Towards real-time visualization of a juggler’s brain

Abstract

Surpassing the initial ‘wow’ effect of a complex juggling trick and producing long-lasting engaging performances are the main goals of any juggling act. Conveying to the audience the skill and the effort required for a performance is often difficult. In this paper, we use a wearable EEG headset to investigate how juggling skills can be inferred from a juggler’s brain. We observed characteristic brain activity and synchronization while juggling in both an expert and an intermediate juggler. We also found that processing of visuomotor skills and memory retention can be distinguished during motor imagery and simulated juggling conditions. For the first time, we were able to monitor a juggler’s brain in action. We have shown that using EEG while juggling could both improve our understanding of neuronal mechanisms governing visuomotor control and, importantly, represent a potential to enrich artistic performance and increase audience engagement.

Full text

Towards real-time visualization of a juggler’s brain
Giuseppina Schiavone
a
*, Ulf Großekathöfer
a
, Simon à Campo
b
and Vojkan Mihajlović
a
a
Wearable Health Solutions, Holst Centre / imec The Netherlands, High Tech Campus 31, Eindhoven, The Netherlands;
b
Juggle
Company: Simon à Campo, Eindhoven, The Netherlands
(Received 13 April 2015; accepted 27 September 2015)
Surpassing the initial ‘wow’effect of a complex juggling trick and producing long-lasting engaging performances are
the main goals of any juggling act. Conveying to the audience the skill and the effort required for a performance is often
difficult. In this paper, we use a wearable EEG headset to investigate how juggling skills can be inferred from a juggler’s
brain. We observed characteristic brain activity and synchronization while juggling in both an expert and an intermediate
juggler. We also found that processing of visuomotor skills and memory retention can be distinguished during motor
imagery and simulated juggling conditions. For the first time, we were able to monitor a juggler’s brain in action. We
have shown that using EEG while juggling could both improve our understanding of neuronal mechanisms governing
visuomotor control and, importantly, represent a potential to enrich artistic performance and increase audience
engagement.
Keywords: juggling; motor imagery; visuomotor coordination; memory retention; gamma/theta coupling
1. Introduction
Juggling requires coordination and synchronization of
repetitive hand and body gestures to produce periodic
throwing and catching of a number of objects (e.g.,
balls). Professional juggling involves complex technical
and aesthetic visuomotor skills that are acquired through
rigorous practice and exercise over the course of months
and years. As such it can be considered a highly relevant
tool to investigate neuroplasticity associated with motor-
learning and spatiotemporal dynamics of task-dependent
perceptual-motor coordination.
Although juggling with more than five balls and
introducing juggling tricks typically produces admiration
and appreciation among observers of a juggler perform-
ing, the complexity of the performance and the skills
required often remain unrecognized by a naive audience.
To capture these hidden aspects of juggling, jugglers
have introduced auxiliary means of conveying the art of
performance in recent years. These means are typically
presented as artistic concepts based on synergic interac-
tion between body movements and objects in the 3D
space. While these interpretations of a juggler’s
performance can illustrate the complexity of aspects of
juggling, the skill required for the routines still remains
hidden. To tap into this hidden legacy of a juggler, one
must investigate his brain activity.
Here, we show a proof of concept of how a wearable
EEG system can be used for inquiring into the neuronal
mechanisms that underlie visuomotor processing during
juggling, and we discuss how brain-computer interfaces
(BCI) can offer alternative opportunities for training and
for enhancement of juggling performances. Two subjects
participated in our study, one intermediate and one
expert juggler. Two experimental conditions were
designed. The first one was intended to characterize
brain activity and connectivity during three-ball cascade
juggling compared to conditions such as rest, imagery
juggling, and juggling movements without balls in both
intermediate and expert juggler. Due to the limited num-
ber of subjects, we could not run a statistical analysis for
comparing the two categories intermediate and expert; at
the same time we were expecting to find differences
between the two jugglers due to network adaptation and
functional specialization induced by several years of jug-
gling practice. The second experimental condition was
intended to investigate whether the difficulty of a jug-
gling trick was reflected in the EEG of the expert juggler
while performing a juggling cascade with three, five, and
seven balls. This latter experiment was related to perfor-
mance execution and was formulated together with a
professional juggler to solve an apparently common
problem during performance on stage: ‘both during com-
petitions and standard shows, the audience is not always
able to understand how difficult it is for a juggler to per-
form certain trick, especially those that appear relatively
simple in ball patterns, but require high attention and
visuomotor control’–quoted from a brainstorm meeting.
The possibility of visualizing brain activity of a juggler
*Corresponding author. Email: giuseppina.schiavone@imec-nl.nl
© 2015 Taylor & Francis
Brain-Computer Interfaces, 2015
http://dx.doi.org/10.1080/2326263X.2015.1101656

in a way that reflects the complexity of the executed
trick could help to increase the engagement and tuning
of the audience during a show.
The paper is organized as follows. In the section
‘State-of-the-art’we report the most relevant works
related to juggling from a neuroscience perspective. In
the ‘Methods and materials’section we present the
experimental set-up and protocols, the techniques
adopted for processing the EEG signals and the statistical
analysis performed on the collected data. Subsequently,
the ‘Results’section illustrates the main outcomes of our
research, followed by the ‘Discussion’section, in which
we contextualize our results with reference to our
research questions and related works. Finally, in the
‘Conclusion’section we outline follow-up studies and
propose interesting directions for the use of wearable
EEG during juggling.
2. State-of-the-art: juggling-induced neuroplasticity
The effect of juggling training on cortical organization
and brain functioning has been addressed using
neuroimaging techniques, other than EEG, before and
after training. Scholz et al. [1] used diffusion tensor
imaging (DTI) to measure the variation of fractional
anisotropy, correlates of white-matter microstructure vari-
ation, and voxel-based morphometry (VBM) to measure
gray-matter changes in response to the learning of a
complex visuomotor skill in juggling. They reported sig-
nificant training-related increases in fractional anisotropy
in white-matter underlying the right posterior intrapari-
etal sulcus and significant increases of gray-matter den-
sity in the medial occipital and parietal lobe in cortical
regions that overlie the white matter area. They also
showed that, in general, structural changes did not corre-
late significantly with training progress or the perfor-
mance level,[2,3] suggesting that the majority of
structural changes might be related to the amount of time
spent training and learning a new task and not to the
training outcome.
Similar conclusions were drawn earlier by Driemeyer
et al. [2] where VBM was applied to investigate activity-
dependent gray-matter changes using the same experi-
mental paradigm as in Scholz et al. [1] i.e., three-ball
cascade juggling. Driemeyer et al. [2] showed that learn-
ing to juggle can induce gray-matter changes in the
occipito-temporal cortex as early as after 7 days of train-
ing. These changes were referred to as transient because
further alterations of brain structure were not observed in
conjunction with improvement of the juggling skills over
time due to training.
Additionally, while comparing expert and non-expert
jugglers, Gerber et al. [4] found a significant increase of
gray-matter density in regions involved in visual motion
perception and eye-hand coordination in expert jugglers.
For this group gray-matter density in right visual areas
(hMT+/V5) was found to correlate with juggling
performance.
Attentive screening of literature research on this topic
failed to find studies where EEG activity was recorded
during execution of juggling performances. The most
trivial reason preventing studies is the delicate nature of
EEG systems and the requirement that the artists have to
be able to freely move while juggling. Cumbersome
EEG systems that require the usage of conductive gel
and wires that connect electrodes to the EEG acquisition
system, which are highly sensitive to motion-induced
noise, are among the main obstacles. Furthermore, exper-
iments confined to a limited working space are unsuit-
able for containing juggling posture and movements.
In this work, we explore the advantages of using a
wearable EEG headset to visualize the electrical brain
activity of a juggler, showing how wearable, wireless
EEG systems with dry electrodes offer a viable alterna-
tive to traditionally used gel-based wired EEG systems
[5] for out-of-the-lab BCI research and entertainment.
3. Materials and methods
3.1. Participants
Two subjects participated in our study, an intermediate
(left-handed male, 40 years old) and an expert juggler
(right-handed male, 22 years old). The intermediate ama-
teur juggler was recruited among colleagues, while the
expert juggler was a professional juggler with more than
10 years juggling experience. We consider the difference
between intermediate and expert juggler in line with pre-
vious definitions: experts are defined as those who could
juggle five or more balls; intermediate jugglers are
defined as those who could comfortably maintain a
three-ball juggle for more than a minute.[6]
3.2. EEG head set
The experiments were performed using the wireless imec
EEG headset, shown in Figure 1A. It has the capability
of continuously measuring EEG and electrode-tissue
impedance signals at up to 1024 Hz.[7,8] In this study a
sampling rate of 256 Hz was used. To ensure the correct
positioning of the headset and its signal quality, the
impedance was checked at the beginning of each trial
and whenever a noisy EEG signal was observed. In addi-
tion, impedance was used to annotate noisy segments,
considering that acceptable impedance values for dry
electrodes should be kept below 100 kΩ, i.e., about 10
times higher than impedance values for gel electrodes.
Commercially available dry Ag/AgCl electrodes with
pins were used to penetrate the hair and contact the
scalp. The electrodes were mounted on a spring-loaded
support to ensure good contact with the skin and more
2G. Schiavone et al.

comfort to the user (see Figure 1A). Active electrode
chips that buffer the electrical signal are placed directly
on top of the spring-loaded contact to prevent noise from
entering the EEG system as much as possible. The head-
set measures the potential difference between measure-
ment electrodes at locations C3, C4, Cz, and Pz of the
International 10–20 System for EEG measurements and
the reference electrode positioned at the right mastoid.
The subject bias electrode or ground electrode is located
behind the left mastoid.
3.3. Procedure
Two experimental protocols were performed. The first
protocol involved both intermediate and expert jugglers
and consisted of five conditions:
•Rest: rest and think of something not related to
juggling
•Imagery: imagine juggling
•Juggle: perform three-ball cascade pattern
•ImageryHands: move arms without balls in a
juggle-like fashion and imagine juggling
•NoBalls: move arms in a juggle-like fashion and
think of something not related to juggling
An illustration of juggling performance (Juggle) of
an expert juggler is depicted in Figure 1B. Each condi-
tion lasted 20 seconds and was repeated 15 times (trials)
for the intermediate juggler and 10 times (trials) for the
expert juggler. The number of trials was less in the
expert juggler due to time availability. In each trial the
sequence of conditions was randomized. For all condi-
tions the subjects had their eyes open. They were asked
to keep their head as still as possible and to limit overall
upper body movement in order to reduce the impact of
motion on the EEG signal. During the Juggle condition,
whenever a ball fell down before the established execu-
tion time was reached the trial was discarded and the
condition was repeated. There were pauses of a few
minutes between trials whenever requested by the sub-
jects. The second protocol involved only the expert jug-
gler and consisted of three conditions with incremental
difficulty:
•3 Balls: perform three-ball cascade pattern
•5 Balls: perform five-ball cascade pattern
•7 Balls: perform seven-ball cascade pattern
Each condition was repeated three times (trials) in a
randomized order at each trial. Because of the difficulty
of sustaining the seven-ball game for longer periods, a
duration threshold of 15 seconds was defined for each
trial. If a ball fell down within this threshold the trial
was discarded and repeated; if the juggler was able to
sustain the game for a longer period the recording was
continued until the first ball dropped. The cascade pat-
tern [9] was chosen because of the consistency of the
ball pattern across task difficulties induced by increasing
the number of props (horizontal figure-eight above the
hands, produced by throwing one prop in an arc-like
fashion before catching another on its way down).
Figure 1. Experimental setup: (A) imec’s wireless EEG headset with dry electrodes; (B) snapshot of expert juggler wearing the EEG
while juggling in a five-ball cascade pattern.
Brain-Computer Interfaces 3

3.4. Processing
3.4.1. Filtering and artifact removal
EEG signals collected in both experiments were
processed in similar ways. To ensure the integrity of the
data we initially visually inspected all signals and manu-
ally removed segments containing spike-like artifacts that
are representative of non-physiological signal distur-
bances. Segments with impedance higher than 40 kΩ
were also considered affected by artifacts and removed.
We then band-pass-filtered the signal with a third-order
Butterworth filter in a 3–70 Hz frequency band, and
applied a 50 Hz notch filter. Considering that juggling
movement can reach up to 2–3 Hz for intermediate and
advanced jugglers,[10] we chose a low-pass cutoff fre-
quency of 3 Hz to remove EOG and other movement
artifacts.
Our empirical evaluation showed that in all cases in
this study (except for juggling with seven balls), motion
artifacts mostly impacted EEG content below 4 Hz.
Given that the impact of automatic motion artifact reduc-
tion on the EEG content is unknown,[11] we decided to
not use any motion artifact reduction methods, but
instead exclude the delta band from our analysis. It is
worth mentioning that, according to observations from a
previous study [11] in which the impact of head move-
ments on EEG spectral content was investigated, we
expected that motion artifacts could cause increases in
the EEG power spectrum for a frequency range lower
than 20 Hz.
Artifact removal processing resulted in a total of
about 31 minutes of recording for each EEG channel
((15 for intermediate +10 for expert) trials * 15 seconds
(clean signal) * 5 conditions) for the first experiment
and a total of about 1.5 minutes for each EEG channel
(3 trials * 10 seconds (clean signal) * 3 conditions) for
the second experiment. For the latter, signals in Pz were
excluded from further analysis (power, coherence, and
statistical analysis) due to the high electrode-tissue impe-
dance signal, caused by skin-electrode contact loss.
3.4.2. Spectral analysis
Cleaned and filtered signals were segmented into
epochs of 4 seconds in length with 75% overlap and
Welch spectral analysis [12] was applied to each epoch.
After visual inspection of peaks in the band power
spectra profiles, averaged across trials (Figure 2), five
frequency bands were considered for analysis of the
power spectrum: theta (3.5–8 Hz), alpha (8–13 Hz),
beta (13–29 Hz), low gamma (29–35 Hz), and high
gamma (35–45 Hz) (Figure 3). The average of the log
power spectra across each frequency band was com-
puted for each electrode and used for further statistical
analysis.
3.4.3. Coherence analysis
Spectral coherence between pairs of EEG channels was
considered to measure synchrony of oscillations between
electrodes in each frequency band, as defined for the
spectral analysis. The mscohere MATLAB function was
used. It computes the magnitude squared coherence esti-
mate of the two EEG signals using Welch’s averaged
modified periodogram method and measures linear syn-
chronization between two series. A value of 1 indicates a
perfect linear relationship, meaning perfect agreement in
phase difference, while a value of 0 denotes that the ser-
ies are uncorrelated, meaning completely random phase
differences.
3.5. Statistical analysis
Repeated-measure ANOVAs were used for the statistical
analysis of the first experiment. For both the power and
coherence modulations within-subject, three-factors
repeated-measure ANOVA was performed with the
factors being: electrode site (C3/C4/Cz/Pz) for power
spectra and electrodes pairs (C3-C4/C3-Cz/C3-Pz/C4-Cz/
C4-Pz/Cz-Pz) for coherence measure, task condition
(Rest/Imagery/Juggle/ImageryHands/NoBalls), and
Figure 2. Examples of 4-s epochs of the EEG signal after cleaning and pre-processing. Oscillations in time are illustrated for each
electrode and each condition of the first experimental protocol.
4G. Schiavone et al.

Figure 3. Broad-band power spectra of the expert (1) and intermediate (2) juggler during the first experimental protocol.
Brain-Computer Interfaces 5

frequency (theta/alpha/beta/low gamma/high gamma),
with subjects as random factor (1 = expert juggler, 2 =
intermediate juggler). Furthermore, two-factor ANOVAs
(with factor frequency and task condition) for each sub-
ject were used in consequent analysis. Tukey’s test was
performed for post-hoc analysis where differences
between Juggle and other conditions, Rest and Imagery
conditions, and ImageryHands and NoBalls conditions
were tested.
For the second experiment, grand averages across
trial repetitions for frequency bands and different condi-
tions were reported without performing further statistical
analysis, due to the low number of trials per condition.
All statistical analysis was performed in R. The sta-
tistical analysis performed for this work should be con-
sidered as indicative not assertive given that only two
subjects participated in the study.
4. Results
4.1. Results of Experiment 1
4.1.1. Power spectra modulation
Figure 4shows the modulation in power across task con-
dition and frequency. The three-way within-subject
repeated-measure ANOVA showed significant main
effects of frequency (F(4,4) = 33.58, p= .002), condition
(F(4,4) = 6.572, p= .047), and channel (F(3,3) = 12.76,
p= .032), as well as an interaction between frequency
and condition (F(16,16) = 4.84, p= .0015). We
proceeded by investigating Tukey’s test two-factor
ANOVAs, with factor task condition (Rest/Imagery/
Juggle/ImageryHands/NoBalls) and frequency (theta/
alpha/beta/low gamma/high gamma), for each subject,
and considered results of the comparison between Juggle
condition and other conditions, Rest and Imagery
conditions, and ImageryHands and NoBalls conditions.
Qualitative assessment on effects at each electrode sites
is also reported (Figure 5depicts observation at the
electrodes level for the intermediate subject).
While comparing Juggle vs other conditions, we
found that, for the expert juggler, the power in the Juggle
condition is significantly higher (p< .005) than in other
conditions in all frequency bands. For the intermediate
juggler, the power in the Juggle condition is significantly
higher (p< .001) than in other conditions in high gamma
and theta (across all electrodes); it is significantly higher
(p< .001) than in Rest and Imagery conditions in low
gamma (in C3, C4, Cz); it significantly differs
(p< .005) in alpha (being higher in C3, C4, Cz) from all
other conditions apart from the NoBalls condition; no sig-
nificant differences are observed for the beta band.
In the Rest vs Imagery conditions comparison, we
found overall significantly (p< .001) higher power in
the alpha band for Rest condition compared to Imagery
condition for the expert juggler; no significant differ-
ences between these two conditions were found for the
intermediate juggler.
In the ImageryHands vs NoBalls conditions compar-
ison, we found overall significantly higher (p< .05)
alpha power in the NoBalls condition compared to the
ImageryHands condition for the intermediate juggler; no
significant differences were found for the expert juggler.
4.1.2. Coherence modulation
Figure 6shows the modulation in coherence across task
condition and frequency. The three-way repeated-
measure ANOVA showed significant main effects of
condition (F(4,4) = 19.27, p= .007) and paired channel
(F(5,5) = 36.31, p= .0006), and an interaction frequency-
condition (F(16,16) = 2.685, p= .028). We proceeded by
investigating Tukey’s test two-factor ANOVAs, with factor
task condition (Rest/Imagery/Juggle/ImageryHands/
NoBalls) and frequency (theta/alpha/beta/low gamma/high
Figure 4. Log of power spectral density for expert (1) and intermediate (2) juggler, averaged across electrodes and trials, for
different task conditions and frequency bands in the first experimental protocol.
6G. Schiavone et al.

Figure 6. Coherence measure for expert (1) and intermediate (2) juggler, averaged across electrodes and trials, for different task
conditions and frequency bands in the first experimental protocol.
Figure 5. Log of power spectral density for intermediate juggler, averaged across trials, for different electrode locations, task
conditions, and frequency bands in the first experimental protocol.
Brain-Computer Interfaces 7

gamma), for each subject, and considered results of the
comparison between Juggle condition and other condi-
tions, Rest and Imagery conditions, ImageryHands and
NoBalls. Qualitative assessment on effects at each elec-
trode site is also reported (Figure 7depicts observations at
the electrode level for the expert subject).
In the Juggle vs other conditions comparison, for the
expert juggler, the theta coherence during Juggle is sig-
nificantly different (p< .001) from theta coherence in
the ImageryHands condition (being higher in Juggle in
CzPz, C4Cz, C4Pz; lower in Juggle in CzC3); the alpha
coherence in Juggle significantly differs (p< .05) from
other conditions (lower in CzC3 compared to NoBalls,
higher in C4Cz, C4C3, C4Pz compared to other); the
beta coherence in Juggle significantly differs from the
beta coherence in the Imagery condition ( p< .05) and in
Rest condition (p< .001) ( being higher for all electrode
pairs apart from CzPz); the same was found for high
gamma coherence (p< .001) (here coherence while jug-
gling is higher for all pairs). For the intermediate juggler,
gamma-band coherence during Juggle is overall
significantly higher (p< .001) than in other conditions;
for beta and theta coherence Juggle significantly differs
(p< .001) from the Rest condition (being higher in beta
and lower in theta).
In the Rest vs Imagery conditions comparison, no
significant differences in coherence across frequencies
were found for both expert and intermediate juggler.
In the ImageryHands vs NoBalls conditions compar-
ison, significant differences (p< .001) were found only
for the expert juggler in alpha coherence and in theta
coherence (with higher coherence in NoBalls compared
to ImageryHands).
4.2. Results of Experiment 2
Grand averages of power spectra across trial repetitions
(Figure 8) show that increasing difficulty due to
increased number of balls for the cascade pattern is
reflected in an increase of power across all frequency
bands and channels. Grand averages of coherence across
trial repetitions (Figures 9and 10) show that for the 7
Balls compared to the 3 and 5 Balls cascade, coherence
between central electrodes increases in theta and alpha
bands and decreases in beta and gamma bands. Interest-
ingly, for the 5 Balls cascade coherence increases across
Figure 7. Coherence measure for expert juggler, averaged across trials, for different electrode locations, task conditions, and
frequency bands in the first experimental protocol.
8G. Schiavone et al.

all frequency bands, apart from beta band, and for all
channel pairs compared to the 3 Balls condition, while
in the beta band coherence between Cz-C4 and Cz-C3 is
attenuated compared to the 3 Balls condition.
5. Discussion
5.1. Experiment 1: the expert juggler brain
5.1.1. Juggle vs other conditions
Our results show that the experienced juggler exhibits
specific brain activation while juggling compared to
other conditions. In particular, the power of oscillations
across the scalp was higher while juggling for all consid-
ered frequency bands, even when juggling movement
was mimicked and isolated in the NoBalls and Imagery-
Hands conditions, demonstrating that this effect could
only partially be explained by the presence of artifacts
for frequencies below 20 Hz.[11] Higher theta coherence
between electrodes in the right hemisphere compared to
mimic movements in the ImageryHands condition could
be attributable to the involvement of theta band in reach-
ing movements [13] and to the degree of motor learning
and retention.[14–16] Higher interhemispheric gamma
and beta coherence in the Juggle condition compared to
the Imagery and Rest conditions could represent
synchronous oscillations associated with intrinsic ( body-
related) and extrinsic (object-centered) coordinate trans-
formations during right-to-left and left-to-right movement
[17] in both the presence (Juggling) and absence of goal-
directed motor planning (ImageryHands and NoBalls).
Overall higher alpha coherence compared to other condi-
tions can be associated with hemispheric synchronization
in the control and coordination of bimanual movement.
In particular, here we notice dominance in synchronous
activity in the right hemisphere, possibly reflecting the
Figure 10. Coherence measure for expert juggler, averaged across trials, for different electrode locations, task conditions, and
frequency bands in the second experimental protocol.
Figure 8. Boxplot of log power spectral density averaged
across electrodes and trials for each frequency band and
condition in the second experimental protocol.
Figure 9. Boxplot of coherence measure averaged across
electrodes and trials for each frequency band and condition in
the second experimental protocol.
Brain-Computer Interfaces 9

established efficient bimanual motor routine or internal
models [18] and a stronger visuomotor adaptation [19]
due to extensive practice. Interpretations on lateralization
in this paper should be carefully considered given the
reduced number of electrodes and the absence of source
modeling.
In the intermediate juggler, the Juggle condition was
specifically characterized by higher power in theta and
high gamma frequency bands and higher interhemi-
spheric gamma (low and high) coherence. Interestingly,
when comparing Juggle and Rest conditions, as seen in
the expert juggler, higher beta coherence was found in
Juggle compared to Rest, while, in contrast to the expert
juggler, lower theta, instead of higher gamma coherence,
was found. It is possible that for both experienced and
intermediate jugglers, synchronous activations of beta
oscillations are associated with intrinsic coordination
coding, while interplay of synchronous gamma and theta
oscillation and modulation of gamma and theta spectral
amplitude are attributable to extrinsic visuomotor coordi-
nation and movement planning, allowing a switch from
already acquired internal models to formation of new
motor memories, more for the intermediate than for the
expert juggler.[13] The complexity and the variety of
synchronous neuronal activities observed in the expert
juggler could represent the results of a learning process
characterized by temporal hierarchy and multiform
dynamics. As reported by Huys et al.[20] and Mapelli
et al.,[10] the level of expertise of a juggler and his per-
formance are shaped by the successive acquisition of dif-
ferent visuomotor skills from control of postural sway to
eye-head and hand movement coordination,[21,22]ata
monotonically increasing frequency-locked ratio to the
balls trajectories. This allows the proficient juggler to
switch adaptively between functional organizations
involving distinct perceptual systems.[20,23]
5.1.2. Rest vs Imagery conditions
Rest and Imagery conditions were included in the first
experimental protocol to control for neuronal activity
involved in the mental representation of the juggling
movement. As previously reported, our results show that
for both the expert and the intermediate juggler, neuronal
activity distinguishing Imagery and Rest conditions from
actual juggling movement resulted in higher beta coher-
ence during movement. This result might also be attribu-
ted to differences in functional interactions of different
areas across the scalp during visuomotor processing [24]
of real and imagined movement. Comparing the act of
imagining the movement with the rest condition we
could not find differences in synchronous activity across
frequency bands for both jugglers. On the other hand,
significantly higher power in alpha oscillation during
Rest compared to the Imagery condition was found only
in the experienced juggler. Alpha suppression during
mental rehearsal of movement is associated with the abil-
ity to generate motor imagery, and is commonly used in
BCI applications.[25] It is possible that the intermediate
juggler had more difficulty in generating motor imagery
of juggling while the expert juggler had a stronger and
stable internal representation of motor movement
induced by extensive practice.[26–29]
Interestingly, higher alpha power during the Juggling
condition compared to Rest and Imagery seems in dis-
agreement with the established role of alpha suppression
in motor control. This finding could be partially a result
of unaccounted movement artifacts below 20 Hz, and
partially attributed to alpha synchronization induced by
top-down modulation as seen in tasks involving attention
towards moving objects, decoding of visual space, or
access to long-term memories.[30]
5.1.3. ImageryHands vs NoBalls conditions
ImageryHands and NoBalls conditions were introduced
in the first experimental protocol to control for neuronal
activity involved in movement control and right-left hand
coordination, isolating the visual component of juggling
as a goal-directed task. As previously reported, our
results show that, for the expert juggler, juggling is
clearly distinguishable from other juggling-like move-
ments by looking at the increase of the power spectra
across different frequency bands. For the intermediate
juggler no statistical difference between the power of
alpha oscillation during Juggling and NoBalls condition
was found, and higher alpha suppression in the Imagery-
Hands condition compared to Juggle and NoBalls condi-
tions was observed. This result may derive from the
ability of the intermediate juggler to produce motor ima-
gery during the motor action, even better than during
isolated mental rehearsal. For the expert juggler higher
alpha and theta coherence in the NoBalls condition com-
pared to ImageryHands was found. Also alpha coherence
while juggling was found to be significantly lower com-
pared to NoBalls for the expert juggler. This reduced
synchronization of theta and alpha oscillation during
motor imagery combined with motor movement and
actual juggling in the expert juggler could be interpreted
as the ability to quickly represent motor imagery as
retrieved from strong memory retention of the juggling
movement.
5.2. Experiment 2: visualising cascade juggling
difficulty
The second experiment was designed to determine
whether the increasing difficulty in performing cascade
juggling with three, five, and seven balls is reflected in
the EEG of the experienced juggler. We found that the
10 G. Schiavone et al.

power of neuronal oscillations across all frequency bands
generally increases with the task difficulty. Also,
synchronous activity was found to increase across all fre-
quencies when passing from the 3 Balls to the 5 Balls
condition. Synchronization of activities for the 7 Balls
was higher in alpha and theta bands and lower in beta
and gamma bands compared to the 3 and 5 Balls condi-
tions. Broadband increase of power and coherence might
be the product of unaccounted artifactual components,
due to increased body movements and instability of the
juggler tossing seven balls. It is also possible that mecha-
nisms related to higher attentional demands and motor
control contribute to increased synchronous brain activity.
5.3. Limitations
The study described in this paper investigated EEG mon-
itoring in an entirely different setting than traditional
controlled lab environments with a clearly specified user
behavior protocol. In our case, the user is allowed to
freely move and perform juggling activities, while his/
her EEG is recorded with a wireless dry-electrode EEG
headset. Even though such a setup minimizes the impact
on natural user behavior and performance, it also gives
rise to a number of limitations that stem from fragile
captured EEG signals. The recorded signal is prone to
various interferences that are difficult to characterize and
isolate from the EEG. We believe that body, eye, and
face muscle movement artifacts have the most detrimen-
tal effect on the signal.[5] Given that those artifacts
mainly impact low-frequency components of the EEG,
we excluded the signal in the delta band from the analy-
sis. However, this does not ensure that all artifact com-
ponents are removed from the signal. For example, a
large increase in the theta and alpha bands with the
increased complexity of juggling performance or increase
in the theta band due to performing tasks involving hand
movements can partially be attributed to the amount of
movement required. Also, facial muscle and neck muscle
tension can result in beta and gamma EEG artifacts.
Further studies in this domain should address proper
characterization of artifacts and should incorporate arti-
fact-handling techniques during signal analysis, e.g., by
using electrode-tissue contact impedance.[11]
The quantitative EEG analysis performed in this
paper represents a first standard attempt to uncover pos-
sible biomarkers of juggling skills. Further analysis
should also include relative power, which is better suited
to assess between-frequency effects and allows lower
inter-subject variability. Phase coherence should also be
considered in further processing, having the advantage of
measuring phase synchronization independently from
amplitude correlations and being a better representative
of directional flow of information and functional interac-
tions between different brain areas.
Including only two participants in the study, one
expert and one intermediate, also limits the interpretation
power of the results reported. Although with a larger
number of repetitions we were able to extract clear pat-
terns across activities in different EEG frequency bands
for both participants, it is yet to be determined to what
degree the same effects can be observed in other jug-
glers. This concerns three aspects: jugglers with different
skill levels, different juggling techniques used, and dif-
ference in jugglers’handedness. Furthermore, we
exploited only a short segment of 20 s of juggling
performance, which represents only snapshots of EEG
activity during performances. Continuous monitoring
over longer periods, before, during, and after practice or
performance, could provide many more insights into a
juggler’s brain.
5.4. Implications for juggling performance
The preliminary results obtained in our analysis open up
new opportunities for a juggler to increase the engage-
ment of the audience during a performance. The identi-
fied EEG biomarkers of juggling skills and game
difficulty could be visualized in real time to allow the
audience to have a better perception and understanding
of the mental efforts required for juggling. Other possi-
bilities of performance enhancement could include map-
ping of these biomarkers into sound or other visual
outputs, as explored in other studies by means of move-
ment mapping.[31–33] Expanding this work towards a
BCI system which includes neuro-feedback for the user
could represent a new and powerful tool for training
beginner jugglers.
6. Conclusion
In this work we have shown that monitoring the electri-
cal activity of the brain while juggling using dry elec-
trodes mounted into a wearable and wireless EEG
headset is possible and can offer new solutions for out-
of-the-lab BCI research and entertainment. Despite the
fact that additional work is required to isolate neurophys-
iological signals from movement artifacts, we have
shown for the first time the potential of using wearable
EEG for studying neuro-plasticity induced by the jug-
gling practice.
Different brain activity was observed in the two
jugglers in particular in alpha and theta power and coher-
ence while performing imagined juggling movements
compared to performing the simulated movement only,
possibly attributable to different memory retrieval pro-
cesses and motor coordination. Higher beta coherence
while performing the movement was also observed in
both participants, suggesting functional interactions
between different brain areas.
Brain-Computer Interfaces 11

Interpretations of the reported results are still
speculative, given the limited number of jugglers
involved and the simplistic methods used. More sus-
tained considerations can be achieved by involving a lar-
ger number of jugglers with different skill levels and/or
by comparing different juggling techniques. The use of
more sophisticated quantitative methods for EEG analy-
sis, such as phase coherence and detrended fluctuation
analysis, should also be considered in future studies to
better characterize the juggler’s brain’s temporal dynam-
ics and functional connectivity.
We hope that these preliminary results will stimulate
further research towards the understanding of cognitive
processes of juggling beyond body movement analysis.
Analyzing attention and mental engagement, measuring
stress levels, or estimating emotional states before, dur-
ing, and after performances and training sessions might
allow exciting new insights into the mind of a juggler.
Similarly, investigating the brain activity during juggling
in combination with co-tasks such as reading or arith-
metic tasks, and inducing evoked potentials, might pro-
vide an understanding of the cognitive processes that
accompany juggling.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Giuseppina Schiavone http://orcid.org/0000-0003-3584-0678
References
[1] Scholz J, Klein MC, Behrens TE, et al. Training induces
changes in white-matter architecture. Nature Neurosci.
2009;12(11):1370–1371.
[2] Driemeyer J, Boyke J, Gaser C, et al. Changes in gray
matter induced by learning—revisited. PLoS One. 2008;7:
e2669.
[3] Draganski B, Gaser C, Busch V, et al. Neuroplasticity:
changes in grey matter induced by training. Nature.
2004;27(6972):311–312.
[4] Gerber P, Schlaffke L, Heba S, et al. Juggling revisited—a
voxel–based morphometry study with expert jugglers.
NeuroImage. 2014;5:320–325.
[5] MihajlovićV, Grundlehner B, Vullers R, et al. Wearable,
wireless EEG solutions in daily life applications: what are
we missing? J Biomed Health Inform. 2015;19(1):6–21.
[6] Huys R, Beek PJ. The coupling between point-of-gaze and
ball movements in three-ball cascade juggling: the effects
of expertise, pattern and tempo. J Sports Sci. 2002;
(3):171–186.
[7] Patki, S, Grundlehner, B, Verwegen, A, et al. Wireless
EEG system with real time impedance monitoring and
active electrodes. Biomedical Circuits and Systems Con-
ference (BioCAS). 2012:108–111.
[8] Chen YH, de Beeck MO, Vanderheyden L, et al. Soft,
comfortable polymer dry electrodes for high quality ECG
and EEG recording. Sensors. 2014;4(12):23758–23780.
[9] Beek PJ, Santvoord AV. Learning the cascade juggle: a
dynamical systems analysis. J Motor Behav. 1992;4(1):
85–94.
[10] Mapelli A, Galante D, Paganoni S, et al. Three-dimen-
sional hand movements during the execution of ball jug-
gling: effect of expertise in street performers. J
Electromyogr Kinesiol. 2012;2(6):859–865.
[11] Mihajlović, V, PatkiS, Grundlehner, B. The impact of
head movements on EEG and contact impedance: An
adaptive filtering solution for motion artifact reduction.
36th Annual International Conference of the IEEE Engi-
neering in Medicine and Biology Society (EMBC).
2014:5064–5067.
[12] Welch PD. The use of fast Fourier transform for the esti-
mation of power spectra: a method based on time averag-
ing over short, modified periodograms. IEEE Trans Audio
Electroacoust. 1967;15(2):70–73.
[13] Perfetti B, Moisello C, Landsness EC, et al. Ghilardi MF.
Modulation of gamma and theta spectral amplitude
and phase synchronization is associated with the
development of visuo-motor learning. J Neurosci. 2011;1
(41):14810–14819.
[14] Caplan JB, Madsen JR, Schulze-Bonhage A, et al. Human
theta oscillations related to senso-rimotor integration and
spatial learning. J Neurosci. 2003;23:4726–4736.
[15] Tombini M, Zappasodi F, Zollo L, et al. Brain activity
preceding a 2D manual catching task. NeuroImage.
2009;47:1735–1746.
[16] Gentili RJ, Bradberry TJ, Oh H, et al. Cerebral cortical
dynamics during visuomotor transformation: adaptation to
a cognitive-motor executive challenge. Psychophysiology.
2011;48:813–824.
[17] Lange RK, Braun C, Godde B. Coordinate processing dur-
ing the left-to-right hand transfer investigated by EEG.
Exp Brain Res. 2006;68(4):547–556.
[18] Gerloff C, Andres FG. Bimanual coordination and
interhemispheric interaction. Acta Psychol. 2002;10
(2):161–186.
[19] Ghilardi M, Ghez C, Dhawan V, et al. Patterns of regional
brain activation associated with different forms of motor
learning. Brain Res. 2000;871:127–145.
[20] Huys R, Daffertshofer A, Beek PJ. Multiple time scales
and subsystem embedding in the learning of juggling.
Hum Movement Sci. 2004;3(3):315–336.
[21] Dessing JC, Daffertshofer A, Peper CLE, et al. Pattern
stability and error correction during in-phase and antiphase
four-ball juggling. J Motor Behav. 2007;9(5):433–446.
[22] Dessing JC, Rey FP, Beek PJ. Gaze fixation improves the
stability of expert juggling. Exp Brain Res. 2012;16
(4):635–644.
[23] Huys R, Daffertshofer A, Beek PJ. Learning to juggle: on
the assembly of functional subsystems into a task-specific
dynamical organization. Biol Cybern. 2003;8(4):302–318.
[24] Kilner JM, Paulignan Y, Boussaoud D. Functional connec-
tivity during real vs imagined visuomotor tasks: an EEG
study. Neuroreport. 2004;5(4):637–642.
[25] Pfurtscheller G, Neuper C. Motor imagery and direct
brain-computer communication. Proc IEEE. 2001;89
(7):1123–1134.
[26] Barrett J, Ehrlichman H. Bilateral hemispheric alpha
activity during visual imagery. Neuropsychologia. 1982;
(6):703–708.
[27] Haibach PS, Daniels GL, Newell KM. Coordination
changes in the early stages of learning to cascade juggle.
Hum Movement Sci. 2004;3(2):185–206.
12 G. Schiavone et al.

[28] Ichikawa, J, Miwa, K, Terai, H. Analysis of motor skill
acquisition in novice jugglers by three-dimensional motion
recording system. Annual Meeting of Cognitive Science
Society COGSCI. 2014.
[29] Leroy D, Thouvarecq R, Gautier G. Postural organisation
during cascade juggling: Influence of expertise. Gait
Posture. 2008;8(2):265–270.
[30] Klimesch W. Alpha-band oscillations, attention, and con-
trolled access to stored information. Trends Cogn Sci.
2012;6(12):606–617.
[31] Willier, A, Marque, C. Juggling gestures analysis for
music control. In: Gesture and sign language in human-
computer interaction. Berlin: Springer; 2002. p. 296–306.
[32] Bovermann, T, Groten, J, De Campo, A, et al. Juggling
sounds. In Proc 2nd International Workshop on Interactive
Sonification. 2007 February; York.
[33] Reynolds, M, Schoner, B, Richards, J, et al. An immersive,
multi-user, musical stage environment. In Proc 28th Annual
Conference on Computer Graphics and Interactive Tech-
niques. New York, NY: ACM; 2001, August. p. 553–560.
Brain-Computer Interfaces 13