15-01-2013, 02:56 PM
A Brain–Computer Interface Based on Mental Tasks
with a Zero False Activation Rate
A Brain–Computer Interface Based.pdf (Size: 370.47 KB / Downloads: 35)
Abstract—
Most brain–computer interface applications in real-life
suffer from the high rate of false activations. The ultimate goal
when designing brain–computer interfaces is to reach the zero
false activation rate while the true activation rate is kept at a high
level. In this study, a brain–computer interface design is shown to
have a zero false activation rate. The interface is based on
different mental tasks. It is custom designed to every subject and
to every mental task. The most discriminatory mental task for
each subject is determined. We use the autoregressive modeling
as the feature extraction method. The classification is performed
by a radial basis function neural network. The EEG signals of
four subjects during five mental tasks are used. The order of
autoregressive model is varied from 2 to 20 and custom designed
for each mental task and each subject in the cross-validation
stage. The performance of the brain–computer interfaces based
on the most discriminatory mental tasks is shown to be highly
promising since the false positive rate reaches zero while the
mean of the true positive rate obtained is above 70%.
Keywords-brain–computer interface; BCI; mental task; custom
design; EEG; false activation rate; autoregressive modeling; radial
basis function; neural network
I. INTRODUCTION
The Amyotrophic Lateral Sclerosis (ALS, sometimes called
Lou Gehrig's disease or Maladie de Charcot) and the Spinal
Cord Injury (SCI) are two common diseases that result in
motor disabilities. In the most severe cases, people with motor
disabilities cannot move any single limb of their bodies; hence,
these people are not able to communicate with their
environments. Therefore, an alternative way of communication
for these individuals is of great help. Brain–computer interfaces
(BCIs) form such an alternative tool.
The output of a BCI system has two operational states: the
intentional control (IC) state and the no control (NC) state. The
IC state is the state in which the BCI output is activated by the
user, while the NC state is the state in which the BCI output
remains inactive.
Two rates for evaluating the performance of BCIs are
defined. These are the true positive rate (TPR) and the false
positive rate (FPR). EEG signals are sampled at a certain rate.
Each sample is examined to determine as to which state IC or
NC it belongs to. TPR is the percentage of the correctly
classified IC states and FPR is the percentage of the
misclassified NC states.
A neurological phenomenon is a specific feature in brain
signals. Such a phenomenon is time-locked to activities in the
brain. Various neurological phenomena are exploited in
different BCI systems. The mental task is one of these
neurological phenomena. Refer to [1]-[3] for a review of the
field.
A BCI system based on mental tasks is proposed in this
study. This BCI system is custom designed for each subject and
for each mental task. The EEG signals of five mental tasks used
in our work have been collected by Keirn and Aunon [4]. This
dataset has been used in many studies, some of which appear in
[5]-[24]. The false positive rate is of great importance in BCI
applications; however, unfortunately, this rate has not received
enough attention. The confusion matrix or the false positive
rate has been only reported in very few studies [17]-[20]. In
this study, along with true positive rates, false positive rates are
also considered.
This paper is organized as follows. The methods used in the
design are proposed in section II. The dataset, the feature
selection method and the classification method are also
discussed in this section. Section III presents the results and
some discussion about their significance. The conclusions and
some suggestions for future research are given in section IV.
II. METHODS
A. Data
The dataset has been collected by Keirn and Aunon [4]. The
EEG signals of seven subjects, during five different mental
tasks have been recorded. These mental tasks are:
This work was supported in part by the Natural Sciences and Engineering
Research Council of Canada (NSERC) under Grant 90278-06 and in part by
Canadian Institutes of Health Research (CIHR) under Grant MOP-62711.
Proceedings of the 4th International FrD3.4
IEEE EMBS Conference on Neural Engineering
Antalya, Turkey, April 29 - May 2, 2009
978-1-4244-2073-5/09/$25.00 ©2009 IEEE 355
• The baseline: The subjects were asked to think of
nothing;
• The multiplication task: A nontrivial multiplication
problem (such as 43 times 85) was given to subjects to
solve mentally;
• The letter composing task: The subjects were asked to
mentally compose a letter to a friend;
• The rotation task: A 3D object were shown to the
subjects and they were asked to rotate it around an axis
mentally;
• The counting task: The subjects were asked to
visualize a sequence of numbers being written on a
blackboard.
Five trials of each mental task were performed in each
recording session. Each trial was 10 seconds long. The number
of completed trials is different among the subjects as shown in
Table I. Subject 5 completed three sessions. We used the data
of all subjects who completed at least 10 trials. This is with the
exception of subject 4 whose EEG signals contain some
missing data. Thus, the data of four subjects were used in this
study. The numbers assigned to the subjects in the original
study and in this study are presented in Table I.
The recording room was sound controlled with dim
lighting. Six electrodes were placed on central, parietal, and
occipital areas (C3, C4, P3, P4, O1, and O2 based on the
International 10-20 System). Electrically linked mastoids, A1
and A2 were the references.