Spatio-temporal transition of EEG activities during insight occurrence

The 21st Annual Conference of the Japanese Neural Network Society (December, 2011)
[P3-12]
Spatio-temporal transition of EEG activities during insight
occurrence
Ryo Kondo (PY)1 , Tomohiro Shibata1 , Naoya Oosugi2 , and Kazushi Ikeda1
1
Graduate School of Information Science, Nara Institute of Science and Technology
2
RIKEN-BSI / The University of Tokyo
E-mail: {r-kondo,tom,kazushi}@is.naist.jp
Abstract— Insight is a method for solving intellectual problems without any explicit algorithm. The
analyses in the literature identified the areas in the
brain that are related to insight but they have not
elucidated the temporal structure of insight due to
low time-precision. This study clarified the spatiotemporal transition of insight-related information in
EEG activities, where the amount of information was
quantified by the performance of a decoder predicting
insight occurence and a dimension-reduction method
is applied for extracting the spatio-temporal structure.
The results implies that the origin of insight is O2 in
international 10-20 system.
Keywords— Insight, EEG, Decoding
Introduction
Insight is characterized as a method for human to
solve intellectual problems suddenly without any explicit algorithm and it has been reported that activities in the anterior cingulate cortex and prefrontal
cortex increase when insight occurs [1, 2, 3]. This
study aims at clarifying the spatio-temporal transition
of Electroencephalography (EEG) activities during insight occurrence using decoding methods.
The performance of a decoder can measure the
amount of information contained in its input signals [4]. For example, if a decoder of EEG signals
at a specific channel in a specific time-window can discriminate an occurrence of insight from an algorithmic
method, the EEG signals can be said to have enough
information on insight. Hence, we consider the time
course of channels that show high prediction performance in single channel decoding.
Applying dimension-reduction methods to the performance matrix, we can visualize the spatio-temporal
transition of EEG activities related to insight. The
results implies that the origin of insight is O2 in international 10-20 system.
Figure 1: The flow of the anagram test in the experiment.
1
2 Material and Methods
2.1 Experimental Setup
Eleven healthy right-handed participants (age 19–
35, three males) participated in the study.
All
were graduate/undergrad students and had normal or
corrected-to-normal vision. Subjects were paid for
their participation. The experiment was conducted
with the approval of the Research Ethics Committee
in NAIST, and written informed consent was given by
all participants.
The participants were seated and faced to an LCD in
a shielded room. Their task was anagram test of English words with four letters on the display. The participants were instructed to solve anagram test, push a
button when they completed it, and orally report how
they solved. Figure 1 shows the flow of the task in one
trial. One block consists of twenty trials followed by
an interval for rest and ten blocks were conducted.
2.2 Data Acquisition and Preprocessing
EEG was acquired at 200 Hz sampling frequency
from fifteen scalp sites (Fp1, Fp2, F3, F4, C3, C4, P3,
P4, T3, T4, Fz, Cz, Pz, O1 and O2 in international
10-20 system) using electrodes. The EEG signals
were amplified and digitized using Polymate AP1132
(TEAC, Japan) and bandpass-filtered between 7 and
31 Hz (fourth-order Butterworth filter) using Matlab
(The Mathworks, USA).
2.3 Decoding and AUC Analysis
We constructed a decoder that classifies the EEG
signals to one of the two classes, insight or not. The input of a decoder is the set of single-channel single-timewindow band powers (alpha waves, 7–13 Hz;, beta1
waves, 14–20 Hz; beta2 waves, 21–30 Hz), where a
time-window is 750–500 msec, 700–450 msec, . . ., or
250-0 msec before the button-push in a trial.
The performance of a decoder was measured by evaluating the area under the specificity-sensitivity curve
(AUC), obtained by varying the detection threshold
with the five-folded cross validation. Note that the
AUC is a non-parametric statistics that estimates the
probability that a randomly chosen target has a higher
feature value than a randomly chosen non-target [5].
Hence, we can regard the score as a quantity of information on insight included in the channel and timewindow.
2.4 Spatio-Temporal Transition
We extracted the spatio-temporal transition of brain
activities that appear in common during insight process. To do so, we apply the singular-value decomposition (SVD) method to the AUC matrix, each row
of which consists of a participant’s class separation
(CS) scores, defined as CS = AUC − 0.5. Note that
the principal singular vector expresses the most typical
activation in insight.
3 Results
3.1 Behavioral Results
Two participants reported that they solved the
problem by insight at rate of only 1.2–1.4%, whose
data were removed since the rates were too small to
make classifiers. The others have 20–50% insight rate.
Note that we measured the participants’ response
time when they were asked to push a button, which
were between 430–510 msec.
3.2 AUC Values for Single Channels
Figure 2 shows the AUC values of single-channel
decoders for each participant. The values are significantly higher than the chance level, 0.5.
1
relationship to the anterior cingulate cortex or the prefrontal cortex [3] is not clear. Moreover, the effectiveness of each band discussed in [1, 2] is not clear either
since we constructed the classifiers with band powers
being input signals.
0.9
0.8
0.7
0.6
AUC
Figure 3: Spatio-temporal transition of activities.
0.5
5
0.4
0.3
0.2
0.1
0
Fp1 Fp2 F3 F4 C3 C4 P3 P4 T3 T4 Fz Cz Pz O1 O2
channels
Figure 2: The AUC values of single-channel decoders
for each participant.
3.3 Spatio-Temporal Transition
Figure 3 shows the principal spatio-temporal transition of activities, where topographical maps of CS
scores in three disjoint temporal intervals are depicted.
O2 was the are firstly activated as seen in the first
time-window (c). Then, the activation was widely
spread to centro-parietal areas in the second timewindow (b). Finally, the brain activities have little
relationship to insight during the response time (a).
4
Discussions
The spatio-temporal transition during insight occurrence agrees in part with the previous studies that insight is related to the activity in the right hemisphere
anterior superior temporal gyrus [1] or in the parietooccipital and centro-temporal areas [2]. However, the
Conclusions
This study clarified the spatio-temporal transition
of insight-related information in EEG activities by regarding the CS scores of insight as the amount of information on insight and by applying the SVD method for
extracting the common feature through participants.
The results implies that the activities of insight are
originated at O2 in international 10-20 system and
spread to centro-parietal areas during insight occurrence.
References
[1] Jung-Beeman, M. et al. (2004). Neural activity
when people solve verbal problems with insight,
PLoS Biology, 2, 0500–0510.
[2] Sheth, B. R., et al. (2009). Posterior beta and anterior gamma oscillations predict cognitive insight,
J. Cognitive Neuroscience, 21, 1269–1279.
[3] Dietrich, A., & Kanso, R. (2010). A review of EEG,
ERP, and neuroimaging studies of creativity and
insight, Psychological Bulletin, 136, 822–848.
[4] Oosugi, N., et al. (2010). A BCI study on the detection of insight occurrence from EEG, IEICE Technical Report, NC2010-59.
[5] Fawcett, T. (2006). An introduction to ROC analysis, Pattern Recognit. Lett., 27, 861–874.