCE RADIOLOGY
Pattern Recognition
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 51-55
IMAGE PROCESSING APPLICATION FOR ENHANCEMENT OF
MEDICAL DIAGNOSTIC FEATURES
JOANNA JAWOREK, ELIASZ KAC
TOCH
Students Scientific Society  Implant
Faculty of Electrical Engineering, Automatics, IT and Electronics
AGH University of Science and Technology
Abstract: Automatic analysis of medical imaging is the permanently developed discipline requiring the interdisciplinary
collaboration linking medicine and computer science. The technique oriented on the increasing the resolution of the medical
images is presented.
Introduction
Nowadays medical images are a very important and very through the improvement of the image quality. The program s
popular source of information, used by medical doctors for interface can be seen in Figure 1 and contains the most com-
disease detection and location. Medical imaging refers to the mon image processing operations such as: binarization, filtra-
techniques and processes that present the human body in two tion and histogram equalization but also morphological opera-
and three dimensions[1]. Recently, a great advancement has tions (erosion, dilation, closing and opening) [3],[4]. A special
been made in the field of medical imaging. New techniques button  mammography launching a set of dedicated image
and processes are used to create images of the human body transformations simplifies the interpretation of the mammogra-
for clinical purposes, e.g. for visible range imaging better cam- phy images (Section 4). We can compose the operation chain
eras and new microscopy techniques have been developed. and change every single parameter of each operation.
Although the amount of medical images produced in the world A properly chosen sequence of operation is very helpful in
increases each year, their quality for diagnosis and therapy eliminating the majority of noise that always appears in the im-
remains suboptimal [1],[2]. In our application we propose the age (compare Figure 1 and Figure 2). It also helps in image
application of the latest medical imaging technologies and al- stabilization, by the extraction of the most important and sig-
gorithms to increase the quality of the images. nificant features (e.g. sharpening of the organ edges, separat-
This paper is arranged as follows. In section 2 we de- ing interesting organs from the rest of the image) [2]. All these
scribed the software application, image processing operations transformations can be important for the location of tumors or
and presented the algorithm on a flowchart. Section 3 shows other pathologies, study of the anatomical structures before
an example result of tomography. In Section 4 we described the surgery, and planning of the treatment.
a function dedicated for mammograms. Implementation and In Figure 3 a simplified flowchart [5] has been presented
statistical conclusion are presented in Section 5. Finally, Sec- including only the main steps. In every single moment while
tion 6 summarizes the conclusions and presents our plans for using the application, the operator can also use functions like:
the future. " save image, open image, delete image
" execute the list of operations, remove all chosen opera-
tions, remove a single operation
Image processing and the specification " compose the operation chain and change every single pa-
rameter of each operation
of the software application
" import operation chain from an XML fi le (default list)
" export the list of operation to an XML file
The aim of our research was to develop a software application
that will help doctors to diagnose a disease more efficiently,
Pattern Recognition
Image processing application for enhancement of medical diagnostic features
52
Fig. 1. Interface of the application with the input medical image of the backbone
Fig. 2. Output image of the backbone after negation and histogram equalization operation
An acceptable result Mammography image
Computer tomography was introduced in the 1970s and since Mammography is a low-power X-ray technique (about 0.7 mSv)
that time it has become a very important tool in medical imag- that examines the human breast and gives a picture of the in-
ing [1]. It is currently commonly used to detect many different ternal structure (Figure 6) [12]. These particular special pic-
diseases like: cancer, acute and chronic changes in the lung tures may help radiologists in the diagnosis of breast problems
and polyps, so it is very important to have a acceptable quality including cancer, milk ducts or small abnormal tissue [7]. Inter-
of the images [6]. For the input image let us take as an exam- pretations of mammograms can be problematic firstly because
ple abdomen tomography image (Figure 4), which is very un- a normal breast can appear differently for each woman and
clear. After applying a sequence of operations including: ne- secondly because of the quality and gray scale representation
gation and filtration, our output image (Figure 5) has changed of the tissue that makes the decision much more difficult.
dramatically.
Pattern Recognition
Image processing application for enhancement of medical diagnostic features 53
Fig. 3. Main diagram of the application
Fig. 4. Input image of the Computed Tomography (CT) scan of Fig. 5. Output image of the tomography usable to locate patholo-
cross section of the human abdomen. It is impossible to study the gies and study the structure thanks to the processing performed
anatomical structure
Pattern Recognition
Image processing application for enhancement of medical diagnostic features
54
The colors in the image make objects easier to distinguish
for doctors while diagnosing. In the new GUI we provide a but-
ton to enlarge the places which can contain abnormal tissue
(Figure 7). It is helpful for us by measuring the object and to
assess whether the tissue is abnormal or not.
Implementation and testing
Keeping in mind that the application was developed for doctors
and not for computer scientists, we provided a special button
for automatic selection of a file in the XML format [8] where the
most common sets of operations for the medical images are
stored, for example: computer tomography, MRI, ultrasonog-
raphy.
The application has been developed in the C++ language
with QT, OpenCV (Open Source Computer Vision Library)
[9],[10],[11] and Matlab. It is cross-platform and can be com-
piled for the operating systems from Windows and Linux fami-
Fig. 6. Regular mammography which shows the internal structure lies.
of the breast Testing of correct modification for each kind of image was
carried on several images (about 50). It is surely insufficient
to draw or write any statistical conclusion, however our input
Our proposal of the function chain dedicated for the inter- and output images were judged by doctors, which marked the
pretation of the mammography images contains the following quality of the output image as promising and helpful during the
steps: diagnostic of the diseases (Figure 8).
" Reading the input image
" Converting the image to a grayscale
" Filtration (reducing the noise) Conclusion and perspectives
" High-pass filtering
" Binarization Image processing offers many advantages increasing the
" Converting from intensity format to RGB quality of the image. It should never be forgotten that it is only
" Segmentation a computer application and the last decision should always be-
" Displaying the output image long to the doctor. In this paper, we propose an image process-
Fig 7. Interface of the application for mammography images. We can see the input and output image and the zoom function.
Pattern Recognition
Image processing application for enhancement of medical diagnostic features 55
Fig. 8. The quality of 50 output images judged by 20 doctors.
ing application that supports doctors in diagnosing. The authors 2. Tadeusiewicz R., Ogiela M.: Automatic Image Understan-
hope that it will be useful not only for clinical purposes but also ding New Paradigm for Intelligent Medical Image Analysis,
for young doctors and students, for which it is not always easy Bio-Algorithms and Med-Systems, 2006
to determine the diagnosis correctly. A second promising per- 3. Tadeusiewicz R., Flasiński M.: Rozpoznawanie obrazów,
spective is to broaden the amount of functions and the use of Warszawa 1991
new methods to automatically detect and mark small tumors or 4. Malina W.: Metody cyfrowego przetwarzania obrazów,
pathological images. Warszawa 2005
5. Wirth N.: Algorytmy + struktury danych = programy, War-
szawa 2002
Acknowlegements 6. http://en.wikipedia.org/wiki/Computed_tomography
7. http://www.nlm.nih.gov/medlineplus/ency/imagepages/
We would like to thank our research tutor dr hab. inż. Piotr Au- 1105.htm
gustyniak for inspiring discussions and scientific guidance and 8. McLaughlin B., Edelson J.: Java i XML. Wydanie III, 2007
the Department of Bioinformatics and Telemedicine Medical 9. Zieliński T.P.: Cyfrowe przetwarzanie sygnałów, Warszawa
College Jagiellonian University for supplying the images. We 2007
are also grateful to Tomasz Knap for helping us with the imple- 10. http://opencvlibrary.sourceforge.net/ (OpenCV library)
mentation of the software. 11. Ezust A., Ezust P.: Introduction to design patterns in C++
with Qt4, http://cartan.cas.suffolk.edu/oopdocbook/open-
source/
References 12. Chmielewski L., Kulikowski J.L., Nowakowski A.: Obrazo-
wanie biomedyczne, 2004
1. Cytowski J., Gielecki J., Gola A.: Cyfrowe przetwarzanie
obrazów medycznych. Algorytmy technologie zastosowa- Research tutor
nia, Warszawa 2008 dr hab. inż. Piotr Augustyniak
COMPUTER SCIENCE RADIOLOGY
Pattern Recognition
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 57-61
DIGITAL SKELETONIZATION AS A PROBE OF CORRELATION
BETWEEN SUTURAL BONES AND DIFFUSION LIMITED
AGREGGATION CLUSTERS
JANUSZ SKRZAT, JERZY WALOCHA
Department of Anatomy
Jagiellonian University, Medical College, Kopernika 12, 31-034 Kraków, Poland
Abstract: Skeletonization of the digital images of the sutural bones and computer generated clusters formed by diffusion
limited aggregation process (DLA) revealed branch-like patterns underlying the shape of these objects. Density of DLA cluster
is determined by the ratio of particle size to distance a particle moves in one step. Skeletons of DLA clusters aggregated from
particles whose size was 20 pixels show the most similar fractal dimension to the skeletons of sutural bones. The skeletons
obtained from DLA clusters formed by lesser (5, 10 pixels) or bigger (30, 40, 50 pixels) particles are less accurate to model basic
pattern underlying shape of the sutural bones because of divergences in their fractal dimension.
Key words: Diffusion limited aggregation, Wormian bones, Cranial sutures
Introduction
A digital skeleton is a compact representation that allows
mathematical analysis of the object. This is homotopic, thin
and median in relation to the object it represents. Skeletoniza-
tion is a technique, which allows to obtain basic structure of
the object and to preserve most topological information [7, 8,
12]. This technique is a convenient tool to get a simplified re-
presentation of shapes and therefore can help to understand
framework of composite objects like biological patterns.
Small bones found in human cranial sutures, particularly in
the lambdoid suture are example of intricate, irregular objects. Fig. 1. Close up view on sutural bones
Sutural bones also termed wormian bones are ossicles, which
develop from accessory ossification centers within cranial su-
ture and fontanelles. These inconstant bones vary in size and Morphological appearance of sutural bones resembles
number, which is generally limited to few or several entities, clusters that are obtained by the diffusion-limited aggregation
though they may exist as singular isolated bones [2, 3, 10]. algorithm. Diffusion-limited aggregation (DLA) is the process
Sutural bones have usually irregular shape, which often re- whereby particles undergoing a random walk due to Brown-
sembles stellate or branch patterns (fig. 1). ian motion cluster together to form aggregates of a branch-like
structures of fractal nature (fig. 2). As the cluster grows, there
is a greater probability for particles to stick to the ends than
to penetrate the interior. The shape of the DLA cluster is con-
trolled by the possibility of particles to reach the cluster. Some
Pattern Recognition
Digital Skeletonization as a Probe of Correlation between Sutural Bones and Diffusion Limited Agreggation Clusters
58
variations are also observed depending on the geometry of
the growth, whether it is from a single point radially outward or
from a plane or line for example [4, 11]. Diffusion-limited aggre-
gation as a model of pattern formation has attracted consider-
able attention since it was introduced by Witten and Sander in
1981. It was used not only to physical systems in the Laplacian
growth class but it has been also applied to biological systems
[1, 9].
Fig. 2. A typical cluster generated by diffusion-limited aggregation
algorithm containing 5000 particles.
Up till now, neither skeletonization nor diffusion-limited ag-
gregation procedures have been applied to investigate geo-
metrical properties of the sutural bones. Therefore purpose of
this study is to determine whether skeletonization of the im-
ages of the sutural bones lead to similar patterns, which can
be obtained from clusters formed by diffusion limited aggrega-
tion process. The second goal was to find out how far these
structures present similar complexity expressed by the value
of fractal dimension.
Material and methods
Sutural bones were found in the adult human skulls that are
housed in the Department of Anatomy of the Medical College
of the Jagiellonian University. Morphometric analysis of the 14
sutural bones located within lambdoid suture was performed in
a following way: first, the posterior aspect of the skull was pho-
tographed using digital camera (Canon EOS D30), zooming on
the sutural bones. Further their contours were traced as a 1
pixel line (0,35 mm) using the Graphire tablet (Wacom). Re-
solution of each image was 72 ppi. Enclosed contour of each
sutural bone was filled with black colour, saved as a grayscale
bitmap file and skeletonized using ImageJ software. This tech-
nique extracted a region-based shape feature representing
the general form of an object, according to the algorithm that
erodes repeatedly the image until only the skeleton remains
(fig. 3).
Digital skeletonization was also performed on images of
DLA clusters, which were generated on the computer with im-
Fig. 3. Subsequent steps leading to the digital skeleton of the su-
plemented ImageJ software, including Diffusion Limited Aggre-
tural bone; a  posterior view of the skull, sutural bone demar-
gate Models plugin (by A. Karperien and W. Rasband, Charles
cated by the rectangle, b  magnifi ed image of the sutural bone,
Sturt University, Australia) [7]. As a reference, we generated
c  isolated shape of the sutural bone, d  digital skeleton of the
DLA clusters, which vary in size of initial particle measured
sutural bone.
Pattern Recognition
Digital Skeletonization as a Probe of Correlation between Sutural Bones and Diffusion Limited Agreggation Clusters 59
by the number of pixels (px). The size of initial particles was
following: 5, 10, 20, 30, 40, and 50 pixels. Thus we obtained
6 sets of DLA clusters, each included 14 aggregates. All ag-
gregates were composed of round shaped particles, whose
number was arbitrary selected as 5000. The example of such
cluster and its digital skeleton shows fig. 4.
Next images (black skeleton, white background) obtained
from sutural bones and DLA clusters were subjected to the
fractal analysis using FracLac plugin for ImageJ software (de-
veloped by A. Karperien, Charles Sturt University, Australia)
[6]. Fractal dimension was calculated as box-dimension. The
equation for finding a fractal dimension (Df) approximates scale
and details of the object:
Ą#log N(� )ń#
=
D lim
f ó# Ą#
log�
� 0
Ł# Ś#
where the limit is found as the slope of the regression line for
the data, and N denotes number of pieces that compose the
object at scale �.
To assess if the fractal dimensions of digital skeletons ob-
tained from DLA clusters and sutural bones have the same
shape of distribution, Kruskal-Wallis test was applied and fol-
lowed by multivariate comparison test to find out statistically
significant differences among samples.
Fig. 5. The less and the most complicated skeletons of the DLA
clusters and the sutural bones, scored by fractal dimension (Df).
Fig. 4. Computer generated DLA cluster (size of aggregated par-
ticles  20 pixels, number of particles  5000) and its digital skel- 10, 20 and 30 pixels do not differ significantly from the distribu-
eton presented as 1 pixel thick segmented lines. tion of fractal dimension of the skeletons of the sutural bones
(table 2), the most adequate to simulate skeletons of the su-
tural bones seems to be DLA clusters aggregated from par-
Results ticles of size 20 pixels. The skeletons of these clusters yield
fractal dimensions (1,16  1,25), which completely fits within
Digital skeletons of the sutural bones and DLA clusters show the range of variation of fractal dimension (1,09  1,32) esti-
similar branch-like pattern but of various complexity expressed mated for skeletons of the sutural bones. In this case, the dif-
by fractal dimension (fig. 5). Estimated fractal dimension of ference between mean fractal dimension (1,20 for skeletons of
the digital skeletons of the 14 sutural bones range from 1,09 DLA clusters and 1,18 for skeletons of the sutural bones) is the
 1,32 with median 1,16 and mean equal to 1,18. Skeletons smallest (" = 0,02).
obtained from generated DLA clusters deliver bigger span of Skeletons of DLA clusters made of particles, whose size
fractal dimension (1,03  1,50) than those of the sutural bones. is 30 pixels, yield lower mean fractal dimensions (Df = 1,12)
However, in each group of the DLA clusters fractal dimension than skeletons of the sutural bones (Df = 1,20) and thus big-
is less variable than in the case of sutural bones (table 1). ger difference (" = 0,08). Hence, range of variation of fractal
In the light of the Kruskal-Wallis test (H = 88, 30; p < 0,05) dimensions estimated for the skeletons of these clusters (1,07
analyzed samples do not come from the same distribution  1,18) correspond partially only to the variation of the fractal
(fig. 6). The result of multiple comparison test (table 2) indi- dimension presented by the skeletons of sutural bones (1,09
cates statistically significant difference between fractal dimen-  1,32).
sions of skeletons of the sutural bones and skeletons of the On the other hand, skeletons of DLA clusters obtained
DLA clusters aggregated from the particles, which size was 5, from smaller particles (10 pixels) show higher mean fractal di-
40 and 50 pixels. mension (Df = 1,33) than skeletons of the sutural bones (Df =
Although the distributions of fractal dimensions of the skel- 1,18) what makes the biggest difference between means ("
etons of DLA clusters generated from particles, whose size is = 0,15). Hence, the range of variation of fractal dimension of
Pattern Recognition
Digital Skeletonization as a Probe of Correlation between Sutural Bones and Diffusion Limited Agreggation Clusters
60
Fig. 6. Medians and variation of the fractal dimensions of the digital skeletons  DLA clusters versus sutural bones (sb).
Table. 1. Statistics for fractal dimensions of the digital skeletons of the DLA clusters and the sutural bones (sb).
Cluster Sample Median Mean SD Min Max
5 px 14 1,48 1,47 0,015 1,45 1,50
10 px 14 1,34 1,33 0,015 1,30 1,35
20 px 14 1,21 1,20 0,024 1,16 1,25
30 px 14 1,12 1,12 0,033 1,07 1,18
40 px 14 1,07 1,07 0,018 1,03 1,10
50 px 14 1,06 1,06 0,021 1,03 1,10
sb 14 1,16 1,18 0,070 1,09 1,32
Table 2. Results of multiple comparison test for fractal dimensions of the digital skeletons of DLA clusters and sutural bones (sb).
Statistically significant p < 0,05.
Cluster 5 px 10 px 20 px 30 px 40 px 50 px sb
5 px  1,0000 0,0469 0,0000 0,0000 0,0000 0,0047
10 px 1,0000  1,0000 0,0024 0,0000 0,0000 0,3844
20 px 0,0469 1,0000  0,6906 0,0039 0,0005 1,0000
30 px 0,0000 0,0024 0,6906  1,0000 0,7377 1,0000
40 px 0,0000 0,0000 0,0039 1,0000  1,0000 0,0401
50 px 0,0000 0,0000 0,0005 0,7377 1,0000  0,0065
sb 0,0047 0,3844 1,0000 1,0000 0,0401 0,0065 
Pattern Recognition
Digital Skeletonization as a Probe of Correlation between Sutural Bones and Diffusion Limited Agreggation Clusters 61
these clusters (1,30  1,35) overlap only higher values of frac- References
tal dimensions estimated for skeletons of sutural bones (1,09
 1,32). 1. Barra F., Davidovitch B., Levermann A., Procaccia I., Lapla-
DLA clusters composed of particles which size was 5, 40 cian growth and diffusion limited aggregation: different
and 50 pixels yield fractal dimensions, which are different sta- universality classes. Phys Rev Lett 87: 134501-134504,
tistically from fractal dimensions of the skeletons of the sutural 2001.
bones. Therefore these clusters do not properly model basic 2. Bennett K.A., The etiology and genetics of wormian bones.
pattern of the examined sutural bones. Am. J. Phys. Anthropol. 23: 255-260, 1965.
3. Berry A.C., Berry R.J., Epigenetic variation in the human
cranium. J. Anat. 101: 361-379, 1967.
Conclusions 4. Halsey T.C., Diffusion-limited aggregation: A model for pat-
tern formation, Physics Today 53: 36, 2001.
Diffusion-limited aggregation is a fractal algorithm that can pro- 5. Karperien A., Fractal dimension and Lacunarity. FracLac
duce similar patterns, which can be found among the sutural plugin for ImageJ: http://rsb.info.nih.gov/ij/ plugins/fraclac/
bones. Digital skeletons of the sutural bones and computer fraclac.html, 2007.
generated DLA clusters present branch-like shapes but of vari- 6. Karperien A., Rasband W., Diffusion Limited Aggregated
ous fractal dimension. Complexity of the DLA cluster skeletons Models: http://rsb.info.nih.gov/ij/plugins/DLA.html, 2006.
changes according to the initial size of the aggregating parti- 7. Lee S.W., Lam L., Suen C.Y., A systematic evaluation of
cles. Cluster composed of small particles shows higher fractal skeletonization algorithms. Int. J. Pattern Recognition Arti-
dimension than clusters aggregated from bigger particles. Size fi cial Intelligence 7: 1203-1225, 1993.
of particles 20 pixels found as the most adequate to simulate 8. Murthy I.S.N., Udupa K.J., A search algorithm for skeleton-
skeletons of the sutural bones corresponds to the diameter ization of thick patterns. Computer Graphics Image Pro-
of distal finger-like processes that are peripheral elements of cessing 3: 247-259, 1974.
the sutural bones. This could imply that random process may 9. Noeim F., Moatamed F., Sahimi M., Morphogenesis of
underlie sutural bone formation and similarities in the digital the bone marrow: fractal structures and diffusion limited
skeletons may suggest common mechanism of their formation growth. Blood 87: 5027-5031, 1996.
based on fractal algorithm. Branch-like shapes of the sutural 10. Pal G.P., Routal R.V., A study of sutural bones in different
bones supports an idea of application of fractal geometry in morphological forms of skulls. Anthrop. Anz. 44: 169-73,
description of their morphology. Hence, relationship between 1986.
skeletons of the sutural bones and DLA cluster may shad new 11. Witten T.A., Sander L.M., Diffusion-limited aggregation.
light on the style of pattern formation during cranial morpho- Phys. Rev. B 27: 5686-5697, 1983.
genesis. 12. Wright M.W., Cipolla R., Giblin P.J., Skeletonization using
an extended Euclidean distance transform. Image Vision
Computing 13: 367-375, 1995.
ANATOMY COMPUTER SCIENCE
MEDICINE STATISTICS
Pattern Recognition
BIO-ALGORITHMS AND MED-SYSTEMS
JOURNAL EDITED BY MEDICAL COLLEGE  JAGIELLONIAN UNIVERSITY
Vol. 4, No. 7, 2008, pp. 63-70
LEARNING-BY-E-TEACHING: EXPERIENCE FROM INVOLVING
STUDENTS IN PREPARATION OF E-LEARNING MATERIALS
ANDRZEJ A. KONONOWICZ, ALEKSANDRA J. STACHOC
,
ANNA ROMANOWSKA-PAWLICZEK, PIOTR OBTUA
OWICZ, WIESA
AW PYRCZAK
Department of Bioinformatics and Telemedicine, Madical College, Jagiellonian University,
31-530 Krakow, Św. A
azarza 16, Poland, www.bit.cm-uj.krakow.pl
Abstract: Creation of e-learning materials is considered as a time consuming and costly process. Involving students in preparation
of educational content may be benefi cial in many ways. Firstly, it motivates students to self-study following a similar pattern as
it is the case in learning-by-teaching sessions. Secondly, some of the students projects may be reused in development of more
advanced e-learning resources. The aim of this paper is to summarize three-year experiences gained in two elective courses
on e-learning technologies offered to students of medicine and applied computer science. The spectrum of topics taken up by
students is outlined. Possibilities of applying the outcomes of these courses in practice are discussed.
Introduction Method
Courses on e-learning technologies are still quite rare in aca- The Department of Bioinformatics and Telemedicine at Jagiel-
demic curricula. Considering the fact that the demand on e- lonian University Medical College offers its students two elec-
learning resources worldwide has increased significantly over tive courses on creating e-learning resources. These courses
the past few years it becomes obvious that this situation re- are geared toward two different target groups: students of
quires a change. In this context it seems to be worthwhile to medicine and students of applied computer science.
make an attempt to involve students not only in passive usage The elective course on e-learning technologies offered
of e-learning materials but also in creation of their content. to medical students is organized around elementary IT tech-
nologies useful for creation of electronic educational materials.
Within this course, students of the fourth year learn principles
Background of web languages (HTML), authoring environments (Macrome-
dia1 Dreamweaver) and computer graphics (Corel Paint Shop
The idea to involve students in preparation of didactic re- Pro). Outcomes of this activity are e-learning web-pages on
sources derives from the long-established tradition of Lear- selected medical topics created single-handedly by groups of
ning-by-Teaching (LbT). Various studies have already reported students (consisting of 1-4 students). Our e-learning tutors are
outcomes of including senior students (4-6 year) in teaching encouraged to search for information concerning a self-chosen
activities organized for junior students (1-3 year). Effects of medical topic, transform and integrate it into a web-adjusted
peer-assisted learning (PAL) have been tested among others form, which is destined for a previously selected target group
in Physiotherapy [6], Diversity Issues in Healthcare [7], Clini- (e.g. lay person, patients, students of medicine, specialists).
cal Examination [1] and Early Patient Contact [2] programmes. The course offered to fourth year students of applied com-
J. Secomb [5] presents in her work a systematic review of peer puter science is more technical-oriented. In this course an
teaching and learning in clinical education. The cited work, fo- overview of e-learning tools and standards is made. One of the
cusing mainly on the nursery discipline, summarizes the key most interesting, due to its multidisciplinary character, tasks
benefits of learning-by-teaching  e.g. increase in cognitive assigned to students of computer science is to prepare anima-
development, improvement in students motivation for learn- tions in Flash technology on a selected biological process or
ing and building up confidence in clinical practice. In almost all medical procedures.
cases the polled students were satisfied with their experiences
gained in peer teaching and learning sessions.
1
Currently Adobe
Telemedicine
Learning-by-e-teaching: Experience from involving students in preparation of e-learning materials
64
Additionally, students might participate in a students scien- e-learning projects. The course for students of computer sci-
tific workgroup as an extracurricular activity. Discussing there ence took place for the first time last year (2007) and was com-
many telemedicine related subjects, some of them took also pleted by twelve students. Fourteen Flash animations have
the opportunity to create new e-learning sites or refine existing been created within the course. The student s scientific work-
ones which were prepared within e-learning courses. group delivered four additional web projects. Fig.1 presents an
Significant differences between a classical Learning-by- overview of graphical layouts taken from selected web pages
Teaching method and the approach selected for our study can and animations created by students within these e-learning
be noticed. E-learning does not require a direct link between courses.
the teacher and students. Educational resources in electronic Students took up many interesting and actual subjects from
form replace usually the personal contact with a tutor. In con- the field of healthcare. Among these were e.g. physiology and
sequence senior students taking part in our activities were not pathology of cardiovascular system, digestive system, nervous
brought in direct contact with their target groups but created system and vertebral column. Projects concerning treatment
their work  off-line for future use. methods of heart related problems were present, as well as
This study reports on the content obtained as a result of materials about resuscitation methods (e.g. hearth massage).
students e-learning activities in the Department of Bioinformat- Students took a keen interest in topics of fertility, contracep-
ics and Telemedicine. Unlike many previous studies, this pa- tion and HIV/AIDS (fig. 2). Some of the students web-pages
per does not concentrate on quantitative or qualitative aspects and presentations explain basic biological structures and pro-
proving effectiveness of Learning-by-Teaching or examining cesses as urine production, circulation of iron in human and
students satisfaction with peer teaching. Instead, the aim of protein s synthesis. Also such noteworthy topics as carbon
this paper is to outline the spectrum of topics taken up by stu- monoxide poisoning, drugs addiction and history of medicine
dents and to discuss the possibilities of applying the outcomes were described by students.
of these courses in practice.
Tab. 1 summarizes the branches of medicine covered by
students in their works. From a large set of projects prepared
Results by students within LbT sessions a few examples has been
chosen and described below.
In the years 2005-2008, forty-five students completed the elec-
tive course destined for medical student delivering nineteen
Fig. 1. Overview of graphical layouts taken from selected web pages and animations
created by students within our e-learning courses [13-22].
Telemedicine
Learning-by-e-teaching: Experience from involving students in preparation of e-learning materials 65
Fig. 2. HIV s lifecycle in Flash animation [23].
of this global pandemic it is crucial to make people aware of
Tab. 1. Specification of selected topics in e-learning courses. causes of cardiovascular illness, develop methods enabling
their early diagnosis and effective treatment as well as promote
No. of web No. of
healthy lifestyle.
Branch of medicine
pages animations
The most frequent among various cardiovascular disorders
is the ischemic heart disease. This medical condition often
Cardiology 2 1
leads to heart attack and death. Prophylaxis and treatment
Gynaecology 2 -
of atheroembolism is therefore an interesting topic for future
doctors. Coronary angioplasty, a common treatment method of
Neurology 2 -
myocardial ischemia, has been described by students in form
of a web-page entitled miłE-PTCA (nicE-PTCA). PTCA  an
Oncology 1 -
acronym of Percutaneous Transluminal Coronary Angioblasty
Infectious Diseases 1 1
 is a medical procedure consisting in recanalization of stenotic
artery with use of small balloon and stent. This method is cur-
Dermatology 1 -
rently very common and is regarded as an alternative method
Internal Medicine 6 1
to various kinds of myocardial ischemia therapy [4].
Authors of the above described a web-page which com-
Surgery 1 -
bined in an interesting way specialist knowledge with informa-
Paediatrics 1 - tion for lay readers. Their major goal was to familiarize patients
waiting for PTCA with concepts related to this procedure.
Emergency medicine - 5
Content of the miłE-PTCA has been logically and comprehen-
sible prepared. It includes basic knowledge (e.g. anatomy of
Toxicology 3 -
coronary vessels), subject related information (e.g. definition
Basic Sciences 1 7
of PTCA, recommendations and contradictions) and practi-
cal advices useful for patients (e.g. preoperative preparation,
Others 1 -
postoperative recommendations). Web page visitors may learn
that angioplasties undergo annually several thousands people
Cardiovascular diseases are the number one cause of in Poland, and about half million worldwide. Mortality rate in
death worldwide. High rates of morbidity are observed both in this kind of operation is less than 1%, whereas effectiveness
industrialized and developing countries. WHO experts estimate reaches even 90% [11].
that in the course of the next few years 20 millions people will Other heart related problems discussed by students are
die yearly due to cardiovascular diseases [8]. In Poland car- congenital heart diseases, prepared in the form of a web-page:
diovascular diseases cause half of all deaths [9,10]. Because E-Wady Serca (E-heart defects, fig. 3). This site is destined for
Telemedicine
Learning-by-e-teaching: Experience from involving students in preparation of e-learning materials
66
medical students and to parents/carers of children with heart
disorders. Pictures of cardiovascular system illustrate mecha-
nisms leading to pathological symptoms of such disorders as:
atrial septal defect (ASD), ventricular septal defect (VSD), pat-
ent ductus arteriosis  Botall s duct (PDA), tetralogy of Fallot
(TOF) and hypoplastic left heart syndrome (HLHS). The re-
viewed web presentation has a clear educational character.
In the introductory part normal physiological blood circulation
is outlined. Characteristics of the above mentioned congeni-
tal heart disorders include information on their spread, typical
symptoms and various treatment methods. A closer look at the
web-page will give the opportunity to discover many interesting
facts like that the frequency of some cardiac abnormalities is
higher in females than males (ASD, VSD, PDA), or a detailed
explanation of the Norwood procedure (i.e. a series of opera-
tions in HLHS) [3].
The presented page is also a practical example on that
how projects of students from different background may com-
plement each other. A static diagram proposed by medical stu-
Fig. 6. Examples of students projects: Resuscitation [27].
dents showing blood circulation in physiological and various
pathological states (fig. 3) has been extended into an animated
version by a student of computer science (fig. 4).
Some of the students attending courses on e-learning Discussion
technologies went into the subject of first aid techniques (ani-
mations fig. 5, fig. 6). This topic is always of great importance Students involvement in preparation of educational e-learning
and up-to-date, but knowledge in this subject still not enough materials (medical web-pages and Flash animations) moti-
widespread. Life-saving techniques should be taught not only vates to self-directed information search, logical reasoning and
to medical students but also be addressed to lager parts of the concept association. The proposed activities gave students
society. possibility to present both their knowledge and artistic skills.
An excellent example of that is a web-page on neurophysiol-
ogy which includes original illustrations of high quality drawn
single-handedly by a student (fig. 7).
Fig. 3. A web-page within an e-LbT session discussing the subject of congenital heart disorders [24].
Telemedicine
Learning-by-e-teaching: Experience from involving students in preparation of e-learning materials 67
Fig. 4. Enrichment of medical e-learning pages by students of computer science [25].
Fig. 5. Examples of students projects: First aid tutorial [26].
Telemedicine
Learning-by-e-teaching: Experience from involving students in preparation of e-learning materials
68
Fig 7. Web-page in neurophysiology [28].
Students creating their own educational materials are the faculty in many different ways. One of the possibilities is
believed to memorize better the presented information. The to present these materials on students portals or e-learning
strong IT character of this class may be perceived as an inspi- platforms as a source for self-directed learning. A further option
ring change for medical students not used to this form of lear- is to apply this content for creation and enrichment of more
ning. We believe that it has the potential to increase positive advanced e-learning resources like virtual patients. Indexed
attitude to science and foster the ability of life-long learning. learning objects could be gathered in specialized repositories
Students seemed to be keen on taking up this kind of activity and reused by the faculty as extension of traditional lectures
and in most cases they performed well in preparing e-learning or textbooks. After profound review it is also thinkable to offer
materials. A few of them presented their projects at students these materials as informational material for patients or their
scientific conferences. Among them were two consecutive In- families. Before publishing, it is of course necessary to obtain
ternational Scientific Conferences of Medical Students held authors consent and clarify the intellectual property rights
in Krakow (2007, 2008). The students web-pages have been (IPR) of all content available in these educational materials.
presented within Telemedicine & Biocybernetics sessions. Inclusion of best projects in teaching practice at the univer-
Some of the projects are presented on [12]. sity raises also some concerns. Faculty teachers in charge of
Students activity during e-learning courses and later on the e-learning classes have technical background and assisted
the presentation of their work at conferences gave them the the students mainly in technology related questions. Therefore,
chance to consolidate and broaden their bio-medical knowl- all students projects should undergo before publishing a de-
edge. These activities constitute also great opportunity to inter- tailed review by the clinical teachers. With the increased num-
change ideas and experiences achieved as a result of self-pre- ber of projects, providing additional incentives for the clinical
paring educational materials as well as writing and presenting faculty staff to do the review could be very helpful.
scientific dissertations. Students had also a chance to present When looking at the courses from the technical perspec-
their work to medical specialists, who evaluated their activities. tive it can be noticed, that they were organized on the basis
Some of the e-learning courses prepared by students at the of professional authoring tools like Dreamweaver or Flash.
Department of Bioinformatics and Telemedicine JUMC are reg- This software is on the one hand very powerful and flexible,
ularly updated and expanded. They seem to have the potential but on the other also expensive and (for some students) dif-
to be an interesting extension to traditional learning activities. ficult to learn. Working on the level of HTML editors, even
The obtained results are good examples of practical con- of WYSIWYG2 type, may seem nowadays to be a too deep-
nection of technology and medicine. E-learning materials level approach. However, relying solely on the authoring tools
prepared by medical students were taken in many cases as available in Learning Content Management Systems (LCMS),
a knowledge base for computer science students while prepar- like Blackboard or Moodle, may create portability problems.
ing their animations. The ready animations enrich the existing
pages ensuring their constant evolution.
2
WYSIWYG Editor (What-You-See-Is-What-You-Get) is an editor
Well prepared animations and web-pages taken from stu-
class enabling visual authoring of web pages without the need to deal
dents projects may be re-used by peer students and also by
directly with HTML code.
Telemedicine
Learning-by-e-teaching: Experience from involving students in preparation of e-learning materials 69
E-learning interoperability standards (like SCORM or IMS) are żenia w rocznej obserwacji, Postępy w Kardiologii Inter-
in the authors belief not yet the perfect remedy for this prob- wencyjnej, 2/3(5), 199-206, 2006.
lem. More technically apt students may in addition find the [5] Secomb J.: A systematic review of peer teaching and lear-
technical constraints of LCMS limiting. The huge differences ning in clinical education, 17(6), 703-716, 2008.
in IT-knowledge among medical students make it additionally [6] Solomon P., Crowe J.: Perceptions of student peer tutors
difficult to target the classes. in a problem-based learing programme, Medical Teacher,
23(2), 181-186, 2001.
[7] Tang T., Hernandez E., Adams B.:  Learning by Teaching :
Further work A Peer-Teaching Model for Diversity Training in Medical
School, Teaching and Learning in Medicine, 16(1), 60-63,
All web-pages and animations created within the course have 2004.
been gathered on one single e-learning portal, which is cur-
rently waiting for a faculty review. Projects which have been Web pages
accepted are intended to be published on-line. A next study will
concentrate on examination of the educational impact of these [8] WHO Homepage, Cardiovascular diseases  http://www.
web-pages on junior students. who.int/topics/cardiovascular_diseases/en/ (last visited
Immense progress in e-learning technologies makes it 9.05.2008)
necessary to update the course annually. A modification of the [9] Polskie Towarzystwo Kardiologiczne, Epidemiologia cho-
programme to include more user-friendly, free-of-charge soft- rób układu krążenia  http://www.ptkardio.pl/pl/archiwum/
ware should definitely be considered. 213.html, In Polish (last visited 9.05.2008)
[10] Poradnik Medyczny, Choroby układu krążenia w Polsce
 http://www.poradnikmedyczny.pl/mod/archiwum/4779.
Summary html, in Polish (last visited 9.05.2008)
[11] Resmedica, Angioplastyka w leczeniu choroby wieńco-
The study presented outcomes of two elective courses on e- wej  http://www.resmedica.pl/zdart3991.html, in Polish
learning technologies destined for students of medicine and (last visited 9.05.2008)
applied computer science. Senior medical students created [12] Koło Telemedycyny i Biocybernetyki  http://www.teleme-
their own e-learning web sites from various medical fields (e.g. dycyna.cm-uj.krakow.pl, In Polish (last visited 9.05.2008)
cardiology, internal medicine or neurology) geared towards dif-
ferent target groups. Students of computer science extended Authors of described projects
the web sites by creating Flash animations. Taking the role of
content authors, students practiced how to structure knowl- [13] Liszka M., Pastuszak M., Sporek M., 2006. Web page An-
edge for learning. Examples of future application of the gath- tykoncepcja.
ered resources were given (e.g. self-directed study, virtual [14] Wolak P., 2006. Web page Historia medycyny.
patients, extension of lectures). The authors would welcome [15] Opałka M., Gontkiewicz A
., Nawrot M., 2008. Web page
to collaborate with medical teachers in the review of the edu- Choroba Alzheimera.
cational content created by students and their later application [16] Godyń P., 2007. Animation Endogeneza trzustki.
in didactic practice. [17] Tomanek M., 2007. Animation Ruchy jelit.
[18] Paluch J., 2007. Animation Synteza białka  translacja.
[19] Kuśmierski P., 2007. Animation Proces powstawania mo-
References czu.
[20] Walas M., 2007. Animation Reumatoidalne zapalenie sta-
[1] Field M., Burke J., McAllister D., Lloyd D.: Peer-assisted wów.
learning: a novel approach to clinical skills learning for me- [21] Opoczyński P., 2007. Animation Bandażowanie ręki.
dical students, 41(4), 411-418, 2007. [22] Brodecki K., 2007. Animation Skurcz pęcherzyka żółcio-
[2] Glynn L., MacFarlande, Kelly M., Cantillon P., Murphy wego.
A.: Helping each other to learn  a process evaluation of [23] Tomanek M., 2007. Animation Etapy cyklu życia wirusa
peer assisted learning, 6(18), http://www.biomedcentral. HIV-1.
com/1472-6920/6/18. [24] Mirek A., Czajka M., Popławski M., 2006. Web page E-
[3] Moszura T. i in.: Zastosowanie stentów wewnątrznaczynio- Wady Serca.
wych w wieloetapowym leczeniu dzieci z zespołem niedo- [25] Pękala I., 2007. Animation Wady serca.
rozwoju lewego serca  doświadczenia własne, Kardiochi- [26] Godyń P., 2007. Animation Przywracanie oddechu.
rurgia i Torakochirurgia Polska, 4(2), 148-154, 2007. [27] Opoczyński P., 2007. Animation Masaż serca.
[4] Pawłowska-Jenerowicz W., Lechowicz W., Dąbrowski M.: [28] Potocki P., 2006. Web page Neurofizjologia.
Korzystny wpływ leczenia angioplastyką wieńcową ostrej [29] Lubojemska P., Zając T., Sarnecki B., Polańska E., 2006.
fazy zawału mięśnia sercowego na wydolność układu krą- Web page miłE-PTCA.
Telemedicine
Learning-by-e-teaching: Experience from involving students in preparation of e-learning materials
70
MEDICAL STUDENT COMPUTER SCIENCE STUDENT
PREPARATION OF MEDICAL PREPARATION OF MEDICAL
WEB-PAGES ANIMATIONS
TEACHER
EVALUATION OF E-LEARNING
MATERIALS
Telemedicine
1. The original and one Photostat copy of the manuscript should be mailed to: Managing Editor Zdzisław Wisniowski.
Authors are strongly urged to submit a CD containing the manuscript in Word for PCs format along with the hard
copies. Articles already published or those under consideration for publication in other journals or periodicals should
not be submitted.


Wyszukiwarka

Podobne podstrony:
Bio Algorythms and Med Systems vol 5 no 10 2009
Cwiczenia Zarzadzanie w systemie Windows Server 2008
04a?5 Power Supply and Bus Systems
(Ebook Occult) Oto Agape Vol 2 No 4
Audel Hvac Fundamentals, Air Conditioning, Heat Pumps And Distribution Systems (Malestrom)
Magazine Fantasy and Science Fiction [Vol 111]
The new drilling control and monitoring system
03a?6 Power Supply and Bus Systems
Alchemy Journal Vol 1 No 1
TAB 6 Navigational Safety Chapter 28 Global Maritime Distress and Safety System
Alchemy Journal Vol 2 No 4
87 151 Heating and ventilation system servicing
Alchemy Journal Vol 4 No 2
Alchemy Journal Vol 3 No 1
Magazine Fantasy and Science Fiction [Vol 110] Issue 05 May 2006 (v1 0) [html]

więcej podobnych podstron