Systems, Man & Cybernetics - January 2016 - 9

hemispheres were inversely changed.
Furthermore, the potential changes in T7
Discrete
Continuous
and T8 were almost linearly proportional
L
to the direction of the tongue. We also
F
plotted x d ^ t h, which is the difference
R
between the potential levels of T8 and T7
Fp1
in the bottom of the figure. It can be seen
70 µV
that x d ^ t h well follows the shape of the
Fp2
cue presented at the top of the figure, and
T7
it might be used as a feature to trace the
position of the tongue.
T8
To evaluate the response time of the
O1
potential change for tongue movements,
which is a critical factor to enable a realO2
time interface, we measured the time difπ
µL -
2
ference between a cue and an evoked
xd (t)
µF 0
potential response. We showed the sub(T8−T7)
µR π
jects a visual cue that changed from L to
2
F and F to R with recording a signal from
2 4 6 8 10 12
17
22
27
31
Time (s)
T7. We calculated the mean value n and
standard deviation v from each signal
segment recorded for 1 s before the cue Figure 2. the gKP response for horizontal tongue movements. the
given cues to a subject indicating the horizontal position of the
was changed. To specify the time when
tongue are plotted at the top of the figure. the cues pointing out one
the potential response was actually of three positions (L, F, and r) were presented for 12 s, then a new cue
observed, we detected the time point continuously traveling along a horizontal line was presented for 19 s.
when the potential level first exceeded the results showed that, when the tongue moved to the right, the
the range from n - 3v to n + 3v. Since potential levels in the left hemisphere (Fp1, t7) increased while the
levels in the right hemisphere (Fp2, t8) decreased. xd(t), which is the
p (n - 3v # x # n + 3v) . 0.9973 in a nordifference of the potential levels in t8 and t7, followed the shape of
ma l d i st r ibut ion, we a s su med t hat the presented cues.
exceeding this range would be caused
by an external factor, i.e., tongue movements. The average of the response times measured
tongue (L, F, and R) were used as the referential points
from 16 trials for two subjects was 577 ms. Since the
to calculate the angle.
mean reaction time to detect a visual cue is approxiTo implement the linear model, we segmented x d ^ t h
mately 190 ms for college-age individuals [10], it can be
from 3 to 4 s and calculated its mean value as nL. During
assumed that the GKP response can be observed in
this period, the tongue had touched the position L, and nL
EEG signals within 400 ms after a tongue movement.
represents the potential difference between T7 and T8 in
In this article, we only described the signals recorded
this condition. We discarded x d ^ t h from 2 to 3 s since it
near the facial region from a single subject. Please
was the unstable potential change that occurred during
refer to [5] for detailed information about the results
the transition state. The mean values of x d ^ t h from 5 to
from all EEG channels across the scalp, obtained from
6 s, and from 7 to 8 s were also measured in the same way
multiple subjects.
as n F and n R , respectively. To translate xt, which is the
potential difference between T8 and T7 on a time point t to
corresponding it, we exploited a linear model such that
TMI Detecting the Horizontal Position
^
h
We previously showed that the signal x d t changes
Z r xt - n F
according to the position of the tongue. Since x d ^ t h is
]- .
^ n L # x t 1 n F h,
] 2 nL - nF
almost linearly proportional to the position, a simple
(1)
it = [ r xt - n F
linear model could be utilized to translate x d ^ t h into
] 2 . n R - n F ^n F # x t # n R h,
]
the position. To represent the horizontal position of the
Error,
etc.
\
tongue, we exploited the angle between two directions:
1) the direction where the tip of the tongue is pointing
Our experimental results in [5], utilizing this linear model,
at and 2) the direction from posterior to anterior.
showed that this approach could detect the horizontal
Instead of the actual angle between them, we exploited
direction of the tongue in angular measurements. To meaan angular value i var ying from - r/2 , 0, to r/2,
sure detection accuracy, we prepared the angular cue
pointing out from one point on an upper semicircle. In
according to the tongue movements from L, F, to R, as
each trial, this cue denoted one random angle from -90° to
shown in Figure 1(a). That is, three positions of the
Ja nu a r y 2016

IEEE SyStEmS, man, & CybErnEtICS magazInE

Table of Contents for the Digital Edition of Systems, Man & Cybernetics - January 2016