Spatio-temporal Comparison between ERD/ERS and MRCP-based

Movement Prediction

Anett Seeland

, Laura Manca

, Frank Kirchner

1,3

and Elsa Andrea Kirchner

1,3

Robotics Innovation Center (RIC), German Research Center for Artiﬁcial Intelligence (DFKI GmbH),

Robert-Hooke-Straße 1, 28359 Bremen, Germany

Faculty of Biology and Chemistry, University of Bremen, Leobener Str., 28359 Bremen, Germany

Robotics Group, Faculty of Mathematics and Computer Science, University of Bremen,

Robert-Hooke-Straße 1, 28359 Bremen, Germany

Keywords:

Movement Prediction, ERD/ERS, MRCP, Brain-computer Interface, BCI.

Abstract:

In brain-computer interfaces (BCIs) based on electroencephalography (EEG), two distinct types of EEG pat-

terns related to movement have been used for detecting the brain’s preparation for voluntary movements: a)

event-related patterns in the time domain named movement related cortical potentials (MRCPs) and b) patterns

in the frequency domain named event-related desynchronization/synchronization (ERD/ERS). The applicabil-

ity of those patterns in BCIs is often evaluated by the classiﬁcation performance. To this end, the known

spatio-temporal differences in EEG activity can be of interest, since they might inﬂuence the classiﬁcation

performance of the two different patterns. In this paper, we compared the classiﬁcation performance based

on ERD/ERS and MRCP while varying the time point of prediction as well as the used electrode sites. Em-

pirical results were obtained from eight subjects performing voluntary right arm movements. Results show:

a) classiﬁcation based on MRCP is superior compared to ERD/ERS close to the movement onset whereas the

opposite results farther away from the movement onset, b) the performance maximum of MRCP is located at

central electrodes whereas it is at fronto-central electrodes for ERD/ERS. In summary, the results contribute to

a better insight into the spatial and temporal differences between ERD/ERS and MRCP in terms of prediction

performance.

1 INTRODUCTION

In the event-related brain activity, two patterns are

commonly associated with movement, the move-

ment related cortical potentials (MRCPs) and the

event related desynchronization and synchronization

(ERD/ERS) (Pfurtscheller and Lopes da Silva, 1999).

Each of these patterns has components in the prepara-

tory phase of movements (Pfurtscheller and Lopes da

Silva, 1999; Shibasaki and Hallett, 2006).

The MRCPs are slow changes in the amplitude of

the recorded brain activity elicited by movement plan-

ning and execution. Among the pre-movement MR-

CPs several components can be distinguished. The

early readiness potential (RP) starts about 2-1.5 s be-

fore voluntary movement (Stanc

ak et al., 2000; Par-

adiso et al., 2004; Shibasaki and Hallett, 2006). It is

followed by a steep increase in negativity about 500-

400 ms before movement onset (Deecke et al., 1976;

Stanc

ak et al., 2000; Shibasaki and Hallett, 2006),

called late RP. The term pre-motor positivity (PMP)

commonly indicates a positive increase of potential

occurring between 100 and 50 ms before movement

onset (Deecke et al., 1976; Shibasaki and Hallett,

2006; Santucci and Balconi, 2009), which is between

the late RP and the motor potential (MP). The latter

is a steep increase in negativity starting shortly be-

fore movement onset (∼ 50 ms) (Deecke et al., 1976;

Stanc

ak et al., 2000).

The ERD/ERSs are the reﬂection of changes in

the oscillatory activity of neural networks in form

of an attenuation/increase in the power of speciﬁc

frequency bands, which correspond to a desynchro-

nization/synchronization of neural populations in re-

sponse to an event. The brain rhythms commonly

associated with movement, including pre-movement

components, are the µ (8-13 Hz), β (13-30 Hz) and

γ (over 30 Hz) rhythms (Pfurtscheller and Lopes da

Silva, 1999; Pfurtscheller, 1981; Pfurtscheller and

Neuper, 1992; Pfurtscheller et al., 1993; Bai et al.,

219

Seeland A., Manca L., Kirchner F. and Kirchner E..

Spatio-temporal Comparison between ERD/ERS and MRCP-based Movement Prediction.

DOI: 10.5220/0005214002190226

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2015), pages 219-226

ISBN: 978-989-758-069-7

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

2005). The µ ERD has been reported having its onset

about −2 s with respect to movement onset (Stanc

et al., 2000; Shibasaki and Hallett, 2006). The onset

of the ERD in the β-band has been reported between 2

and 1 s before movement onset (Stanc

ak et al., 2000;

Bai et al., 2005; Shibasaki and Hallett, 2006). Con-

cerning the γ-band a pre-movement ERS has been

reported starting about 1 s before movement onset

(Pfurtscheller and Lopes da Silva, 1999).

Being able to recognize and correctly classify

these pre-movement components allows to predict

forthcoming movements. Exploiting the temporal ad-

vantage offered by brain signals over normal output

pathways can for instance result in an earlier per-

ceived response of brain computer interfaces (BCIs)

(Morash et al., 2008), thus making BCIs more user

friendly. Furthermore the detection of motor intention

may be relevant for neuro-rehabilitation by promoting

activity-dependent brain plasticity (Niazi et al., 2011).

Motivated by these possibilities, BCI-researchers

often used the two patterns for the classiﬁcation of

movement intention (e.g. Wang et al., 2004; Li et

al., 2004; Morash et al., 2008; Bai et al., 2011; Fol-

gheraiter et al., 2011; Lew et al., 2012; Seeland et

al., 2013; Ibanez et al., 2014; Jiang et al, 2014).

Some studies used MRCP or ERD/ERS separately

(e.g. Mller-Gerking et al., 1999; Morash et al., 2008;

Wang and Wan, 2009; Bai et al., 2011; Niazi et al.,

2011; Lew et al., 2012; Seeland et al., 2013; Jiang et

al., 2014), others combined them (e.g. Wang et al.,

2004; Li et al., 2004; Ibanez et al., 2014). . However

in only a few studies the reasons behind the choice of

the applied pattern are discussed—a choice that may

rather derive from experience than having a solid the-

oretical basis. Hence, one aim of this work is to em-

pirically compare the two patterns concerning their

efﬁciency for movement prediction.

For such a comparison the time point of predic-

tion as well as the used electrode sites can be rel-

evant, since one of the most observed differences

between the two patterns is their spatio-temporal

evolution. The ERD/ERS has a contralateral on-

set and evolves to a bilateral spread around move-

ment onset (Pfurtscheller and Berghold, 1989; Ba-

biloni et al., 1999; Leocani et al., 2001), while MRCP

starts as bilateral spread and shifts to the contralat-

eral hemisphere as the onset of movement approaches

(Shibasaki and Hallett, 2006). Our intent was to in-

vestigate whether this difference has an impact on the

prediction of voluntary movements in order to get a

better insight into the two EEG patterns that are com-

monly applied for movement prediction. Found dif-

ferences between ERD/ERS and MRCP can be ad-

vantageous for the applicability of movement predic-

Figure 1: Experimental Setup: During the experiment, sub-

jects performed 120 similar movements with their right arm

from a push button to a buzzer.

tion systems. For example they may help in selecting

electrode setups or choosing the target pattern for de-

tection, depending on the application at hand.

2 MATERIALS AND METHODS

The comparison of the two signals is based on an

empirical analysis of a previously conducted study

(Tabie and Kirchner, 2013). Data and methods used

in this analysis are described in the following.

2.1 Data Description

EEG data were recorded via 128 electrodes (acti-

CAP Brain Products GmbH, Munich, Germany) lo-

cated according to the extended international 10-20

system. The predeﬁnded sampling frequency of the

hardware (four BrainAmp DC ampliﬁers; Brain Prod-

ucts GmbH, Munich, Germany) was 5 kHz. The elec-

trode FCz was used as reference and the data were ﬁl-

tered between 0.1 and 1000 Hz. Simultaneously, eight

bipolar channels placed on the right arm of the partic-

ipant were used to record the electromyogram (EMG)

in order to monitor muscular activity (ampliﬁed with

BrainExG MR; Brain Products GmbH, Munich, Ger-

many). Moreover, a motion capturing system (ProRe-

ﬂex 1000, three cameras; Qualisys AB, Gothenburg,

Sweden) was used at 500 Hz to mark the mechanical

movement onset in the EEG.

Eight healthy subjects (29.9 ± 3.3 years) partic-

ipated in the study. Handedness and gender were

ﬁxed to right-handed males to avoid an inﬂuence of

these factors on the experimental results of the rather

small number of subjects. Subjects sat comfortably at

a table in a shielded room and performed three ses-

sions interleaved with breaks of 10 min. They were

BIOSIGNALS2015-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

220

instructed to perform voluntary self-paced right arm

movements from a push button to a buzzer (see Fig-

ure 1). A resting time of 5 s between two consecutive

movements had to be maintained for a valid move-

ment. A visual feedback was shown to subjects when

the resting time was shorter. Each session was con-

sidered as completed when 40 valid movements were

performed. For a detailed description of data and

paradigm it is referred to (Tabie and Kirchner, 2013).

2.2 Physiological Movement Onset

The point in time when the muscles get the electri-

cal signal to move is deﬁned as physiological move-

ment onset. Commonly in neurobiological studies

the physiological movement onset, derived from the

EMG, is used to deﬁne the timescale, i.e., time point

zero corresponds to the point in time when an increase

in electric activity is measured at the muscles. Here,

in an ofﬂine procedure the EMG data of M. biceps

brachii and M. brachioradialis, which contained the

highest signal-to-noise-ratios, were used to label the

physiological movement onset (EMG onset) of each

trial. First, each trial was normalized by subtracting

the mean and dividing by the standard deviation of

the resting period [−2500, −500 ms] preceding the

release of the push button at 0 ms, respectively. Then,

these two normalized EMG channels were averaged.

Next, a variance ﬁlter (Nikolic and Krarup, 2011;

Tabie and Kirchner, 2013) with a window length of

20 ms was applied. Finally, the physiological move-

ment onset was set to the ﬁrst point in time where the

data value exceeded a threshold T for at least 30 ms.

This threshold was deﬁned as

T = m

[−2500,−500]

+ 5 ∗ s

[−2500,−500]

(1)

with mean m and standard deviation s computed for

the resting period from −2500 to −500 ms with re-

spect to the release of the push button.

2.3 Mechanical Movement Onset

The electromechanical delay (Cavanagh and Komi,

1979; Zhou et al., 1995) describes the time between

the physiological movement onset and the produc-

tion of force that yields to an actually visible move-

ment. The latter onset, here named mechanical move-

ment onset, can also be of interest for applications.

Hence, its relation to the physiological movement on-

set shall be given. For that, EEG/EMG and tracking

data streams were synchronized. After calculation of

the speed per sample from the tracking data, the data

were analyzed beginning from the release of the push

button backwards in time. Whenever the speed went

below a threshold of 0.15 mm/sample, which corre-

sponds to the accuracy of the tracking system, a me-

chanical movement onset was marked in the EEG.

The median distance between physiological and

mechanical movement onset across trials was 70 ms

(interquartile range 69 ms).

2.4 EEG Processing

Signal processing and classiﬁcation of the EEG data

were performed with the open source software frame-

work pySPACE (Krell et al., 2013).

Both processing chains, one to detect MRCP and

one to detect ERD/ERS, were provided with the same

training and testing data. The training data contained

supposedly clearly distinguishable instances for the

two classes: instances close to or short after the EMG

onset for the movement preparation class (positive

class) and instances far away from the EMG onsets for

the rest class (negative class). Accordingly, 1 s win-

dows ending at −100 and 50 ms with respect to EMG

onset belonged to the movement preparation class,

and windows of the same length cut every 200 ms, if

no movement occurred 3 s before and 2 s after them,

belonged to the rest class. For the testing data of the

rest class the same extraction rules were applied, but

in order to investigate the temporal evolution of the

performance to detect a pre-movement pattern, the

testing data for the movement preparation class had

to be varied (see Section 2.5).

After window extraction the data were standard-

ized channel-wise (subtraction of mean and division

by standard deviation). Then, the data were processed

differently, depending on the type of patterns (MR-

CPs or ERD/ERSs) that should be detected. This dif-

ferent processing, including dimensionality reduction

and feature extraction, was required due to the differ-

ent properties of the pattern types.

2.4.1 Processing for MRCP

The processing used to detect MRCP pre-movement

components has already be described (Kirchner et al.,

2013a; Kirchner et al., 2013b; Seeland et al., 2013).

Since the RP is a low-frequency component, a ﬁrst

noise reduction step comprised a decimation of the

sampling rate to 20 Hz together with an anti-alias ﬁ-

nite impulse response ﬁlter. Subsequently, another

band pass ﬁlter was applied to reduce the frequen-

cies contained in the signal to 0.1-4 Hz. After this

preprocessing, the 1 s window was reduced to the

last 200 ms. A second noise reduction step was ac-

complished by training the spatial ﬁltering algorithm

xDAWN (Rivet et al., 2009), that is speciﬁcally de-

signed to enhance evoked potentials. Then, the am-

Spatio-temporalComparisonbetweenERD/ERSandMRCP-basedMovementPrediction

221

plitude values of four channels retained from xDAWN

were extracted as features.

2.4.2 Processing for ERD/ERS

The data were decimated to 125 Hz since higher

frequencies are of importance for detection of

ERD/ERS. Accordingly, the signal was ﬁltered in the

broad band from 8 to 40 Hz, which has been chosen

from previous investigations. Broad bands are often

employed for ERD/ERS classiﬁcation in the literature

(Wang et al., 2004; Li et al., 2004; Bai et al., 2011).

The window was then reduced to the last 800 ms and a

common spatial pattern ﬁlter (Blankertz et al., 2008)

was applied with 16 retained pseudo-channels. The

mean and variance of each channel were used as fea-

tures, which has been suggested before (Liao et al.,

2007).

2.4.3 Classiﬁcation and Postprocessing

Each feature dimension was normalized to have zero

mean and a standard deviation of one on the training

data, before a linear support vector machine (SVM)

was trained on the data (Chang and Lin, 2011). The

complexity parameter of the SVM was optimized

for each split using a grid search strategy (grid val-

ues: 10

,10

−1

,...,10

−6

) together with a nested 5-

fold cross validation on the training data. Further,

to account for the class imbalance and to emphasize

the importance of the movement preparation class, the

weight parameter of the SVM for the positive class

was set to 2. Finally, an optimal decision criterion

(threshold), that maximizes the performance on the

training data, was determined to assign a class label

based on the SVM score.

2.5 Evaluation

For evaluation, the data of each subject (three ses-

sions) were merged and results were obtained with a

stratiﬁed 10-fold cross-validation. Accordingly, 108

trials labeled as movement preparation were used in

each cross-validation split for training and 12 trials

for testing. The total number of trials for the rest class

varied across subjects between 140 and 647 since

there was no upper limit of the resting time between

two consecutive movements. As for the movement

preparation class, training in each cross-validation

split was based on 90 % of the trials and the remaining

10 % were used for evaluation.

The balanced accuracy was used as metric, i.e.,

the average of true positive and true negative rate

(BA = (TPR + T NR)/2). The BA compensates for

Figure 2: Electrode locations of a 128-channel actiCAP sys-

tem. Used electrodes in this analysis are numbered. The

number denotes the cluster (1–15) to which the electrode

was grouped.

the different unbalanced ratios of the number of trials

for the positive and negative class (Straube and Krell,

2014).

To investigate the spatio-temporal differences in

performance of the two brain patterns, the end time of

the windows for the movement preparation class and

the electrode locations used for classiﬁcation were

varied. 68 electrodes, that cover a broad area around

the motor cortex, were considered in this analysis.

Electrodes were grouped into 15 different spatial clus-

ters of four or ﬁve electrodes to account for the nois-

iness of a single EEG channel (Figure 2). In the time

dimension, time points from −2.5 to 0.2 s with re-

spect to the EMG onset were analyzed. In addition,

the performances based on all 68 channels as well

as the performance of a random classiﬁer were com-

puted and served as baselines.

3 RESULTS & DISCUSSION

The results are structured into two parts. First,

the movement prediction performance of MRCP and

ERD/ERS components was compared in the temporal

and spatial domain by analyzing the performance at

different electrode groups and time points. However,

a suboptimal performance was expected in this anal-

ysis since only four or ﬁve electrodes were used to

train each classiﬁer. Hence, the second part presents a

comparison of the two patterns using all electrodes

BIOSIGNALS2015-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

222

Figure 3: Spatio-temporal evolution of movement prediction performance in terms of averaged balanced accuracy (BA)

across splits and subjects based on ERD/ERS (top) and MRCP (bottom). Topologies are displayed for 15 electrode clusters

(each black dot indicates the center of a cluster). Time points, relative to the EMG onset at zero, were selected to illustrate

exemplarily the different spatial stages for each pattern as well as the main spatial differences.

for training, i.e., only the time dimension is var-

ied. Additionally, this second analysis gives a more

application-oriented view, since one common objec-

tive in an application is to maximize classiﬁcation

performance.

3.1 Spatio-temporal Comparison

The averaged performance results of MRCP and

ERD/ERS showed differences in their spatio-

temporal evolution as illustrated for selected time

points in Figure 3. The performance of ERD/ERS

exceeded the baseline (0.5 BA) earlier than that of

MRCP (ERD/ERS at about −1.5 s, MRCP at about

−1.0 s).

Furthermore, the performance increase for

ERD/ERS emerged as a recurrent process consisting

of a local maximum at cluster 2 and partly also

cluster 12, as well as a spread of this maximum to

surrounding clusters 8, 3 and sometimes 4 and 7.

For example, 0.09 s before EMG onset the highest

performance is obtained at cluster 2 (Figure 3 third

column). On the other hand the stage of a more

distributed performance can be seen, e.g., 0.51 s

before movement onset (Figure 3 ﬁrst column).

Compared with this, accuracy of MRCP detection

increased simultaneously, i.e., being widespread over

clusters even 0.46 s before EMG onset. Then, a local

maximum emerged at −0.43 s at cluster 2 that spread

to clusters 3 and 4 during the following 200 ms. The

next increase in performance is obtained at −0.24 s

at cluster 4 that spread again widely over clusters 3,

7, 8, and 2 within 0.23 and 0.18 s before movement

onset (Figure 3 second column). Finally, a maximum

occurred at cluster 7 that expanded to clusters 8 and

12 (Figure 3 at −0.09 and −0.05 s).

Optimal classiﬁcation accuracies for ERD/ERS

and MRCP were observed at time point −0.04 s and

zero, respectively, whereby classiﬁcation of MRCP

outperformed detection of ERD/ERS (average BA

± standard error; ERD/ERS: 0.67 ± 0.01, MRCP:

0.71 ± 0.01).

3.2 Comparison using All Channels

As expected, the overall classiﬁcation performance

increased for both brain signals when all 68 chan-

nels were used during training. A possible rea-

son for this increase could be that performance of

MRCP and ERD/ERS is at least partly distributed

over electrodes, as indicated by the results explained

above (Section 3.1 and Figure 3). Thus, adding

more electrodes for training might reveal more rel-

evant information for the detection of the speciﬁc

Spatio-temporalComparisonbetweenERD/ERSandMRCP-basedMovementPrediction

223

Figure 4: Averaged classiﬁcation performance across splits

and subjects over time for classiﬁers based on ERD/ERS

(dashed line) and MRCP (solid line). As baseline per-

formance of a random classiﬁer is depicted. Shaded area

around curves represent the standard error.

brain pattern. The increase was slightly weaker

for ERD/ERS than for MRCP, e.g., at EMG onset

(ERD/ERS: 0.65 ± 0.014 BA with electrodes in clus-

ter 2 and 0.72 ±0.012 BA with 68 electrodes, MRCP:

0.71±0.014 BA with electrodes in cluster 7 and 0.8±

0.012 BA with 68 electrodes).

Figure 4 shows the time course of performance for

the classiﬁcation of MRCP and ERD/ERS, respec-

tively. In addition, the performance of a classiﬁer

that randomly assigns a class label with equal prob-

ability is depicted. The classiﬁcation performance of

both, MRCP and ERD/ERS, performed clearly better

than random after −1.5 s. Further, both performance

curves can be subdivided into three slopes with differ-

ent timings: For MRCP a slow raise in performance

up to −500 ms was observed followed by a steeper in-

crease from −500 to −200 ms, that was even steeper

from −200 ms until the movement approached. In

comparison, accuracy of ERD/ERS detection also

slowly raised up to −900 ms, but already then in-

creased more steeply from −900 to −150 ms, having

the strongest increase from −150 ms until movement

onset. However, this last slope was weaker than the

corresponding one of the MRCP performance curve.

Therefore the two curves intersect at around −130 ms.

After EMG onset the raise in performance for both

patterns continued, resulting for example at the aver-

aged mechanical movement onset (70 ms after EMG

onset) in an accuracy of 0.85 for MRCP and 0.75 for

ERD/ERS.

4 CONCLUSION & OUTLOOK

In this work, we compared the classiﬁcation perfor-

mance of pre-movement components based on MRCP

and ERD/ERS at different time points and differ-

ent electrode groups. This comparison allows to in-

vestigate if the reported neurobiological differences

(Pfurtscheller and Berghold, 1989; Leocani et al.,

2001; Babiloni et al., 1999; Shibasaki and Hal-

lett, 2006) have an inﬂuence on the effectiveness of

movement prediction. Indeed, we obtained spatio-

temporal differences for ERD/ERS and MRCP in the

performances which indicate a higher effectiveness

of ERD/ERS far away from the movement onset,

whereas MRCP performed better near the movement

onset. In the spatial domain, ERD/ERS performance

peaked rather locally at fronto-central electrodes

(contra-medial to the side of movement). This peak

spread to central electrodes. On the contrary, MRCP

performance distribution was more widespread, peak-

ing at central electrodes (contra-medial to the move-

ment side). These ﬁndings do not map one-to-one to

the neurobiological literature, but this is not expected

since methodology (average analysis vs. single trial

processing) and independent variable (voltages differ-

ences vs. classiﬁcation performance) differed.

As outlined in the introduction, ERD/ERS and

MRCP are used separately or sometimes combined.

However, there is no consistent opinion in the litera-

ture on whether ERD/ERS and MRCP should be com-

bined or not. Some authors are in favor of a combi-

nation (Wang et al., 2004; Li et al., 2004) while some

others believe that there is no beneﬁt from the extrac-

tion of features from both patterns (Wang and Wan,

2009). Our results, i.e., the obtained spatio-temporal

differences, indicate that an improvement by combin-

ing features from both patterns may also depend on

the time point of classiﬁcation and the used electrode

sites. Taking a closer look at these two aspects (time

point and sites) may help to resolve existing contro-

versies. Furthermore, knowledge about the spatio-

temporal differences can facilitate the design of novel

combining strategies.

ACKNOWLEDGEMENTS

This work was supported by the German Bundes-

ministerium f

ur Wirtschaft und Technologie (BMWi,

grant FKZ 50 RA 1012 and grant FKZ 50 RA 1011).

The authors like to thank Marc Tabie for providing us

with the data.

BIOSIGNALS2015-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

224

REFERENCES

Babiloni, C., Carducci, F., Cincotti, F., Rossini, P. M., Neu-

per, C., Pfurtscheller, G., and Babiloni, F. (1999). Hu-

man movement-related potentials vs desynchroniza-

tion of EEG alpha rhythm: a high-resolution EEG

study. Neuroimage, 10(6):658–65.

Bai, O., Mari, Z., Vorbach, S., and Hallett, M. (2005).

Asymmetric spatiotemporal patterns of event-related

desynchronization preceding voluntary sequential ﬁn-

ger movements: a high-resolution EEG study. Clin.

Neurophysiol., 116(5):1213–1221.

Bai, O., Rathi, V., Lin, P., and Huang, D. (2011). Prediction

of human voluntary movement before it occurs. Clin.

Neurophysiol., 122(2):364–372.

Blankertz, B., Tomioka, R., Lemm, S., Kawanabe, M., and

uller, K.-R. (2008). Optimizing spatial ﬁlters for ro-

bust EEG single-trial analysis. IEEE Signal Process-

ing Magazine, pages 41–56.

Cavanagh, P. R. and Komi, P. V. (1979). Electromechanical

delay in human skeletal muscle under concentric and

eccentric contractions. European Journal of Applied

Physiology and Occupational Physiology, 42(3):159–

163.

Chang, C.-C. and Lin, C.-J. (2011). LIBSVM:

A library for support vector machines. ACM

Transactions on Intelligent Systems and Tech-

nology, 2:27:1–27:27. Software available at

http://www.csie.ntu.edu.tw/ cjlin/libsvm.

Deecke, L., Gr

ozinger, B., and Kornhuber, H. H. (1976).

Voluntary ﬁnger movement in man: Cerebral poten-

tials and theory. Biol. Cybern., 23(2):99–119.

Folgheraiter, M., Kirchner, E., and Seeland, A. (2011). A

multimodal brain-arm interface for operation of com-

plex robotic systems and upper limb motor recovery.

In BIODEVICES 2011 - International Conference on

Biomedical Electronics and Devices, pages 150–162.

Ibanez, J., Serrano, J. I., del Castillo, M. D., Monge-

Pereira, E., Molina-Rueda, F., Alguacil-Diego, I.,

and Pons, J. L. (2014). Detection of the onset of

upper-limb movements based on the combined anal-

ysis of changes in the sensorimotor rhythms and slow

cortical potentials. Journal of Neural Engineering,

11(5):056009.

Jiang, N., Gizzi, L., Mrachacz-Kersting, N., Dremstrup, K.,

and Farina, D. (2014). A brain-computer interface for

single-trial detection of gait initiation from movement

related cortical potentials. Clin Neurophysiol.

Kirchner, E. A., Albiez, J., Seeland, A., Jordan, M., and

Kirchner, F. (2013a). Towards assistive robotics for

home rehabilitation. In Chimeno, M. F., Sol

e-Casals,

J., Fred, A., and Gamboa, H., editors, Proceedings of

the 6th International Conference on Biomedical Elec-

tronics and Devices (BIODEVICES-13), pages 168–

177, Barcelona. ScitePress.

Kirchner, E. A., Kim, S. K., Straube, S., Seeland, A.,

ohrle, H., Krell, M. M., Tabie, M., and Fahle, M.

(2013b). On the applicability of brain reading for pre-

dictive human-machine interfaces in robotics. PLoS

ONE, 8(12):e81732.

Krell, M. M., Straube, S., Seeland, A., W

ohrle, H., Teiwes,

J., Metzen, J. H., Kirchner, E. A., and Kirchner, F.

(2013). pySPACE - a signal processing and classiﬁca-

tion environment in Python. Frontiers in Neuroinfor-

matics, 7(40). https://github.com/pyspace.

Leocani, L., Toro, C., Zhuang, P., Gerloff, C., and Hallett,

M. (2001). Event-related desynchronization in reac-

tion time paradigms: a comparison with event-related

potentials and corticospinal excitability. Clin. Neuro-

physiol., 112(5):923–30.

Lew, E., Chavarriaga, R., Silvoni, S., and Mill

an, J. D. R.

(2012). Detection of self-paced reaching movement

intention from EEG signals. Front. Neuroeng., 5:13.

Li, Y., Gao, X., Liu, H., and Gao, S. (2004). Classiﬁca-

tion of single-trial electroencephalogram during ﬁnger

movement. IEEE Trans. Biomed. Eng., 51(6):1019–

1025.

Liao, X., Yao, D., Wu, D., and Li, C. (2007). Combining

spatial ﬁlters for the classiﬁcation of single-trial EEG

in a ﬁnger movement task. IEEE Trans. on Bio-med.

Eng., 54(5):821–31.

Morash, V., Bai, O., Furlani, S., Lin, P., and Hallett, M.

(2008). Classifying EEG signals preceding right hand,

left hand, tongue, and right foot movements and motor

imageries. Clin. Neurophysiol., 119(11):2570–2578.

uller-Gerking, J., Pfurtscheller, G., and Flyvbjerg, H.

(1999). Designing optimal spatial ﬁlters for single-

trial EEG classiﬁcation in a movement task. Clin.

Neurophysiol., 110(5):787–798.

Niazi, I. K., Jiang, N., Tiberghien, O., Nielsen, J. r. F. k.,

Dremstrup, K., and Farina, D. (2011). Detection

of movement intention from single-trial movement-

related cortical potentials. J Neural Eng, 8(6).

Nikolic, M. and Krarup, C. (2011). EMGTools, an adaptive

and versatile tool for detailed EMG analysis. Biomed-

ical Engineering, IEEE Transactions on, 58(10):2707

–2718.

Paradiso, G., Cunic, D., Saint-Cyr, J. a., Hoque, T., Lozano,

A. M., Lang, A. E., and Chen, R. (2004). Involve-

ment of human thalamus in the preparation of self-

paced movement. Brain, 127(Pt 12):2717–2731.

Pfurtscheller, G. (1981). Central beta rythm during senso-

rimotor activities in man. Electroencephalogr. Clin.

Neurophysiol., 51(3535):253–264.

Pfurtscheller, G. and Berghold, a. (1989). Patterns of corti-

cal activation during planning of voluntary movement.

Electroencephalogr. Clin. Neurophysiol., 72(3):250–

258.

Pfurtscheller, G. and Lopes da Silva, F. H. (1999). Event-

related EEG/MEG synchronization and desynchro-

nization: basic principles. Clin. Neurophysiol.,

110(11):1842–1857.

Pfurtscheller, G. and Neuper, C. (1992). Simultaneous EEG

10Hz desynchronization and 40 Hz synchronization

during ﬁnger movements. Neuroreport, 3:1057–1060.

Pfurtscheller, G., Neuper, C., and Kalcher, J. (1993). 40-Hz

oscillations during motor behavior in man. Neurosci.

Lett., 164(1-2):179–182.

Rivet, B., Souloumiac, A., Attina, V., and Gibert, G. (2009).

xDAWN algorithm to enhance evoked potentials: ap-

Spatio-temporalComparisonbetweenERD/ERSandMRCP-basedMovementPrediction

225

plication to brain-computer interface. IEEE Trans. on

Bio-med. Eng., 56(8):2035–2043.

Santucci, E. and Balconi, M. (2009). The multicompo-

nential nature of movement-related cortical potentials:

functional generators and psychological factors. Neu-

ropsychol. Trends.

Seeland, A., Woehrle, H., Straube, S., Kirchner, E. A., and

ohrle, H. (2013). Online Movement Prediction in a

Robotic Application Scenario. In Proc. 6th Int. IEEE

EMBS Conf. Neural Eng., pages 41–44, San Diego.

Shibasaki, H. and Hallett, M. (2006). What is

the Bereitschaftspotential? Clin. Neurophysiol.,

117(11):2341–2356.

Stanc

ak, A., Feige, B., L

ucking, C. H., and Kristeva-Feige,

R. (2000). Oscillatory cortical activity and movement-

related potentials in proximal and distal movements.

Clinical neurophysiology : ofﬁcial journal of the In-

ternational Federation of Clinical Neurophysiology,

111(4):636–50.

Straube, S. and Krell, M. M. (2014). How to evaluate an

agent’s behaviour to infrequent events? – reliable per-

formance estimation insensitive to class distribution.

Frontiers in Computational Neuroscience, 8(43).

Tabie, M. and Kirchner, E. A. (2013). EMG onset detec-

tion – comparison of different methods for a move-

ment prediction task based on EMG. In Alvarez,

S., Sol

e-Casals, J., Fred, A., and Gamboa, H., ed-

itors, In Proceedings of the 6th International Con-

ference on Bio-inspired Systems and Signal Process-

ing (BIOSIGNALS-13), pages 242–247, Barcelona.

SciTePress.

Wang, B. and Wan, F. (2009). Classiﬁcation of Single-Trial

EEG based on support vector clustering during ﬁnger

movement. Adv. Neural NetworksISNN 2009, pages

354–363.

Wang, Y., Zhang, Z., Li, Y., Gao, X., Gao, S., and Yang, F.

(2004). BCI Competition 2003–Data set IV: an algo-

rithm based on CSSD and FDA for classifying single-

trial EEG. IEEE Trans. Biomed. Eng., 51(6):1081–6.

Zhou, S., Lawson, D. L., Morrison, W. E., and Fairweather,

I. (1995). Electromechanical delay in isometric mus-

cle contractions evoked by voluntary, reﬂex and elec-

trical stimulation. European Journal of Applied Phys-

iology and Occupational Physiology, 70(2):138–145.

BIOSIGNALS2015-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

226