USING MOBILE AGENTS IN EEG SIGNAL PROCESSING

Roman Mouček and Petr Šolc

Department of Computer Science and Engineering, University of West Bohemia

Univerzitní 8, 306 14 Pilsen, Czech Republic

Keywords: Electroencephalography (EEG), Event related potentials (ERP), EEG/ERP data, EEG/ERP processing

methods, EEG data formats, Artefacts detection, Mobile agents, Java, System aglets.

Abstract: Our EEG/ERP repository contains large EEG/ERP data. The partner institutions would like to work with

these data e.g. to verify their processing methods, but they cannot or they are not allowed to transfer large

data collections over network. The possible solution is to use a mobile agent system. The paper briefly

introduces the agent system Aglets and basic EEG processing methods implemented as mobile agents. The

usability of this approach is tested on selected data files. The implementation of e-mail announcements

within the mobile agent system is mentioned.

1 INTRODUCTION

Our research group at Department of Computer

Science and Engineering, University of West

Bohemia in cooperation with other partner

institutions (Czech Technical University in Prague,

University Hospital in Pilsen, etc.) specializes in the

research of attention, especially attention of drivers

and seriously injured people. With regard to our

research we widely use the methods of

electroencephalography (EEG) and event related

potentials (ERP). Within our partner network we are

responsible for technical and scientific issues, e.g.

EEG/ERP laboratory operation, development of

advanced software tools used in EEG/ERP research,

or analysis and proposal of signal processing

methods.

EEG/ERP experiments take usually long time

and produce a lot of data (sizes of data files are

usually from tens to hundreds of megabytes). These

data, finally stored on a computer in EEG/ERP

laboratory, are then analyzed, converted and

processed using various software tools and

processing methods. Recently, a common tree of

directories and subdirectories served as an

organization scheme for EEG/ERP data. Nowadays,

the first prototype of EEG/ERP database has been

developed. In parallel, we found out that transfer of

huge EEG/ERP data over network could be still an

important trouble for some partner institutions.

Moreover, the suggested network load in our

laboratory would be also enormous.

An easy access to large EEG/ERP data and

a possibility to work with them using own

processing methods are very important requirements

of our partner institutions. The following process is

expected. The partner institution wants to work with

EEG/ERP data, e.g. to verify a processing method

on a large collection of EEG/ERP data stored in our

repository. They cannot or they are not allowed to

transfer this large data collection over network.

Therefore they use data from the repository (create

their local copy) and transfer their software code to

perform a requested operation (e.g. matching pursuit

algorithm). When the operation is finished the result

is announced to user by e-mail.

A possible solution how to ensure this kind of

remote processing of EEG/ERP data is to use the

system of mobile agents based on two ideas: data are

local, and operating software code (agent) is

transferred. Because an agent system serves for

EEG/ERP research (construction of a specific agent

system was not the task), there was important to

look for a system, which is at least proven in some

different domain and seems to be promising for our

needs. Other necessary conditions include

possibilities of easy deployment on various

computer platforms and extensibility of the selected

agent system (open source software).

The following parts of the paper briefly

introduce the selected agent system and some basic

EEG processing methods implemented as mobile

agents. The most important part of the paper

describes the testing of usability of the selected

agent system for our research. The running time of

155

Mou

cek R. and Šolc P. (2010).

USING MOBILE AGENTS IN EEG SIGNAL PROCESSING.

In Proceedings of the 2nd International Conference on Agents and Artiﬁcial Intelligence - Agents, pages 155-158

DOI: 10.5220/0002706301550158

 SciTePress

EEG signal processing methods implemented in one

way as common Java applications and secondly

within a corresponding mobile agent system is

compared. The implementation of e-mail

announcements within the mobile agent system is

mentioned.

2 SYSTEM AGLETS

According to our needs and general

recommendations (Šolc, 2009) we choose the system

Aglets from the variety of existing mobile agent

systems.

System Aglets is a Java mobile agent platform

and library that eases the development of agent

based applications. An aglet is a Java agent able to

autonomously and spontaneously move from one

host to another.

System Aglets was originally developed at the

IBM Tokio Research Laboratory, the Aglets

technology is now hosted at sourceforge.net as open

source project, where it is distributed under the IBM

Public License.

System Aglets includes both a complete Java

mobile agent platform, with a stand-alone server

called Tahiti, and a library that allows developer to

build mobile agents and to embed the Aglets

technology in their applications. The system is based

on callback model (Lange, 1998; “Aglets”, 2004).

3 EEG SIGNAL PROCESSING

METHODS

We selected a limited set of methods for

experimental implementation of mobile agents:

 Methods for conversions between EEG/ERP

data formats;

 Basic methods for detection of artefacts in

EEG signal.

3.1 Data Formats

There exists a variety of data formats for storing

EEG/ERP data. The more spread formats and

formats used in our laboratory include e.g. European

Data Format (EDF and EDF+) (“EDF”, n.d.), Vision

Data Exchange Format (VDEF) (“VDEF”, n.d.), and

KIV format (Kučera, 2008).

European Data Format (EDF) contains an

uninterrupted digitized EEG record stored in one file

(a header record is followed by data records). The

header content has a variable length. It identifies

a testing subject and specifies the technical

characteristics of recorded EEG signal. The data part

contains consecutive fixed-duration epochs of the

record. Despite its drawback this data format has

been probably the most hopeful attempt to

standardize description of EEG data.

Vision Data Exchange Format (VDEF) is used

by the specific technical equipment in our

laboratory. EEG record is divided into three files: a

header file, a marker file and a data file. The header

file describes recorded data and provides a limited

set of corresponding metadata. The marker file

contains information about markers in EEG signal.

The data file contains raw EEG data.

KIV data format is a modification of simple

ASCII format of EEG signal, where metadata (file

header in ASCII) are stored in XML file and data

from electrodes are stored in separate binary files.

3.2 Artefacts Detection

Artefacts contaminate EEG signal; they include e.g.

blinks, skin potentials, and muscle movements.

Since artefacts may lead to misinterpretations during

EEG data analysis there is a high effort to find and

to eliminate them.

The basic methods for artefacts rejection use

a gradient criterion (the absolute difference value for

neighbouring EEG signal values is computed; values

exceeding a threshold are rejected) and an amplitude

criterion (EEG signal values exceeding a maximum

value, or not exceeding a minimum value are

rejected). Both methods are implemented as

standalone Java applications and also as aglets

(Section 4).

EEG signal containing artefacts is labelled using

a selected method. Labelled parts are then stored in

a special list for the next processing.

4 IMPLEMENTATION OF

AGLETS

All methods from Section 3 were implemented as

aglets.

4.1 Aglet for EDF- KIV Conversion

An aglet realizing the conversion between EDF data

format and KIV data format is inherited from

com.Ibm.aglet.Aglet; interface MobilityListener has

to be implemented. When a new aglet instance is

created on a server, the following event method is

ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence

156

invoked (method body implements the conversion

from EDF data format to KIV data format).

public void onArrival(MobilityEvent ev)

{

buffCreator = new BufferCreator

(new File("file.edf"));

KivFormatWriter kivWriter =

new KivFormatWriter();

kivWriter.write(buffCreator.getHeader(

), buffCreator.getBuffer(), (new

File()));

}

The other methods from Section 3 are

implemented in a similar way.

4.2 Java Mail

There is usually useful to inform a user about results

of performed operation. Therefore the possibility to

announce the results of aglet method by e-mail was

designed and implemented. Because system Aglets

does not support this kind of communication

naturally, JavaMail API as a client system was used.

A user has the possibility to define e.g. used mail

server, the message content, which is sent as the

result of aglet operation, and a list of message

recipients.

5 TESTING OF AGLETS SYSTEM

There is important if implemented aglets finish their

operations in time comparable to the running time of

methods implemented as common Java applications.

The testing of both implementations (common Java

classes and aglets) was performed on the computer

with EEG/ERP repository. Tahiti server was

installed on this computer and was run with

extended memory (256 MB).

Each test including one of the methods described

above was run ten times (finally the average running

time was computed). The results (average running

times) of each test performed subsequently on 7 MB

and 22 MB files in EFD format are available in the

following tables.

Table 1: Average running times (in seconds) for

conversion from EDF file to EDF file.

code type/file size 7 MB 22 MB

Java class 24 s 146 s

Aglet 27 s 148 s

Table 2: Average running times (in seconds) for

conversion from EDF file to KIV file.

code type/file size 7 MB 22 MB

Java class 76 s 249 s

Aglet 79 s 250 s

Table 3: Average running times (in seconds) for detection

of artefacts – gradient criterion.

code type/file size 7 MB 22 MB

Java class 3,8 s 10,0 s

Aglet 3,9 s 10,9 s

Table 4: Average running times (in seconds) for detection

of artefacts – amplitude criterion.

code type/file size 7 MB 22 MB

Java class 2,5 s 6,5 s

Aglet 2,5 s 6,7 s

The results show that running of Aglets system

does not slow down the computational process

considerably. The following tests of implemented

aglets using local university network had again

a marginal influence on the system performance.

6 CONCLUSIONS

The important aim of our research was to verify if

mobile agent system can help our research partners

to use our EEG/ERP repository in such a way that

transfers of huge data over network can be

eliminated. The open source system Aglets proved

its usability and it is now installed and prepared in

EEG/ERP laboratory for testing by our research

partners.

The possible disadvantage for the users is the

necessity to install not only Java Virtual Machine,

but also Tahiti server.

ACKNOWLEDGEMENTS

This work was supported by Grant Agency of the

Czech Republic under the grant GA 102/07/1191.

REFERENCES

Šolc, P., 2009. Decentralized processing of EEG signal

(Decentralizované zpracování signálu). Thesis (in

Czech). University of West Bohemia. Pilsen.

USING MOBILE AGENTS IN EEG SIGNAL PROCESSING

157

Lange, D., Oshima, R., 1998. Programming and

Deploying Java Mobile Agents with Aglets. Addison-

Wesley Professional.

Java mobile agent platform Aglets (Aglets), 2004.

Retrieved from http://aglets.sourceforge.net/

European Data Format (EDF). (n.d.). Retrieved from

http://www.edfplus.info/

Vision Data Exchange Format (VDEF). (n.d.). In Brain

Vision Recorder Manual. Retrieved from http://

biomag.uni-

muenster.de/EEGlab/docs/VisionRecorderManual.doc

Kučera, J., 2008. Software for ERP processing, data

reading and presentation (Software pro zpracování

ERP, načítání a zobrazení dat). Thesis. (in Czech).

University of West Bohemia. Pilsen.

ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence

158