Development of a Biosignals Framework for Usability Analysis
Inês Oliveira
Rui Lopes, Nuno M. Guimarães
CICANT
Universidade Lusófona
Campo Grande, 376
LASIGE, Faculty of Sciences,
University of Lisbon
Campo Grande
1749-016, Lisboa, Portugal
1749-024, Lisboa, Portugal
[email protected]
ABSTRACT
The understanding of human physical and physiological signals
and expressions, together with a growing processing and control
capacity allows for new approaches in interactive systems
design. In this poster, we introduce constitutive elements and
steps towards the integration of EEG (Electroencephalography)
signals in a system to analyze reading activities.
The poster presents system design choices that include the
software architecture and the feature extraction and
classification techniques used in the first prototypes.
Categories and Subject Descriptors
H.5.2 [Information Interfaces and Presentation]: User
Interfaces – user centered design, evaluation/methodology.
General Terms
Design, Experimentation, Human Factors.
Keywords
EEG signals, Usability Analysis
1
INTRODUCTION
Neurophysiological signals have become an important, and a
usable source of information. The recent research field of
neuroergonomics relates the disciplines of neurosciences and
ergonomics, and studies the behavior of the brain in the context
of the usage of real world artifacts and situations, i.e. the brain
at work [6]. The integration of this discipline with system design
has an impact on the methods for usability analysis [6].
The effectiveness of a usability study may benefit from the
consideration of the human physiological signals. [1][2]. The
paper describes the experience in the integration of electroencephalography (EEG) signals in a preliminary usability
analysis setting. The goal is to design experiments dealing with
the analysis of “reading” situations and this poster focuses on
the required system architecture and feature extraction solutions.
{rlopes|nmg}@di.fc.ul.pt
2
SYSTEM ARCHITECTURE
Our EEG-based applications are built upon a software
framework (EEGLib) that encapsulates the critical issues of
EEG processing.
The EEGLib components extract relevant features from EEG
data and can be reused in different applications. The object
model includes components for EEG data modeling and tools
for EEG data processing. The core concepts are: EEGFrame
represents a frame data, a numerical matrix with rows for
samples and columns for electrodes. Each frame is synchronized
with a time server. EEGStream models a time continuous stream
of frames, and EEGOperator abstracts the EEG processing
operations in frames and streams, and encapsulates feature
extraction and classification. Each concept is represented by a
(C++) class and integrated in a systematic hierarchy. All the
processing operations are based on a MatLab Engine [5].
EEGLib includes tools that can be classified in the following
categories: (a) Basic operations (Electrode, Sample and
Windowing); (b) Feature extraction; and (c) Classification
operations to categorize a feature vector. EEGLib currently
exploits the following recognition algorithms, based in MatLab
neural network and SPRTOOL [4] toolboxes: (1) Multi-layer
perceptrons, (2) Bayesian classifiers, (3) Linear classifiers, (4)
K-Nearest-Neighbors, (5) K-Means, (6) Support Vector
Machines, and (7) Expectation-Maximization.
The software framework provides the building blocks to
program EEG-based applications and to compose the right
(feature extraction * feature classification). The experiences use
a simple 16-channel EEG device, (MindSet-1000) represented
below in Figure 1 (10-20 International System).
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
SAC’09, March 8-12, 2009, Honolulu, Hawaii, U.S.A.
Copyright 2009 ACM 978-1-60558-166-8/09/03…$5.00.
Figure 1. 16-channel EEG device electrode map
3
EXPERIENCE AND RESULTS
4
Some very simple tools built with EEGLib were applied in test
experiments related with specific aspects of user interaction,
with a restricted number of subjects. The understanding and
observation of the “reading” task was the goal at this stage.
3.1
Reading or Not Reading
The initial experiments on “reading detection” were based in the
presentation of alternate blank and text screens containing a few
lines of news text. The text was never repeated was presented in
such a way that no particular visual effort was required.
Features vectors were based on the mean PSD (Power Spectral
Density) in α, β, θ and δ rhythms obtained in the 16-channel
signal. A full vector is therefore composed of 64 individual 16bit samples (1kHz). The feature vectors were reduced using
PCA or a minimal variance (Var) threshold. This threshold
eliminates features without changing the original signal. The
thresholds of both preprocessing methods were tuned for several
classification tools based on the average error rate in 100 trials.
In each trial, the training and test sets were randomly selected
from the reading and non-reading sets.
Figure 2 shows the best average results obtained. As the figure
shows, K-nn, AdaBoost with MV preprocessing and SVM with
PCA achieved precision and recall rates above 90%.
100.00%
DISCUSSION
This experience shows that EEG-based systems for usability are
feasible. A number of issues require further discussion.
The tests mentioned above were made with a very limited
number of subjects (2-3), so results cannot be generalized. The
focus has been the development and optimization of the
framework. Ideally, a significant number of tests and samples
should produce a robust identification of the channel*band
thresholds associated with a given cognitive process (i.e.
reading). We know however that variance is to be expected due
to task constraints, user differences or brain plasticity.
The effective exploitation use of biosignals such as EEG
requires the capacity for real time processing. The simplest
devices are usually isolated and generate sample files to be
processed off-line. In the experiments, real time processing
requires direct access to the device bus connections.
Our results provide a positive hint that EEG signals can be of
use in evaluating usability. In particular, the results of the
reading state detection open a door to the study of legibility via
EEG processing, provided that appropriate experiments are
designed. The next step will be to exploit these tools in the
analysis of known factors that influence legibility (such as font
size, font or background color). An important goal is to
objectively confirm some usability rules and heuristics.
A further development of the application of this framework is
the extension towards multimodality evaluation [8], an area of
advanced interfaces that requires innovative evaluation
techniques, and is closely coupled with the management of the
awareness and attention mechanisms.
90.00%
80.00%
70.00%
60.00%
5
50.00%
ACKNOWLEDGMENTS
This work was partially supported by Fundação para a Ciência e
Tecnologia (FCT) grants
SFRH/BD/29150/2006 and
SFRH/BD/30681/2006.
40.00%
30.00%
20.00%
6
10.00%
[1] Beer, R.D., et al, The Usability Lab of the Future,
INTERACT’03, 2003
Error Rate
False Positive Rate
A
BF
)-V
ar
(R
BF
)-P
C
(R
os
SV
M
False Negative Rate
SV
M
t-V
ar
A
aB
o
os
t-P
C
Ad
Ad
aB
o
A
nVa
r
Kn
nPC
Kn
ye
sBa
Ba
ye
s-
PC
A
Va
r
0.00%
Precision
Recall
Figure 2. READING : Best average results with PCA
and MV preprocessing
After selection of different bands and electrodes, theoretically
more correlated with the reading task, the results were refined
and re-obtained and showed better quality.
The selection of the best (band * channel) combination requires
further test and integration of previous results which provide
initial filtering options for the reading status information.
Reading tasks, as demonstrated in [3] activate different brain
areas (hemispheres for example), rhythms and also include a
spectrum of tasks like visual, orthographic, phonetic and
semantic).
REFERENCES
[2] Wolpaw, J.R., Brain Computer Interfaces for
Communication and Control, ASSETS’07, Oct 2007, AZ,
USA
[3] Bizas, E., et al, EEG Correlates of Cerebral Engagement in
Reading Tasks, Brain Topography, 12 (9), 1999
[4] Franc, V., Statistical Pattern Recognition Toolbox for
MatLab, http://cmp.felk.cvut.cz/cmp/software/stprtool/,
2008
[5] MathWorks. www.mathworks.com , 2007
[6] Nielsen,J., Usability Engineering, Morgan Kaufmann, 1994
[7] Parasuraman, R., Rizzo, M. Introduction to
Neuroergonomics. Neuroergonomics. The Brain at Work,
Oxford University Press Inc, 2008
[8] Carriço, L., Duarte, C., Lopes, R., Rodrigues M., and
Guimarães, N. Building Rich User Interfaces for Digital
Talking Books, CAD of User Interfaces IV, KAP, 2005